View/download the full transcript from Day 1.
October 30, 2017 | Author: Anonymous | Category: N/A
Short Description
, that there's. 20 a 30 percent chance of encountering a Janet Watkins IN THE CIRCUIT COURT FOR matriculate ......
Description
1
1 2 3
U.S. FOOD AND DRUG ADMINISTRATION (FDA)
4
and
5
AMERICAN ASSOCIATION FOR CANCER RESEARCH (AACR)
6 7
PUBLIC WORKSHOP
8 9
Dose-Finding of Small Molecule Oncology Drugs
10 11 12 13 14 15
Monday, March 18, 2015
16
8:01 a.m. to 5:02 p.m.
17 18
Washington Court Hotel
19
Washington, DC
20 21 22
A Matter of Record (301) 890-4188
2
C O N T E N T S
1 2
AGENDA ITEM
PAGE
3
Welcome and Work Objectives
4
Pasi Jänne, MD, PhD
5
5
Amy McKee, MD
5
6
SESSION 1: Small Molecule Characterization
7
Pharmacology Matters: Adapting the
8
Paradigm of Small Molecule Oncology Drug
9
Development
10
Natalie Simpson, PhD
11
Is It Safe: Understanding the Performance of
12
Nonclinical Safety Assessment Models in
13
Predicting Human Outcomes
14
Thomas Jones, PhD
15
Nonclinical to Clinical Correlation of
16
Adverse Effects of Kinase Inhibitors
17
Richard Brennan, PhD
18
Safety Lead Optimization of Kinase
19
Inhibitors: Learnings from Attrition and
20
Translation
21
Donna Dambach, VMD, PhD
22
A Matter of Record (301) 890-4188
6
22
46
72
3
C O N T E N T S (continued)
1 2
AGENDA ITEM
3
Enhancing the Safety of Kinase Inhibitor
4
Oncolytic Drugs: Preclinical/Clinical
5
Opportunities
6
PAGE
William Kluwe, PhD
7
Moderated Panel Discussion
8
Todd Palmby, PhD
9
Donna Dambach, VMD, PhD
10
Audience Q&A
11
SESSION 2: Design of Phase 2 Dose Finding
12
Dose Selection for Small Molecule
13
Oncology Drugs: Present State and
14
Future Considerations
15
Nitin Mehrotra, PhD Optimal Dosing for Targeted Therapies in
17
Oncology: Drug Development Cases Leading
18
By Example
20 21
125
144
16
19
102
Dinesh De Alwis, PhD
183
198
Optimizing Dose-Finding Trials: Statistics Laura Fernandes, PhD
22
A Matter of Record (301) 890-4188
218
4
C O N T E N T S (continued)
1 2
AGENDA ITEM
3
Best Practices of Adaptive Dose-Finding
4
Studies
5
PAGE
Stuart Bailey, PhD
6
Best Practices of Adaptive Dose-Finding
7
Studies II
8 9 10 11
Jose Pinheiro, PhD
230
262
Pharmacometrics in Industry Amit Roy, PhD Moderated Panel Discussion
12
Lei Nie, PhD
13
Eric Rubin, MD
294 314
14
Audience Q&A
323
15
Wrap Up and Adjourn
380
16 17 18 19 20 21 22
A Matter of Record (301) 890-4188
5
1
P R O C E E D I N G S
2
(8:01 a.m.)
3
Welcome and Workshop Objectives
4
DR. JANNE:
Good morning, everybody, and
5
welcome to our FDA-AACR Dose-finding of Small
6
Molecule Oncology Drugs Workshop.
7
Janne from the Dana Farber Cancer Institute.
8
myself and Dr. Eric Rubin are co-chairing this
9
meeting on behalf of AACR.
10
DR. MCKEE:
11 12
Hi.
My name is Pasi And
I'm Dr. Amy McKee, and I'm
one of the co-chairs with Dr. Geoff Kim from FDA. DR. JANNE:
So hopefully we'll have a robust
13
discussion over the next couple days.
And really,
14
the overall goal of the workshop is to explore best
15
practices, dose-finding, selection for small
16
molecule kinase inhibitors.
17
Goal is to -- we want to foster a robust
18
discussion from a movement away from our typical
19
dose escalation type of studies to move, to think
20
about adaptive designs that can potentially
21
incorporate clinical, pharmacologic,
22
pharmacometric, and when appropriate, nonclinical
A Matter of Record (301) 890-4188
6
1
information to guide dose selections. A long-term goal is to spur initiatives
2 3
where dose-finding and selection are no longer
4
restricted to early phases of drug development and
5
will ultimately become integrated into the life
6
cycle of drug development with continued refinement
7
in evidence as new data emerges. So again, we hope to have robust
8 9
discussions.
We have presentations and panel
10
discussions.
We'd like this to be obviously as
11
interactive as possible throughout the next two
12
days.
13
So with that, I'll turn it over to Amy. DR. MCKEE:
So just on behalf of the FDA,
14
we'd like to welcome you, and I will introduce our
15
first speaker, Dr. Natalie Simpson, who is a
16
pharmacologist/toxicologist at the FDA.
17
DR. JANNE:
18
have the panel at the end.
19 20
Thank you.
And after this session, we'll
Presentation – Natalie Simpson DR. SIMPSON:
Good morning.
My name is
21
Natalie Simpson, and the title of my talk is
22
Pharmacology Matters:
Adapting the Paradigm of
A Matter of Record (301) 890-4188
7
1
Small Molecule Oncology Drug Development.
2
my disclaimer slide stating that these are my
3
opinions and do not necessarily reflect those of
4
the FDA.
5
This is
So we are here today to discuss the fact
6
that it's difficult to optimize therapeutic doses
7
for target therapies like kinase inhibitors using
8
the traditional paradigm for cytotoxic drugs.
9
reasons for this are exposure-response
10
relationships are rarely defined, resulting in
11
frequent dose reductions due to dose-limiting
12
toxicities or DLTs.
13
Also, inter-patient variability is not
14
adequately evaluated during early clinical
15
development, and this results in the fact that
16
sponsors have to conduct additional dose
17
optimization studies postmarketing.
18
So the question becomes, what steps can we
19
take to address this problem and improve dose
20
optimization for kinase inhibitors?
21 22
Some
At this workshop, you will hear from multiple disciplines about the best practices in
A Matter of Record (301) 890-4188
8
1
incorporating various types of data into this
2
process.
3
pharmacology and toxicology data to predict human
4
adverse events.
5
This morning's session will begin using
The goals of this session are to identify
6
the key best practices in the nonclinical
7
evaluation of a compound, including but not limited
8
to, selectivity, pharmacology, secondary
9
pharmacology, and toxicology, and integration of
10 11
this information during dose optimization. On this slide, I would like to highlight the
12
challenges of dose optimization in finding the
13
right dose for the right patient using ponatinib
14
and nilotinib as examples.
15
ABL kinase inhibitors with a favorable risk-benefit
16
profile, but also a history of labeling changes.
17
There drugs are both
This is related to the fact that early
18
clinical trials in animal toxicology studies did
19
not predict the toxicities of thromboembolism and
20
vascular occlusion as adverse events, and also the
21
fact that toxicities are delayed and cumulative.
22
In fact, the incidence and frequency of
A Matter of Record (301) 890-4188
9
1
vascular events collected through the Adverse Event
2
Reporting System are consistently increasing with
3
the use of ponatinib and nilotinib, and industry
4
has made changes and has valuable insight to
5
improve dose optimization based on such examples.
6
So what we are interested in from a
7
nonclinical perspective is, was there other
8
information that could have helped predict the
9
toxicity?
We know that clinical experience and
10
pharmacology data suggests that DLTs may be related
11
to the promiscuous activity of kinase inhibitors.
12
Kinase inhibitors, including ABL kinase
13
inhibitors like ponatinib and nilotinib, target
14
different mutations often with the result of less
15
selectivity.
16
target have a common toxicity.
17
VEGF inhibitors, you typically see hypertension;
18
with EGFR inhibitors, you typically see rash.
19
We also know that drugs with a common For example, with
With ponatinib and nilotinib, both target
20
ABL kinase and various mutations, but this may not
21
be the only explanation for the vascular events,
22
because as I will discuss in the next few slides,
A Matter of Record (301) 890-4188
10
1
there is increased risk for vascular occlusive
2
events with ponatinib and nilotinib compared to
3
other ABL kinase inhibitors.
4
Also, if we dig deeper and explore secondary
5
pharmacology, there appears to be differential
6
targeting among ABL kinase inhibitors for kinases
7
involved in vascular maintenance; for example,
8
those regulating inflammatory responses and
9
angiogenesis pathways.
10
I mentioned there was an increased risk for
11
vascular occlusive events for ponatinib and
12
nilotinib compared to other ABL kinase inhibitors.
13
We took an exploratory look at hierarchical
14
clustering of grade 3-4 cardiovascular events
15
observed with kinase inhibitors obtained using
16
human safety data.
17
The preferred term of grade 3-4
18
cardiovascular treatment emergent adverse events,
19
or AEs, were grouped by type of event.
20
failure, stroke, ischemic heart disease, and
21
peripheral arterial vascular events are on the
22
X-axis of the heat map and each represented by an
A Matter of Record (301) 890-4188
Cardiac
11
1
individual column on the heat map. Unsupervised hierarchical clustering based
2 3
on the type of event was performed for selected ABL
4
VEGFR and EGFR kinase inhibitors that are on the
5
Y-axis.
6
indicating a high incidence of adverse events and
7
green indicating a low incidence, interestingly,
8
the vascular events with high incidence are mainly
9
observed with ponatinib and nilotinib -- you can
And if you look at the heat map, with red
10
see up at the top -- and they cluster away from
11
other ABL kinase inhibitors.
12
I'm not sure if you can see it, but
13
dasatinib and imatinib are also on that heat map
14
and traditional VEGFR inhibitors that are more
15
clustered in the middle of the heat map.
16
So there is increased risk of vascular
17
events with ponatinib and nilotinib.
18
other information that can explain why?
19
Is there
As I mentioned, kinase inhibitors by nature
20
many times are not very selective.
21
as an example of secondary pharmacology data that
22
is still exploratory, but may be useful
A Matter of Record (301) 890-4188
On this slide,
12
1
retrospectively to try to understand the increased
2
risk of vascular events, this is an unsupervised
3
hierarchical clustering analysis of data that was
4
published in PLoS ONE last year, comparing kinase
5
selectivity across drugs. Of note, this type of analysis was not
6 7
sourced from NDA submissions.
It was not available
8
at the time of drug approval for ponatinib and
9
nilotinib. On the left margin of the hierarchical
10 11
clustering are kinase inhibitors we pooled from
12
this publication, mainly for their intent to target
13
ABL VEGFR and EGFR in which their inhibitory
14
activity at one micromolar concentration for
15
300 kinases that are listed on the X-axis had been
16
compared in a single screen.
17
thing.
18
So that's a big
This is all on a single screen. Red on this heat map means high percent
19
enzyme inhibition for a particular kinase on the
20
X-axis, and green indicates low percent inhibition
21
for a particular kinase.
22
So you can that afatinib and erlotinib at
A Matter of Record (301) 890-4188
13
1
the top, but there is not a lot of red, indicating
2
that there is not a high percentage of inhibition
3
for each of those 300 kinases that are on that heat
4
map.
5
for their intended target.
And this translates to being more selective
On the other hand, at the bottom, if you
6 7
look at ponatinib and sunitinib, there is a lot of
8
red on that heat map, indicating that many of those
9
300 kinases are targeted by those drugs.
And so
10
this would be an example of a more promiscuous type
11
of drug that is not very selective for its intended
12
target.
13
of that heat map.
14
And nilotinib is somewhere in the middle
So ponatinib and nilotinib target many
15
kinases.
16
selective by nature of their mechanism of action.
17
But can inhibition of any of these kinases by these
18
drugs possibly explain the increased risk of
19
vascular events?
20
This goes back to them being less
So we took a closer look at the data from
21
that particular paper, particularly focusing on
22
kinases that are involved or known to affect
A Matter of Record (301) 890-4188
14
1
endothelial survival/function and vascular
2
maintenance.
3
of the kinases that we're looking at to see their
4
inhibition for the particular kinase inhibitor
5
that's on the Y-axis.
6
So again, on the X-axis is the list
Again, the same thing as last time, percent
7
inhibition, high percent inhibition is red, low
8
percent inhibition for a particular kinase by the
9
kinase inhibitor is green.
10
So what you can see is drugs with high
11
inhibitory activity for kinases involved in
12
vascular maintenance and function cluster in red at
13
the top, and drugs with little effect cluster in
14
green at the bottom, and ponatinib and nilotinib
15
are up in that red cluster.
16
Another thing that was interesting when we
17
did this type of analysis from the data from this
18
paper was that there is some differential targeting
19
for P38, which is one of the kinases known to
20
effect endothelial survival and function and
21
vascular maintenance.
22
is differential targeting within that cluster.
And you can see that there
A Matter of Record (301) 890-4188
15
So this is the type of pharmacology data
1 2
that could be followed up in other studies to
3
possibly understand and predict the cause of
4
vascular events.
5
Therefore, taking into account the target
6
selectivity of ponatinib and nilotinib and common
7
adverse events, it is not unreasonable to conclude
8
that risk factors for vascular events may be
9
detected early in development using nonclinical
10
pharmacology studies together with available prior
11
clinical safety data, and followed up in
12
appropriate models to address remaining concerns
13
about the potential to cause these effects in
14
humans.
15
The first goal of this morning's session is
16
to discuss the safety evaluation of kinase
17
inhibitors using pharmacology and toxicology.
18
current paradigms recommended per ICH S9 guidance
19
are that prior to phase 1 for advanced oncology
20
indications, we look at primary pharmacology
21
studies to support the mechanism of action and
22
antitumor activity.
The
And we also look at toxicology
A Matter of Record (301) 890-4188
16
1
studies, typically 28 days in duration, in a
2
relevant rodent and non-rodent species.
3
see studies of longer duration typically until
4
prior to phase 3, and those are three months.
We don't
Based on the examples of ponatinib and
5 6
nilotinib, we would also like a discussion as to
7
how better use of secondary pharmacology data in
8
the overall safety assessment of kinase inhibitors
9
can be useful to identify potential adverse events
10
in addition to the current paradigm. The second goal is to identify the best
11 12
practices in industry for the evaluation of lead
13
compounds.
14
determine target selectivity and potency?
15
are many different methods and readouts to
16
determine target selectivity.
What are the best practices used to There
What types of studies are conducted?
17
Is the
18
typical approach to start with direct binding and
19
follow-up only on positive hits and functional
20
assays?
21
interpret the data?
22
inhibition, direct versus allosteric and structure
What is the best way to present and To what extent are the type of
A Matter of Record (301) 890-4188
17
1
activity, relationships used? What are the best methods and threshold to
2 3
define a positive hit?
Is it best to interpret
4
data using percent inhibition or IC50?
5
are the best practices for correlation of animal
6
and human toxicities in the predictivity of DLTs?
Also, what
Are some species better predictors of
7 8
toxicity than others?
Are there limitations of
9
animal studies to identify DLTs for non-cytotoxic
10
therapies?
And are correlations' predictivity
11
dependent on target organs and class events? For example, young, healthy animals on
12 13
normal diets used in general toxicology studies may
14
not be the best predictors of vascular events for
15
patients eating fatty Western diets with possible
16
comorbidities or other risk factors. We would like an open dialogue to define
17 18
value, utility of studies, and dose optimization,
19
and possible standardization. The third goal is to discuss de-risking
20 21
strategies for molecularly targeted anticancer
22
drugs.
How can pharmacology data help fill gaps
A Matter of Record (301) 890-4188
18
1
left by toxicology studies?
2
that is not typically submitted to the agency that
3
could be used in safety assessment?
4
negative data is something that would be useful in
5
safety assessments and knowing how far below the
6
threshold cutoff a certain kinase fell; we could
7
definitely use that data.
8 9
Is there information
For example,
Also, can gaps be filled by understanding the in vivo activity encompassing differences in
10
kinase pathways among species and context-specific
11
function on different organs?
12
evolving and outputting massive amounts of data, to
13
what extent can computational modeling or systems
14
biology be useful in de-risking?
15
And as technology is
Further, is fully preventing an expected
16
toxicity realistic?
17
on management of toxicity, taking into account
18
patient comorbidities?
19
And should the focus also be
The fourth and fifth goals are to discuss
20
approaches to integrating information gleaned from
21
the nonclinical evaluation into the design of phase
22
1 clinical trials and the role of the nonclinical
A Matter of Record (301) 890-4188
19
1
team during the developmental process. With the overall goal, it would be to
2 3
increase safety and efficiency in clinical trials
4
with the prospective use of available nonclinical
5
data.
6
relationships be used in selection of dose and
7
dosing schedule?
8
nonclinical studies were used to select dose and
9
dosing schedule for everolimus.
10 11
For example, can PK/PD and PK toxicity
PD endpoints determined from
Are there other
examples? Are there examples of integrative approaches
12
for dose selection based on animal PK and
13
allometric scaling in the context of various
14
potency assays?
15
clinical studies become available, how might PK
16
activity and toxicity information be integrated to
17
optimize the clinical dose and schedule?
18
And as data from nonclinical and
Also, what kind of role do the nonclinical
19
teams play during clinical trial design?
Is it
20
only prior to phase 1?
21
play a role in the attribution of toxicity to the
22
study drug, and in clinical toxicity management,
And do nonclinical teams
A Matter of Record (301) 890-4188
20
1
and dose optimization?
2
So this is our wish list for nonclinical
3
data that we think could make the small molecule
4
oncology drug development process more efficient in
5
finding the right dose for the right patient.
6
hope to learn in this workshop the best practices
7
for better dose optimization for targeted small
8
molecule oncology drugs.
9
We
We hope you take home from this morning's
10
session that pharmacology matters in the
11
development of these drugs from early lead
12
selection to optimization through to the
13
postmarketing setting.
14
As such, a thorough understanding of the
15
drug target and activity should help to maximize
16
the therapeutic benefit and minimize risk.
17
role of the nonclinical team may extend beyond
18
phase 1.
19
And the
This entire process is dependent on starting
20
with the most reliable and biologically relevant
21
kinase inhibition data.
22
the entire kinome, description of methods,
For example, encompassing
A Matter of Record (301) 890-4188
21
1
standardized platforms, and that can be used to
2
make cross-drug comparisons for risk assessment.
3
At this time, we do not have a requirement
4
to submit this type of data for oncology products.
5
And if they are submitted, they're submitted with
6
the data in relation to efficacy and not safety.
7
So we would like to see data submitted with an
8
explanation of the secondary pharmacology data as
9
it pertains to safety and not just efficacy.
10
An example of what we envision for kinase
11
inhibitors includes a process by which correlation
12
of clinical adverse events with kinase inhibitory
13
profiles – pharmacology -- and clinical
14
exposure -- pharmacometric data -- will be used din
15
addition to the current methods to better predict
16
toxicity and find the right dose for the right
17
patient.
18
Thank you for your attention.
And I just
19
wanted to leave the reminder of the goals for this
20
morning as I'm walking off the stage.
21
(Applause.)
22
DR. MCKEE:
Okay.
Our next speaker is from
A Matter of Record (301) 890-4188
22
1 2 3
Lilly, Dr. Thomas Jones. Presentation – Thomas Jones DR. JONES:
Good morning.
I'd like to first
4
thank the organizers for the opportunity to come
5
and be part of this workshop.
6
specialist or expert in kinase or kinase
7
inhibitors, or even so much an expert in oncology
8
drug development, although clearly my organization
9
is involved in that.
And I come not as a
10
I think the reason that I'm here is that
11
I've spent about as much time thinking about the
12
performance of our nonclinical safety models as
13
anyone, with the possible exception of Bill Kluwe,
14
who is admittedly older than me.
15
(Laughter.)
16
DR. JONES:
So I thought I'd start by just
17
flashing up the meeting goals again that Natalie
18
started to present with you.
19
goal that I'm going to be focused on most
20
prominently is goal 2, to talk about the selection
21
of lead compounds, but in particular this notion of
22
correlation of animal and human toxicities, and
And to highlight the
A Matter of Record (301) 890-4188
23
1
prediction of toxicities from the nonclinical
2
species to predict relevant dose-limiting
3
toxicities.
4
In fact, it's the two words, "correlation,"
5
often used synonymously with concordance and
6
"predictivity," that I want to dissect out a bit
7
with you today and make sure that we are
8
approaching our setting expectations in a realistic
9
fashion around these aspirations.
10
So one thing that we know about prediction
11
is captured in this quote.
"Prediction is very
12
difficult, especially if it is about the future."
13
Now if I were to ask for a show of hands, I predict
14
that most of you would attribute this quote to Yogi
15
Berra, but in fact, this is a quote from Niels
16
Bohr, suggesting that it's much more relevant and
17
we should pay attention to it much more closely.
18
In the context of today, the way I'm going
19
to use the term "prediction" is the application of
20
a nonclinical test to evaluate a novel molecular
21
structure.
22
That combined with information that we've gleaned
From that test we will derive a result.
A Matter of Record (301) 890-4188
24
1
from previous molecules that we've studied will
2
inform a prediction, a prediction as to what the
3
likely human outcomes are to be as we move forward
4
in the clinical development. We actually will treat the prediction, and
5 6
we do treat this prediction, as a window on the
7
future.
8
We accelerate programs; we de-accelerate programs.
We make resourcing decisions around it.
9
But the fact is, until we get into the
10
clinic and realize the true human outcomes, we
11
don't know whether this prediction is correct or
12
incorrect.
13
nonclinical scientists on the back, but incorrect
14
we take a look.
15
test system; what did the test system not tell us
16
that we should have known in making the prediction?
17
Obviously, if it's correct, we pat our
And generally we look back at the
This is where the idea of doing
18
retrospective analyses and finding tests that have
19
a concordance with human outcomes is important and
20
is relevant, and is a big part of how we
21
investigate mechanisms of adverse effects.
22
So it's not that that is a bad thing to do,
A Matter of Record (301) 890-4188
25
1
but we have to keep it in the context that we start
2
that exercise knowing the human outcome, whereas a
3
prediction going forward is absent that
4
information.
5
The other area that we can focus on is
6
reconsidering our use of the information that we
7
gain from the test or our historical database as
8
well, and we'll talk a little bit about that going
9
forward as well.
10
So I thought I'd be provocative to
11
start -- and I love the show MythBusters.
12
thought I would use the myth-busting approach to
13
address the question, are cytotoxic drugs really
14
well predicted by animal studies?
15
So I
I think we've all taken the assumption that
16
they have been.
17
Rozencweig back in 1981, in which he and his
18
co-authors used a Bayesian approach to look at the
19
predictive value in a quantitative way from the
20
findings from tox studies.
21 22
And there is a paper from Marcel
This actually served as the basis for some protocol development at the National Cancer
A Matter of Record (301) 890-4188
26
1
Institute, and their bottom-line conclusion was the
2
data suggests that the most common dose-limiting
3
toxicities in the clinic were well predicted.
4
However, less often cited from this paper
5
was a passage that the positive predictive value of
6
the animal findings for specific human organ
7
systems depended on the prevalence of the human
8
toxicity among the chemo-types being tested.
9
what's that all about?
10
And
Well, to get there, let me just digress for
11
a moment and go back to something that we all
12
touched on as students, the confusion matrix, which
13
is a tool for analyzing and evaluating the
14
performance of any kind of predictive test system.
15
I've laid out for you here 50 positive
16
outcomes and 50 negative outcomes.
And as with any
17
assay, the expectation is we don't see the world
18
clearly in black and white.
19
outcomes that as this assay is set up are seen as
20
negative, and there are negative outcomes that are
21
seen as positive.
22
among those within the confusion matrix that we're
So there are positive
And this is the distribution
A Matter of Record (301) 890-4188
27
1
all familiar with.
2
Now generally, we don't focus on the issue
3
of prevalence, but prevalence is denoted here, and
4
I'll use this throughout the talk, by this 50
5
percent mark, which denotes that 50 percent of the
6
data that was used to populate this matrix was
7
positive outcome data.
8 9
The performance indicators that most of us are familiar with around the confusion matrix are
10
sensitivity and specificity.
11
we just constructed, we're showing a sensitivity of
12
90 percent, a specificity of 80.
13
And using the matrix
Sensitivity and specificity show how well
14
the outcomes that are known are actually
15
partitioned by the assay system.
16
little to do with prediction per se.
17
around how well does the assay actually parse out
18
correct and incorrect findings.
They have very It's more
19
To get at the question of prediction, we
20
have to turn the confusion matrix on its side if
21
you will.
22
in which he introduced the concept of positive and
Thomas Vecchio in 1966 published a paper
A Matter of Record (301) 890-4188
28
1
negative predictive values.
2
here.
3
are measures of how well the test predicts forward,
4
with the data that arise from the assay, that can
5
be measured in terms of consistent performance
6
through sensitivity and specificity.
7
The formula is shown
The positive and negative predictive values
There is a very important element of the
8
positive and negative predictive value, and that is
9
that it is influenced by this notion of prevalence
10
of the outcome being measured.
11
show you this, because this is a very important
12
concept that's often overlooked in assay validation
13
in the nonclinical world.
14
And let me just
This is the same matrix that I just showed
15
you, constructed with the 50 positive and
16
50 negative outcomes as they were seen by our
17
assay, showing the 50 percent prevalence that
18
distributed them within the matrix.
19
If, however, the true incidence of whatever
20
it is we're measuring with this assay is only
21
10 percent in the general population, the
22
sensitivity and specificity you'll see remain the
A Matter of Record (301) 890-4188
29
1
same.
The assay is performing just as it did
2
before, but the positive predictive value goes from
3
82 percent down to 33.
4
predictive value increases accordingly in contrast.
You'll notice the negative
In fact, for every assay, there is a point
5 6
at which the positive and negative predictive value
7
equal each other.
8
point, negative predictive value will always
9
dominate over positive.
At a prevalence below that
Above that point, the
10
positive predictive value will be the strongest
11
performance indicator. So going back to the Rozencweig paper, they
12 13
used a data set produced by Schein's group in the
14
dog to look at the prevalence and the associated
15
positive predictive values of toxicities of 25
16
cytotoxic drugs.
17
toxicities in man against the positive predictive
18
value calculated from the data set generated in
19
dog.
20
Here's the prevalence of the
What you can see is once you get past the
21
gastrointestinal and bone marrow toxicity, so often
22
associated with the attack when normal rapidly
A Matter of Record (301) 890-4188
30
1
proliferating cells and normal tissues, you can see
2
that the predictive value of these test systems,
3
even for cytotoxic drugs, is barely better than a
4
coin flip for liver, and it goes south from there.
5
If you think about how this applies to the
6
general application of nonclinical test systems,
7
I'll take you back to a publication 1944 by Van
8
Winkel et al.
9
by FDA following the 1938 Food, Drug, and Cosmetic
10
Act, which gave FDA the authority to review safety
11
information on new drug candidates.
This was the first guidance offered
Up to that point, there was no guidance on
12 13
how to actually do that.
14
publication offered some suggestions, including
15
chronic tox studies in up to four nonclinical
16
species.
17
technologies have advanced, but the fact is
18
nonclinical testing remains a quintessential black
19
box.
20
The Van Winkel
Clearly, things have changed,
Essentially we take molecules of all types,
21
biologics, small molecules of varying complexity.
22
We put those through the test symptoms, generate
A Matter of Record (301) 890-4188
31
1
lots of data, and from those data we derive
2
conclusions around the outcome in terms of
3
nonclinical toxicity, yes or no.
4
Now, we have spent some time working in
5
various ways to build a framework to try to capture
6
the performance of contemporary nonclinical test
7
systems, and I'll show you one approach that we've
8
taken in our small group at Lilly.
9
This is a highlight of a nonclinical test
10
package for the development of a small molecule for
11
chronic use.
12
oncology drug with guidance from S9, this would be
13
a bit truncated.
14
model that I'm going to share with you, we used the
15
whole kit and caboodle, minus the repro and the
16
oncology study.
Now, as Natalie alluded to, for an
But nevertheless, to build the
17
So all of the general toxicology studies,
18
the safety pharmacology studies, we all distilled
19
down and put it into a box that we called
20
nonclinical testing.
21
If you think about how decisions are made
22
with these test systems, the molecule is advanced
A Matter of Record (301) 890-4188
32
1
from the selection as a candidate into the
2
nonclinical test system, and a result occurs.
3
declare it either not showing nonclinical toxicity
4
or showing nonclinical tox.
We
Now, obviously as complex as these studies
5 6
are, this is a continuum.
But the fact is we treat
7
the output in a dichotomous way, making decisions
8
around the fate of the molecule going forward.
9
if you don't get too hung up on this and work
10
through this with me, I hope to illustrate a few
11
points.
So
In those cases where we declare the compound
12 13
relatively safe or highly safe in the nonclinical
14
test systems, we advance the compound in a pretty
15
straightforward fashion to the clinic.
16
the clinic, we'll get the answer as to whether or
17
not there is clinical toxicity or not.
And once in
18
In those cases where we encounter toxicity
19
in the nonclinical test systems, it's not quite as
20
easy.
21
whether the findings that we're dealing with coming
22
out of the nonclinical tests appear to be
What we do at that point is we determine
A Matter of Record (301) 890-4188
33
1
manageable in the context of a clinical setting or
2
not.
3
If they are not, then we declare the
4
compounds unsafe and we terminate them.
5
advance to man, and for the purposes of our
6
modeling become useless to us, because we never get
7
the true outcome from those compounds.
8
for those compounds that we deem to be safe enough,
9
we do go into the clinic, and we determine clinical
10
They never
However,
toxicity, yes or no.
11
It's from this completed decision tree that
12
we derive our conclusions regarding our prediction.
13
And for every prediction in this tree, there are
14
two possible correct predictions:
15
clinical toxicity having predicted it; one, that we
16
get no clinical toxicity having seen safety in the
17
nonclinical test systems.
18
predictions, as you can see, are illustrated here.
19
one, that we get
The two wrong
From that, we get what we populate our
20
confusion matrix with as true positives and true
21
negatives and the false positives and false
22
negatives, respectively.
A Matter of Record (301) 890-4188
34
I won't go through the process of building
1 2
the model here today, but this is the framework
3
model that we use to illustrate the way the
4
nonclinical test systems operate in a typical
5
fashion.
6
the literature, giving estimates of sensitivity,
7
specificity, positive and negative predictive
8
value.
9
classes including oncology agents.
10
This includes data that was derived from
It included compounds that fall into all
Given the time frame of most of the
11
data sets that we use, I'll admit that this is
12
largely biased towards cytotoxic drugs in the
13
oncology arena.
14
performance of the models in general use is a
15
sensitivity of 70 percent, specificity of 50, with
16
a positive and negative predictive value of 38 and
17
80 percent, respectively.
18
At any rate, we estimate that the
It suggests to us that at the time a
19
molecule matriculates into development before any
20
GLP toxicology studies are conducted, that there's
21
a 30 percent chance of encountering a positive
22
clinical toxicity as we are defining a serious
A Matter of Record (301) 890-4188
35
1 2
adverse event. There's a publication in the literature from
3
Harry Olson and colleagues that was conducted as an
4
ILSI-HESI consortium exercise, and they looked at
5
the frequency of association of serious ADRs, and
6
from their data set estimated that about 40 percent
7
of the time those ADRs lead to termination of
8
development.
9
If that's correct, we would expect, from the
10
framework model, a 12 percent clinical safety
11
failure rate, which actually matches the estimate,
12
the most contemporary estimate, from Kola and
13
Landis, suggesting that while we don't anticipate
14
this model to be truly correct, it's at least
15
generally correct.
16
One thing you'll notice is the negative
17
predictive value because of the lower range of
18
prevalence is the dominant performance feature.
19
That is to say that these models do work to predict
20
human outcomes.
21
human outcome of safety.
22
predicting specific forms of toxicity.
They work best in predicting the They work less well in
A Matter of Record (301) 890-4188
36
One of the aspirations from the workshop is
1 2
to think about how we can improve that, and I've
3
got a few suggestions that we can consider.
4
let me just walk you through this decision tree,
5
because I want to illustrate a couple of important
6
points.
7
But
Again, just looking at the correct and
8
incorrect predictions, it's interesting to note
9
that attrition arises on the two arms of this tree
10
in different fashions.
When we start out with a
11
consideration of safety from the nonclinical test
12
systems, the attrition that we encounter is when
13
we're surprised with a false negative.
14
into clinical studies not expecting toxicity,
15
encountered it, and compounds of attrit.
We went
16
On the positive side of the nonclinical
17
decision tree, the attrition is actually coming
18
from the true positive events that are being
19
predicted by the nonclinical studies, and I want to
20
make that point in just a minute here.
21 22
But where we start, you can go up the tree on either arm.
We're either starting with the
A Matter of Record (301) 890-4188
37
1
presumption of safety, in which case we've got a
2
70 percent chance of being correct, or the
3
prediction of toxicity, the inverse of that, a
4
30 percent chance of being correct.
5
As we go up on the right side of the arm,
6
what we see is that after our GLP tox studies are
7
completed, we go from a prediction of safety of 70
8
to a prediction of safety of 80 percent.
9
doesn't seem like much, but there was only
That
10
30 percent uncertainty left around the safety, and
11
in fact we've just reduced a third of that
12
uncertainty through the course of our nonclinical
13
testing.
14
On the toxicity side of the things, we do
15
much less in terms of moving the needle.
We go
16
from a prediction of toxicity of 30 percent up to a
17
prediction of 38 percent.
18
stop and think about it, we don't go through the
19
process of considering manageability of a
20
nonclinical finding with the goal of showing how
21
clever we are at predicting the true positive
22
outcomes.
And in fact, when you
A Matter of Record (301) 890-4188
38
1
In fact, we go through this arm when we
2
believe the outcomes are going to be manageable in
3
the clinic or relying on the relative uncertainty
4
that remains that in fact what we're dealing with
5
is likely to be proven as a true positive.
6
Now, I want to show you just an example of
7
one such true positive and the investment that has
8
to be made in realizing that.
9
information regarding a TGF-beta R-1 kinase that we
10 11
And this is some
currently have in phase 2 development at Lilly. We were struck very early in the development
12
program around this TGF-beta inhibitor, as were
13
other companies that have explored this target
14
area, at the very dramatic effects that occur in
15
the nonclinical species, particularly in the form
16
of valvular changes in the heart with valvulopathy
17
appearing both in the rat and in the dog, as well
18
as inflammation, and ultimately rupture of the
19
great vessels.
20
This is a toxicity that clearly got our
21
attention.
It's gotten the attention of others
22
that have worked in this area, and we have seen
A Matter of Record (301) 890-4188
39
1
from publications that some companies have moved
2
away from this target out of concern for these
3
toxicities.
4
We did show through some PK/PD work the fact
5
that we could achieve tumor responses even with
6
intermittent dosing schedules, similar to what we
7
were getting with the daily dosing schedules that
8
yielded these nonclinical events.
9
That led us to invest in some additional
10
nonclinical work that showed that when we went to
11
an intermittent dosing schedule two weeks on and
12
two weeks off as opposed to daily dosing, in a
13
three-month study, we were able to show that we
14
could actually achieve no observed effect levels
15
for any change in the valves or great vessels.
16
Using that information, we were able to
17
build models that allowed us to convince ourselves
18
and others that by maintaining a disciplined
19
approach and exposures below that NOEL level, we
20
could achieve sufficient inhibition of SMAD to
21
warrant moving forward into the clinic.
22
That still required a comprehensive and
A Matter of Record (301) 890-4188
40
1
integrated cardiac monitoring program, including
2
echocardiography.
3
publication that I've cited here, we have seen
4
clinical benefit in glioma patients, and to date,
5
no sign of adverse cardiac events that were
6
characterized in the nonclinical model systems.
7
But I can tell you from the
Again, I don't want to suggest to you that
8
the model we've developed is necessarily right.
9
think as George Box said, "All models are wrong.
10
Hopefully some are useful."
11
both for the nonclinical test systems that we use
12
for predicting safety, as well as some of the
13
framework modeling that we've done to explore the
14
performance of these studies.
15
I
I believe that's true,
Just to wrap up on a few things that if we
16
want to get better at thinking about the prediction
17
of toxicity that we can consider -- Natalie alluded
18
to -- the questions of nonclinical species choices.
19
I've referred before to this paper by Harry Olson
20
and colleagues that show the distribution of the
21
correlation of findings in the rodent and
22
non-rodent, the non-rodent in blue, rodent in red,
A Matter of Record (301) 890-4188
41
1
and where they overlap with findings in both
2
species here in purple. That gave an overall concordance of 71
3 4
percent.
5
have a correlate in the nonclinical species.
6
is often misrepresented as a predictive value.
7
is not.
8
worse.
9
So 71 percent of the human toxicities did This It
The predictive value is actually much
We used an approach applying Bayes formula
10
to calculate a posterior probability score from the
11
sensitivity estimates, and from that you get
12
estimates of the predictive power of a finding that
13
occurs in the rodent or non-rodent.
14
sensitivity was 71 percent.
15
predictive value estimate is 38.
16
The
The positive
Here is a rodent finding that does not occur
17
in the non-rodent, a non-rodent finding that does
18
not occur in the rodent, and a finding that occurs
19
in both species associated with their relative
20
predictive estimate.
21 22
We were interested in seeing if this had any bearing in reality, and we were able to find only
A Matter of Record (301) 890-4188
42
1
one single data set in the literature that had
2
sufficient estimates of positive predictive value
3
to be able to compare to.
4
from Litchfield back in 1962.
5
the data or the analysis reveals.
6
That publication was Let me show you what
So again, here are the positive predictive
7
value estimates from the Olson paper and the
8
positive predictive values presented by Litchfield
9
in 1962 for a finding that occurs in either the
10
rodent or the non-rodent, the rodent but not the
11
non-rodent, the non-rodent but occurring in the
12
rodent, and it looks pretty consistent.
13
Look what happens when you get to a finding
14
in the rodent and the non-rodent.
15
this disproportionate predictive power may reflect
16
what toxicologists have long held to be true, that
17
a finding in both nonclinical species has
18
additional predictive capability.
19
We believe that
Also, questions have arisen in some of what
20
Natalie presented, as well as in some of the
21
publications I think leading up to the workshop,
22
around how long we have to dose in nonclinical test
A Matter of Record (301) 890-4188
43
1
systems to determine those findings that have
2
clinical relevance.
3
Going back again to the Olson paper, I think
4
one of the biggest surprises for most people was
5
that the Olson publication suggested that over
6
90 percent of all clinically relevant human
7
toxicities were revealed in studies of 30 days
8
duration or less.
9
incrementally to that to a total of 98 percent.
10
The three-month studies added
I don't think most people knew what to do
11
with this, until a recent publication out of Japan
12
by Chihiro Tamaki and his colleagues, looking at
13
several hundred compounds that were registered in
14
Japan suggested a similar outcome that most
15
findings display themselves in studies of up to
16
three months in duration.
17
One of the things that's also illustrated in
18
the Olson paper is how much of that early phase
19
correlation was derived from the safety
20
pharmacology studies, and again, coming back to
21
Natalie's point to the question of what role they
22
might play going forward.
A Matter of Record (301) 890-4188
44
1
I've got a yellow light here, so I want to
2
move pretty quickly.
One of the things that we're
3
likely to get into is how we can supplement what we
4
do now with other tests that could add to this
5
predictive capability.
6
one warning and caution around that.
And I just want to share
7
I've got two matrices here.
One of them,
8
the framework that we've just talked about,
9
representing standard safety packages; the other,
10
let's call a specialty tox study that we can
11
supplement with that.
12
It's important for us to think about how we
13
combine the data from these multiple test systems.
14
If we combine them using "and" -- so a positive is
15
a positive in both assays -- what you see is you
16
actually maximize your specificity and the positive
17
predictive value of that combination of assays.
18
That raises your false negative risk, the risk that
19
you will have, a finding in the clinic that you
20
were not warned about.
21 22
Combining them with "or" you get the opposite effect.
You increase the sensitivity,
A Matter of Record (301) 890-4188
45
1
which maximizes the NPV, decreasing that false
2
negative risk, but raising the false positive risk. Our approach at Lilly is to focus on the
3 4
negative predictive value estimates, to work with
5
assay systems that are high specificity in
6
discovery, recognizing that we'll pick up some of
7
those false negatives in future testing. One caution is not to overtest because the
8 9
more of these tests you combine, the more you can
10
accumulate false positives, which is another way to
11
deny therapy to patients, and our advice to keep it
12
simple.
13
At Lilly, we've reduced this down to two
14
orthogonal assays, volume of distribution and
15
cytotoxicity in rat primary hepatocytes.
16
is to drive compounds to be less tissue penetrant
17
and less cytotoxic inherently, and as a result,
18
less overall adverse effects.
19
with pretty good success.
20
Our goal
And we've done that
Let me just finish up by acknowledging my
21
colleagues at Lilly, those that have been
22
particularly helpful in this predictivity analysis
A Matter of Record (301) 890-4188
46
1
that I've referred to, my Lilly colleagues that
2
worked on the data sets for the TGF beta R-1 kinase
3
and the toxicity screening approach.
4
I'll end.
Thank you.
5
(Applause.)
6
DR. MCKEE:
7 8 9 10
With that,
And our next speaker is
Dr. Richard Brennan from Sanofi. Presentation – Richard Brennan DR. BRENNAN:
Thanks, Amy.
Good morning, everybody.
I'd like to thank
11
the organizers for the opportunity to speak today.
12
What I thought I'd do today is given that we now
13
have about 15, 16 years of experience with small
14
molecule kinase inhibitors on the market, do a
15
little bit of a retrospective overview about what
16
we've seen in terms of adverse effects of these
17
compounds in the wild, if you like, with broad
18
application, and how that is reflected in our
19
preclinical testing, whether it's -- we do have
20
direct correlations, whether they are indications
21
or secondary assays beyond the standard set of
22
assays that we can use to identify some of those,
A Matter of Record (301) 890-4188
47
1
or the things where we've done historically and did
2
do a pretty bad job of predicting those toxicities.
3
Moving in light of what Natalie was saying,
4
I'll also talk a little but about the problem of
5
compound promiscuity and what that means in terms
6
of adverse effects, and a couple of ideas about how
7
we might be able to address that going forward.
8 9
This is a graph I put together actually from some data that's available on the Web, from the
10
Blue Ridge Medical Institute.
11
the number of compounds that have entered the
12
market at various stages, and these are
13
specifically small molecule kinase inhibitors.
14
And really, it shows
You can see that the first compound was
15
approved by FDA in 1999, and from then on, for the
16
next few years, there were dribs and drabs,
17
occasional compounds coming on the market.
18
last eight or so years, we've really seen quite a
19
flood of these compounds coming on to the market,
20
and that continues.
21
been approved this year, the small molecule kinase
22
inhibitors.
In the
Two compounds have already
A Matter of Record (301) 890-4188
48
1
We tend to think of Gleevec as being the
2
first really paradigmatic small molecule kinase
3
inhibitor.
4
was actually this compound, which is an mTOR
5
inhibitor.
6
It's an actual product.
7
anti-fungal agent, and later came on the market as
8
an immune-suppressing agent.
9
In truth, the first one on the market
This compound was actually discovered. It was discovered as an
So really, Gleevec is the first example of a
10
compound that was specifically targeted against a
11
particular kinase for an oncology indication; so
12
really, it's the paradigmatic beginning of this
13
history, if you like.
14
After Gleevec, a couple of EGFR inhibitors
15
came on the market, again, for oncology
16
indications, non-small cell lung cancer and so on.
17
But really, the story here is that although we do
18
have quite a lot of experience with some kinase
19
inhibitors on the market, more than half of the
20
compounds on the market -- to date there are
21
29 compounds, small molecule kinase inhibitors,
22
currently approved and on the market in the
A Matter of Record (301) 890-4188
49
1
U.S. -- more than half of those have been improved
2
in the last five years.
3
a limited amount of information available with the
4
broadest set of compounds.
So there's actually quite
Just a few thoughts about clinical adverse
5 6
effects of these compounds.
Most of the compounds,
7
kinase inhibitors, have been approved for oncology
8
indications.
9
comparison to the standard cytotoxic therapies.
They are relatively safe in
10
But typically, there's a high tolerance for adverse
11
effects in oncology than you would see in other
12
indications, chronic indications, non-life-
13
threatening indications. So as kinase inhibitors begin to be approved
14 15
for other indications, the tolerance for adverse
16
effects is going to decrease.
17
think more carefully about how we do preclinical
18
prediction of these adverse outcomes and dose
19
setting for these types of indications beyond
20
cancer.
21 22
And we may need to
Many of the adverse effects that we're seeing with these compounds are actually target
A Matter of Record (301) 890-4188
50
1
related, and some of the more predictable ones are
2
target related, at least the ones that are
3
predicted today.
4
epidermal growth factor receptor inhibitors, which
5
typically present with an acneiform rash, and the
6
VEGF inhibitors that Natalie mentioned, which are
7
linked to hypertension.
8 9
Some examples would be the
One observation, though, is that the adverse effect profile of these compounds, even if they're
10
targeted against the same kinase, can be quite
11
different.
12
that are observed is dependent on the profile of
13
activity of those compounds across the spectrum of
14
kinases.
15
And the severity of the adverse effects
As Natalie mentioned, these compounds can be
16
quite promiscuous.
17
kinases.
18
reasons for that given the set of targets.
19
Then they target many different
And there are structural functional
An example here is in the ABL kinase
20
inhibitors.
There's quite a range of adverse
21
effects, although they ostensibly target the same
22
molecules between imatinib, nilotinib, and
A Matter of Record (301) 890-4188
51
1 2
dasatinib. Myelosuppression is seen with all of them,
3
but the severity of that is related to affinity for
4
c-Kit to some level.
5
That's seen again at a high level with some of the
6
compounds than others, and that's related to,
7
again, affinity for a secondary target, PDGFR.
Edema is another side effect.
8
Again, pleural effusion is seen quite
9
commonly with dasatinib, but is not seen with
10
either of the other two ABL kinase inhibitors to at
11
least at the same degree.
12
those compounds.
13
again, an off-target effect of dasatinib.
14
be related to the SAR kinase activity for that
15
compound, but it may actually be a result of the
16
general promiscuity of dasatinib relative to the
17
other two compounds.
18
It's quite rare with
And that's presumably related to, It may
I want to just talk about some of the common
19
adverse effects we've seen clinically with the
20
compounds and how those are reflected or not in
21
preclinical species.
22
Dermatotoxicity is actually pretty well
A Matter of Record (301) 890-4188
52
1
predictable now.
2
with epidermal growth factor receptors inhibitors,
3
and in fact has been used as a clinical correlate
4
for anticancer activity.
5
these clinical trials for these compounds, a
6
dose-to-rash protocol, where you increase the dose
7
until you actually see this adverse effect, as
8
there is some correlation with the anticancer
9
activity for that.
10
This is a very common observation
There's been in some of
These outcomes do have correlates in
11
preclinical species.
12
and dog.
13
what's seen in the clinic.
14
slides from rats treated with an epidermal growth
15
factor receptor inhibitor.
16
we're seeing hyperkeratosis in the skin, as well as
17
acanthosis, and some inflammation in the hair
18
follicles.
19
in a clinical presentation of these adverse
20
effects.
21 22
They're seen in mouse, rat,
And they're actually quite reflective of So this is a couple of
And as you can see,
And that's very similar to what's seen
Other areas where we do a pretty good job in preclinical testing, developmental reproductive
A Matter of Record (301) 890-4188
53
1
toxicity.
2
particularly against oncology indications because
3
of their critical role in proliferative pathways
4
and cell survival pathways and vascularization
5
pathways and differentiation pathways.
6
These compounds are targeted
We would expect those types of pathway
7
effects to have adverse impacts on development of
8
the fetus, and that is actually seen.
9
review of 33 marketed compounds, that includes both
In a recent
10
small molecule and large molecule compounds
11
targeted against kinases, 32 of 33 marketed kinase
12
inhibitors had label indications for pregnancy
13
category C or D, C being that adverse developmental
14
effects are seen in animals, and D being that
15
they're actually seen in humans.
16
The one outlier there is a compound that's
17
targeted for macular degeneration and is dosed by
18
direct injection into the eye.
19
not seeing any effects with that compound.
20
That's why you're
By far, the greatest level of experience is
21
with imatinib.
22
longest time.
Gleevec has been on the market the So in a recent review of
A Matter of Record (301) 890-4188
54
1
developmental effects in clinical practice with
2
this compound, Abruzzese et al. looked at a number
3
of reports of pregnancy for people that had been
4
exposed to imatinib for treatment.
5
reports were available where pregnancy had happened
6
from a male that had been taking Gleevec.
7
seems to not be an issue in terms of developmental
8
effects; 148 normal deliveries resulted from those
9
exposures.
In males, 150
That
10
It's not quite such a good story in women
11
exposed to Gleevec who are pregnant, particularly
12
during periods of organogenesis.
13
exposed during pregnancy, there were 128 normal
14
deliveries, 24 spontaneous abortions -- that's just
15
slightly above the expected normal rate -- and
16
about 15 births with abnormalities that include
17
skeletal organogenesis, vascular defects, as well
18
as low birth weight.
19
At 167 women
Developmental reproductive effects for these
20
compounds are very well predicted.
21
the rat and the rabbit typically in preclinical
22
testing.
A Matter of Record (301) 890-4188
They show up in
55
1
A couple of other areas that are fairly well
2
predicted.
3
pathways that are critical in cellular
4
proliferation, so typically you see GI toxicology,
5
hematopoietic toxicity, and lymphopenia.
6
common in certain classes of kinase inhibitors, but
7
rat, dog, and non-human primates are all
8
susceptible to these effects.
9
preclinical species may be slightly more sensitive
10
in terms of observations of hematopoietic toxicity
11
in preclinical testing that don't show up in the
12
clinic, and that may be actually dose-related in
13
the safety testing.
14
As mentioned, these compounds target
It's more
And in fact, the
So where do we not do quite such a good job?
15
So ocular toxicity, again, it's pretty common with
16
certain classes of kinase inhibitors.
17
not particularly severe or life-threatening.
18
one very severe observation is seen with MEK
19
inhibitors, and that's retinal vein occlusion.
20
It's often The
That doesn't show up in the standard
21
preclinical species, rat, mouse, and dog.
22
retrospective testing has shown that it actually
A Matter of Record (301) 890-4188
But
56
1
can be observed in the rabbit, and that's probably
2
because the rabbit eye structure is more closely to
3
similar to man than the other preclinical species.
4
Other clinical effects on the posterior
5
segment of the eye include hemorrhage and
6
retinopathy.
7
actually seen with several of the kinase inhibitors
8
across different target classes, so it's a little
9
bit difficult to assign a particular mechanistic
10
Those are more rare.
They're
interpretation on those.
11
Epidermal growth factor receptor is
12
expressed quite highly in the eye, and so there are
13
quite common observations with EGFR inhibitors in
14
the eye, and those include conjunctivitis.
15
preclinical correlates of that include corneal
16
thinning in both the rat and the dog, can be
17
observed with these compounds.
18
also observed in the dog.
19
The
And lens opacity is
A less simple example is with crizotinib.
20
About 60 percent of patients in late stage clinical
21
trials report visual disturbance on taking
22
crizotinib.
So obviously, a rat or a dog can't
A Matter of Record (301) 890-4188
57
1
tell you he's having difficulty seeing, strange
2
observations.
3
in the rat, using electroretinography, you can see
4
defects in signaling in the retina that correlate
5
with light/dark adaptation in this animal.
6
there is some ability to detect effects, although
7
it's not reflective necessarily of what's observed
8
in the clinic.
9
But in retrospective testing again
So
Other preclinical effects can be picked up
10
in species, and those include lens opacity seen
11
with mTOR inhibitors in the rat and with other
12
compounds in the dog.
13
to certain effects like keratitis, hemorrhage, and
14
so on, as well.
15
And the rat is susceptible
So cardiovascular effects have been quite
16
problematic with certain of the kinase inhibitors,
17
particularly VEGF receptor inhibitors.
18
many different forms of cardiovascular effects.
19
Since we're talking about small molecules, these
20
small molecules can have the same off-target
21
non-kinase effects that other small molecules may
22
have, and that includes ion channels, cardiac ion
A Matter of Record (301) 890-4188
There are
58
1 2
channels like the hERG channel. Those can cause QT prolongation,
3
arrhythmogenic effects.
Those are typically picked
4
up quite well in preclinical testing.
5
in vitro or safety pharmacology tests will pick up
6
effects along those lines.
Either
7
For the VEGF inhibitors particularly,
8
hypertension, bleeding, decreases in ejection
9
fraction, and ultimately heart failure have been
10
observed.
11
testing and were actually quite a surprise finding
12
for these compounds in the clinic.
13
These were not picked up in preclinical
Again, retrospective testing can go back and
14
find evidence, and you can have additional testing
15
to pick up these effects.
16
hypertrophy and ejection fraction changes and so on
17
can be seen in rat, but only if you dose the
18
compounds neonatally, and then you will see these
19
effects both in the young offspring.
20
those animals progress to adulthood, you can pick
21
them up.
22
So ventricular
And also, as
But again, only when dosed neonatally.
Ventricular functions can be monitored in
A Matter of Record (301) 890-4188
59
1
rats and in dogs.
2
those effects, they can be observed.
3
If you look quite closely at
With the MEK inhibitors, and I think we'll
4
hear some more about these in more detail in one of
5
the later talks, again cardiomyopathy ejection
6
fraction changes and cardiovascular failure have
7
been observed particularly with MEK inhibitors.
8
Those actually can be monitored in rat by looking
9
at ejection fraction changes, and we'll hear a
10 11
little bit more about that later. Congestive heart failure and, again, left
12
ventricular dysfunction are observed with ABL
13
inhibitors.
14
mice, in preclinical testing.
15
dysfunction and cardiomyoctye loss, and apoptosis
16
in those animals.
17
And these actually can be picked up in And up here is
Another preclinical finding, again with
18
quite a broad range of kinase inhibitors is
19
hypothyroidism.
20
VEGF inhibitors.
21
vascularized, and the vasculature in the thyroid
22
remains VEGF dependent into adulthood, in contrast
It's particularly an issue with The thyroid is highly
A Matter of Record (301) 890-4188
60
1
for vascularization in other tissues, so there is
2
actually a concern for these compounds.
3
This can be picked up in rats and mice and
4
observed as regression of the thyroid capillaries,
5
so lack of vascularization of the thyroid and
6
increased thyroid stimulating hormone after 21 days
7
of treatment.
8 9
Interestingly, in the mice, this is reversible on kinase withdrawal, and that's also
10
the clinical experience, that these thyroid effects
11
can be reversible once treatment is withdrawn.
12
So where don't we do very well?
What's
13
poorly predicted?
14
drug-induced liver injury is a continuous issue for
15
all small molecule drugs.
16
kinase inhibitors on the market particularly, 16 of
17
those kinase inhibitors have specific warnings on
18
the label for a potential for hepatotoxicity.
19
there's also clinical evidence for hepatotoxicity
20
for one other kinase inhibitor, dasatinib.
21 22
Well hepatotoxicity,
When we look at the
And
If we look at how that appears in preclinical testing, out of 26 total marketed small
A Matter of Record (301) 890-4188
61
1
molecule kinase inhibitors, 17 of those molecules
2
that were positive for hepatotoxicity in humans,
3
only 7 showed up as hepatotoxic in the rat, and
4
actually 2 false positives were observed in the
5
rat, so a pretty poor predictive value in that
6
case.
7
Looking at second species, out of 11 of
8
those compounds that were also looked at in dog,
9
only 6 of them, so just a little bit more than
10
50 percent, appeared as positive in the dog.
11
again, we don't do a very good job of predicting
12
hepatotoxicity for these compounds in our
13
preclinical species.
14
So
Similar for kidney toxicity, so 10 of the
15
marketed compounds are positive for kidney toxicity
16
in humans.
17
because of vascularization, hypertension, and
18
resulting proteinuria.
19
10 compounds only five appear in either rat or dog
20
as nephrotoxic, and there's quite a high number of
21
false positives in rat and dog.
22
aren't observed in the clinic as toxic in humans
Some of that's driven by VEGF, again,
But out of those
A Matter of Record (301) 890-4188
Six compounds that
62
1 2
appear in rat or dog. A couple of other quite rare but sometimes
3
severe, and life-threatening in some cases, adverse
4
effects.
5
non-infectious pneumonitis is observed with kinase
6
inhibitors in the clinic.
7
is life-threatening.
8
different kinase inhibition.
9
Interstitial lung disease and
This is quite rare, but
It appears with a number of
There are human specific factors that are
10
risk factors for this.
11
includes pre-existing pulmonary fibrosis.
12
actually, people of Japanese ancestry have a higher
13
incidence of this than Caucasians.
14
That includes smoking, it
It's a complex diagnosis.
And
It involves
15
scanning, as well as clinical testing.
You do see
16
occasional lung lesions in the rat in preclinical
17
testing with kinase inhibitors, but they're really
18
not reflective of this interstitial lung disease,
19
and it's quite sporadic that they're observed
20
preclinically.
21
Similarly, for reversible posterior
22
leukoencephalopathy, that's really not picked up in
A Matter of Record (301) 890-4188
63
1
our preclinical testing at all. So it's a mixed bag.
2
Some things we
3
predict.
Some things we don't predict.
As I
4
mentioned, quite often the things that we do a good
5
job of predicting are things that are directly
6
related to the target biology. This complex map here is really a diagram of
7 8
kind of a summary of some of the major points of
9
epidermal growth factor signaling.
So here's
10
epidermal growth factor receptor.
It's upstream
11
signaling from various ligands and then downstream
12
signaling through signal transduction cascades, to
13
changes in gene expression, down to effects on cell
14
proliferation, cell survival, and so on. You can see it's quite complex.
15
A lot of
16
this information, a lot of this knowledge, has been
17
developed post hoc.
18
early testing that epidermal growth factor is
19
important in some of these processes, we've gone
20
back and dissected that in great detail and quite
21
well.
22
Once we've found out from some
So we can do now -- we can say, well, any
A Matter of Record (301) 890-4188
64
1
effects on this pathway hitting the epidermal
2
growth factor, or actually any of the intervening
3
steps in this pathway, could quite easily lead to
4
changes in cell proliferation, changes in cell
5
survival, that will have effects on proliferating
6
tissues where epidermal growth factor is driving
7
proliferation of those tissues.
8
the skin, includes the gut, and so on in those
9
areas, that are quite well detected in a
10
preclinical species.
11
you can see.
12
And that includes
But it is quite complex, as
One issue has been that if we've got a
13
target, we understand it quite well, we can predict
14
the outcome of hitting that target.
15
alluded to, our compounds are typically not hitting
16
just one target, and in fact they can be quite
17
promiscuous.
18
But as Natalie
Natalie showed a very similar analysis
19
across targeted kinase inhibitors.
20
looks across a broader set of kinase inhibitors,
21
includes the drugs on the markets, but a broader
22
set of test compounds of things that are in
A Matter of Record (301) 890-4188
This analysis
65
1
preclinical testing, things that are in early
2
clinical stage testing, looking across 300 kinases
3
for in vitro activity of these compounds. What you can see is the bright purple color
4 5
is where there's a compound that's inhibiting that
6
kinase.
7
diagram is a couple of things.
8
kinases are quite difficult to inhibit, and there
9
are not many compounds, or in some cases, any
So what we can see from this complex One is that certain
10
compounds that inhibit that kinase among our
11
current set of compounds.
12
Conversely, some kinases themselves are
13
quite promiscuous and are inhibited by many
14
compounds.
15
perspective, some compounds are quite specific in
16
their activity and have a very limited range of
17
activities, whereas some compounds can hit a very
18
broad range of kinases.
19
of the drugs that are on the market.
Similarly, from the compound
That's also includes some
20
Imatinib was initially targeted for the
21
BCR-ABL fusion in chronic myelogenous leukemia.
22
also hits ABL and PDGFR.
It
It was already known when
A Matter of Record (301) 890-4188
66
1
it came on the market, but actually later testing
2
has shown it hits a further set of compounds
3
including c-Kit, cFms, Lck, and so on.
4
beneficial in terms of treating cancer; you can hit
5
more targets that may affect different types of
6
cancer or may kind of have synergistic effects on
7
particular cancers.
8
risk of having an unintended off-target, adverse
9
effect.
That can be
But it also increases your
Similarly for sunitinib, it's a VEGF-1
10 11
targeted compound.
Also, it hits a number of other
12
kinases.
13
the potential for adverse effects both on and off
14
target.
And again, that promiscuity can relate to
15
So how do we deal with that idea of
16
promiscuity and predicting the adverse effects that
17
may be related to hitting those targets?
18
is that we don't necessarily know completely the
19
biological function of many of those targets that
20
these compounds are hitting.
21
know completely the spectrum of activities of our
22
compounds.
One issue
And also, we may not
We don't necessarily screen across 300
A Matter of Record (301) 890-4188
67
1
or the full 500-plus kinases that are expressed in
2
the body.
3
This is analysis from Olaharski et al. at
4
Roche a few years ago, addressing a particular
5
issue of genotoxicity of kinase inhibitors,
6
particularly chromosome instability and the
7
appearance of micronuclei in their preclinical
8
testing.
9
What they looked at was a series of
10
compounds that had all been tested for
11
micronucleus, may or may not cause micronucleus in
12
their testing, and then they assayed those
13
compounds, 113 compounds, against a large panel of
14
kinases and looked at the pattern of activity of
15
those compounds across kinases, and then used some
16
machine learning approaches, combination of Random
17
Forest and Support Vector Machine algorithms, to
18
pull out from that data set a pattern of activities
19
that was able to predict the appearance of
20
micronucleus.
21 22
What they found was a pattern across 21 separate kinases that could predict the outcome of
A Matter of Record (301) 890-4188
68
1
micronucleus in their assays with about 85 percent
2
accuracy.
3
From that panel of 21 kinases, it's actually
4
quite informative if you go back and read the
5
papers, some of them were well known to be involved
6
in cell cycle control, mitosis, and genome
7
stability types.
8
surprise.
9
outcome, they weren't previously known.
But other kinases were quite a
Although they were predictive of the And still,
10
some of those are not really -- have been
11
associated with cell cycle with DNA repair, with
12
mitosis, or any of those outcomes.
13
So it can be quite difficult to predict, if
14
you like, the full biological range of those
15
targets that may be causing adverse effects
16
downstream.
17
approaches and build predictive models that can
18
help you address specific issues in this case.
19
But you can take bioinformatic
The second example here is from a
20
publication from Yang et al. in 2010, and it's
21
taking a broader kind of scope, looking across
22
literature, across drug labels, to identify what
A Matter of Record (301) 890-4188
69
1
are the key adverse effects that are being
2
observed, and then looking for the 20 kinases and
3
finding as much about their kinase activity, what
4
kinases inhibit is possible.
5
kinases, there are actually 266 distinct targets;
6
admittedly quite a high concentration in vitro
7
that's probably not clinically relevant in some
8
cases.
9
And across those 20
What they're able to do is do some modeling
10
and connect the adverse effects that are seen in
11
the clinic to the target activities, and use that
12
to build, again, a predictive model that given a
13
scope of target activities for a compound, then
14
links with more or less confidence to particular
15
clinical outcomes.
16
They can assign confidence scores to those.
17
They can go back and identify known outcomes such
18
as EGFR with the appearance of rash, but also
19
identify putative, hypothetical, new adverse
20
effects that maybe will appear with particular
21
patterns of activity.
22
So just to finish up here with some
A Matter of Record (301) 890-4188
70
1
conclusions, small molecule kinase inhibitors,
2
they're a clinically important and an effective
3
addition to our portfolio of pharmaceutical
4
intervention for oncology.
5
increasing market.
6
50 billion a year for kinase inhibitors in terms of
7
commercial activity on the market.
I think we're approaching
For drug companies, it's been a very
8 9
There's actually an
lucrative, effective approach.
For patients, it's
10
been very effective.
11
available that have a lot of efficacy in the
12
clinic.
13
kinases attacked and more kinase inhibitors come on
14
the market.
15
There are new treatments
So we're going to continue to see more
Generally, the adverse effects are less
16
severe and more manageable than we've seen
17
historically with standard cytotoxic chemotherapy.
18
There are, though, some life-threatening adverse
19
reactions.
20
preclinical safety profiling.
21
earlier, as kinase inhibition moves into other
22
indications such as rheumatoid arthritis, diabetes,
They're not always well predicted by a
A Matter of Record (301) 890-4188
And as I mentioned
71
1
Parkinson's disease, those adverse effects may be
2
less acceptable for those indications.
3
have to do a better job of identifying safer drugs
4
for those indications.
And so we
5
Many adverse effects can be understood and
6
predicted based on the target, but we don't always
7
know what the target is or what the full range of
8
targets are.
9
functions of those targets.
10
And we don't always know the
So kinase inhibitors can be promiscuous.
11
The targets can be promiscuous, and there's a lot
12
of things that we don't know yet about the biology
13
of some kinases.
14
There are some potential applications for
15
computational systems, biology approaches.
16
the issues that we face here is the availability of
17
comprehensive, well-annotated data sets.
18
our companies, we all have a lot of information on
19
kinase inhibitors that target profiles that have
20
preclinical effects and some of the early clinical
21
effects.
22
One of
So within
To do a good job of doing these models, we
A Matter of Record (301) 890-4188
72
1
need access to broad range of that data, broad
2
range of compound activities across many targets
3
and be able to associate both with preclinical
4
outcomes and clinical outcomes.
5
potential here for pre-competitive data sharing in
6
this area that could help us build those models.
7
With that, I'll finish.
8
(Applause.)
9
DR. MCKEE:
Okay.
So there's some
Thank you.
So we are scheduled for a
10
break right now, and we'll be starting on the dot
11
at 9:45.
(Whereupon, at 9:22 a.m., a recess was
12 13 14 15 16
Thank you.
taken.) DR. JANNE:
Okay, we'll try to get started
here; if everyone can take their seats please. We have two presentations in this session,
17
and then that will be followed by the panel
18
discussion.
19
Genentech.
20 21 22
Our first speaker, Dr. Dambach from
Presentation – Donna Dambach DR. DAMBACH:
Good morning, everybody.
Again, I'd also like to thank the organizers for
A Matter of Record (301) 890-4188
73
1
inviting me.
I hope what I present today, which is
2
a snapshot of our portfolio at Genentech and our
3
experience with kinase inhibitors, you'll find
4
useful, and will invoke some discussion.
5
One thing I have noticed from the other
6
talks is you'll notice some themes in what I'll
7
present in terms of learnings and experiences, and
8
hopefully, by the way my program is set up, you'll
9
see that I try to touch on some of the key goals or
10 11
themes of those goals. This work is really an evaluation that I
12
took on to determine whether our safety lead
13
optimization strategy, that we had put forward
14
about seven or eight years ago, was actually
15
demonstrating value.
16
So what I hope to show you today is to bring
17
to you an awareness of what our strategic approach
18
is for safety, lead optimization for small
19
molecules.
20
exclusively on kinases, but we use this base case
21
strategy for all of our small molecules.
22
And in particular, this is focused
The other thing that I hope to demonstrate
A Matter of Record (301) 890-4188
74
1
to you is what impact we've had and what challenges
2
we still face, not only with regard to identifying
3
the lead candidate, mitigating risk, but also with
4
regard to translation of the identified toxicities
5
to the clinic. Finally, and most importantly, I'm not going
6 7
to go into specific case studies, but one of the
8
things that we've learned is that it's so very
9
important for us to investigate to try to
10
understand the cause of the toxicities, so that we
11
can really make good decisions around how to move a
12
program forward or identify a backup; as well as,
13
and Tom alluded to this, constantly iterating what
14
happened in the clinic back to our model so that we
15
can improve our surrogate models. Now, I want to acknowledge -- I'm not going
16 17
to acknowledge my staff.
18
the animals that sacrificed their lives that I want
19
us all to remember, that they are the central core
20
of what we do.
21
though.
22
I'm going to acknowledge
My staff does look like that,
(Laughter.)
A Matter of Record (301) 890-4188
75
DR. DAMBACH:
1
So what I'm going to go
2
through is these are various snapshots, high level,
3
I'm going to go through what our portfolio looks
4
like.
5
attrition.
6
each stage of development and where we had
7
attrition, how we made decisions.
8
move on to how we translated AE determination for
9
those molecules in the clinic.
10
I'm going to go through a series of steps of I'm going to spend a lot of time on
And then I'll
I wanted to give you one example of how we
11
make decisions around what large non-rodent species
12
we use and some of the struggles we've had
13
internally, and then I'll bring it all together.
14
As you know, Genentech was founded as a
15
biopharmaceutical company, although Tarceva was a
16
small molecule that had developed and licensed.
17
But the true small molecule development at
18
Genentech is about 10 years old.
19
years, we've predominately focused on kinases, and
20
we've looked at over 30 different targets.
21
are either historical or active.
22
In those 10
These
They've been predominately, as you've heard,
A Matter of Record (301) 890-4188
76
1
in the oncology space, but we have a lot of
2
non-oncology targets.
3
chemical space, diverse target diversity, as well
4
as disease diversity.
5
They've covered diverse
Our overall small molecule lead optimization
6
strategy is one where we want to try to mitigate
7
risk by partnering very early in the discovery
8
process with our efficacy, our research leads, DMPK
9
and pharmaceutics scientists, so that we can
10
actually integrate all of those functions into
11
identification of our lead.
12
dedicating specific scientists, specific
13
toxicologists, as discovery toxicologists and
14
investigative toxicologists.
15
different styles.
16
as well.
And we do that by
So we have three
We have regulatory toxicologists
17
What's really important here is to
18
understand what our goals are because this is what
19
is the flavor of the output.
20
important to what we're trying to do here today,
21
but you'll see some of it may not 100 percent
22
inform what we would like in the clinic for these
And some of this is
A Matter of Record (301) 890-4188
77
1 2
kinase inhibitors. So our goals are to minimize attrition,
3
especially in phase 1, due to toxicity.
So what
4
we're trying to do is attrit molecules before they
5
even get into phase 1.
6
and to understand translatable risk, so that when
7
we decide to advance a molecule, we want to make
8
sure that we're advancing a molecule that has a
9
predictive pattern or that we can understand or
And then to inform hazards
10
characterize the risk.
11
these are things to keep in mind as we go through
12
the data.
13
So those are the goals, and
The other thing I want to reiterate, I'm
14
going to spend a couple of slides talking about the
15
philosophical approach.
16
before, but it's real and it makes the foundation
17
of our decision-making.
18
And you've heard this
In the small molecule space, as we're trying
19
to identify, characterize, and de-risk, we do that
20
during the chemistry design period.
21
chemical characteristics are set in the molecule,
22
there's not much we can do from a safety
A Matter of Record (301) 890-4188
Once the
78
1
perspective.
2
rationale, as Tom mentioned, and we do that.
3
basically, we try to get out as much risk as we can
4
during the design process.
5
We can try to alter dose regimens and But
The two things we focus on very heavily, and
6
you'll see in our design strategy, is we focus on
7
on-target undesired toxicities and the off-target
8
toxicities.
9
off-target toxicities are all based on the chemical
And our philosophy is that the
10
and the phys-chem characteristics.
11
these very closely, and we partner very closely
12
with our chemistry colleagues around these.
13
So we look at
Here's our strategy on one slide.
This is
14
our base strategy.
15
of heavy lifting and a lot of front loading in the
16
discovery space because it's a period when there's
17
the ability to change the molecule, and it's
18
relatively inexpensive.
19
As I've told you, we do a lot
Here's our basic strategy.
One of the first
20
things we do, and we're not alone in doing this, is
21
we do what's called a target safety assessment.
22
This is before there's any chemical matter.
A Matter of Record (301) 890-4188
79
We basically sit down with clinicians, we
1 2
sit down with the researcher, we review the
3
literature, we look at genetically engineered
4
models, and we ask ourselves, what are the
5
potential on-target toxicities, and how do they
6
relate with regard to the therapeutic area?
7
Because as you know, there's different risk-benefit
8
for different areas, or different co-therapies and
9
co-morbidities associated with these different
10
areas. So we very carefully evaluate what are the
11 12
things that we need to be concerned about from a
13
target safety perspective.
14
If you look down, as we get to the target
15
screening, you'll see that number 6, we customize
16
our lead optimization strategy based on the actual
17
target.
18
determined that we have to add a counter screen,
19
for kinases often it's hematopoietic counter
20
screens, we'll add that, but it's specific for each
21
program.
22
So if in our target assessment, we had
In red are the five basic functionalities
A Matter of Record (301) 890-4188
80
1
that we build into every lead optimization
2
strategy.
3
is building on something that Tom talked about.
4
Notice how we're focusing on the NPV.
5
to eliminate as much as possible because we know
6
there's value in advancing a what we call clean
7
molecule because we're not always certain of the
8
translatability of the things that we identify.
9
And what I want you to notice about this
We're trying
So again, a common theme, our number one,
10
especially with kinases, is reduce promiscuity and
11
increase selectivity, and I'll get into that very
12
briefly in the next slide.
13
We use, again, targeted screens, like kinase
14
screens.
If it were a protease, we would use
15
protease screens.
16
pharmacology screens very heavily.
17
again heard from Tom, reduce intrinsic
18
cytotoxicity.
19
between intrinsic cytotoxicity in vitro with some
20
other characterizations and relating it to exposure
21
to translation in vivo, at least in preclinical
22
species.
And we also use the secondary We also, as you
There's a pretty good correlation
A Matter of Record (301) 890-4188
81
We focus on cardiovascular risk, both EFis
1 2
and hemodynamic.
And we also work very closely
3
with our ADME colleagues to reduce ADME risk,
4
really around TDI, around distribution, really
5
understanding distribution, transporters, and
6
reactive metabolites.
7
life-threatening conditions, we put in a screen,
8
again not unlike other pharmaceutical companies,
9
where we try to eliminate gene tox.
And for those non-
When we do identify risk, we do dedicate
10 11
resources to investigate and characterize so we can
12
make decisions, and I'll show you some examples of
13
that.
14
generate, we do use that to inform not only our IND
15
trials, but also our clinical monitoring.
16
And then all of this information that we
Here are some of the high focus areas that I
17
just touched on.
18
know is let's talk about kinases and this idea of
19
high selectivity, low promiscuity.
20
And what I want to just let you
We have some criteria that we set, and again
21
this may not be unique for Genentech, but our
22
criteria is that we like to see at least a
A Matter of Record (301) 890-4188
82
1
hundredfold selectivity over our target, and those
2
assays are usually run at one micromolar.
3
examine at least 270 kinases, and we'll alter that
4
as necessary.
And we
And these are by binding panels.
When we can, we will follow up with a hit
5 6
with a cell-based assay, but as you heard earlier,
7
sometimes we just don't know about the biology of
8
these targets, and sometimes we have to investigate
9
those.
And when we do investigate those, we'll put
10
it out in the public.
11
the LRRK kinase inhibitors.
12
inhibitors looking at species differences.
13
We did that recently with We've done it with MEK
With regard to the off-target evaluations,
14
so let's call it secondary pharmacology, we look
15
across a variety of receptors, enzymes, and ion
16
channels, and we set a criteria.
17
the size of the panel, a panel may be 40 targets, a
18
panel may be 150 targets, we like to keep the
19
number of these hits in these binding assays at
20
less than 10 percent.
21 22
So depending on
So if we have a highly promiscuous molecule, we ask the chemist to go back and look for another
A Matter of Record (301) 890-4188
83
1
scaffold that may not be as promiscuous.
2
all the reasons you've heard, we do it not only for
3
kinases, but we do it for these other off-targets
4
because, frankly, we don't understand all the
5
secondary pharmacology targets and how they impact
6
some of the outcomes.
7
So for
So again, we do this at 10 micromolar.
And
8
a positive hit -- this is pretty standard -- is a
9
15 percent displacement or binding.
But we don't
10
end there.
What we do is when we see the hits, we
11
actually interrogate each of those hits to
12
understand what is the potential for agonism and
13
antagonism.
14
animal-based assays.
15
results, those IC50s, those Kis, to try to
16
understand what the safety margin may be, or the
17
translatability.
We have to do that in cell-based or And then we use those
18
For intrinsic cytotoxicity, we use primary
19
human hepatocytes because there are some enzymatic
20
capabilities.
21
show you a real high snapshot.
22
EFis and hemodynamic, as I've told you before, and
And for cardiovascular risk, I'll
A Matter of Record (301) 890-4188
We look at both
84
1
I alluded to what we do with our DMPK colleagues. So here's a high level table of some of the
2 3
off-target evaluations that we do.
Again, just
4
splitting it up by receptors, ion channels or other
5
targets, the other targets would be customized.
6
And then I have the columns, whether it's
7
life-threatening or non-life-threatening, and in
8
the yellow is what we do for a specific therapeutic
9
area.
So if we know we're going to target the
10
brain, we expand our panel pretty widely from our
11
base panel.
12
Our base panel is a panel of 40 targets
13
only, and the purpose of that panel is twofold.
14
One, it targets some of the most important targets
15
that are associated with safety pharmacology issues
16
in the clinic.
17
we believe this 40 panel is a good indicator of
18
promiscuity.
19
And two, based on our assessment,
So that's how we use our initial screening
20
panel, and then we add to it as needed.
21
once we get a hit, we interrogate that hit to
22
really understand the true clinical relevance and
A Matter of Record (301) 890-4188
And again,
85
1
do follow-up work to understand whether there's a
2
translatable effect.
3
Here's an example of how we would do a
4
tiered approach to build a weight of evidence for
5
some of our off-targets, and again this is not
6
unique to Genentech.
7
with a computational or cell-based assay, if they
8
have it, and then they do some kind of higher
9
throughput screening.
10
Most companies will start
We find the two-point patch clamp for EFis
11
works very well.
12
And then we will often do the IC20, IC50 patch just
13
for confirmation.
14
of evidence.
15
It's very predictive of IC50s.
Again, we're building a weight
We do routinely go into IPS systems, and the
16
reason we do that is because the technology has
17
advanced to a degree where we can least do effects
18
with impedance on beat rate.
19
build out an MEA type platform.
20
to try to qualify our assays so we understand what
21
the value is.
22
We're starting to And we're starting
We don't fully know how to use these assays,
A Matter of Record (301) 890-4188
86
1
and so it takes time for us to build confidence and
2
a weight of evidence of how we can use these and
3
what the confidence is in these.
4
We still utilize ex vivo models, whether
5
it's hanging heart or aortic.
These are functional
6
models where we can glean a lot of information.
7
And finally, we always tether it into some
8
integrated assessment so that we can get a PK/PD
9
effect relationship.
So this is a typical example
10
of how we use the data, how we build the data out
11
to make decisions.
12
Now, I'm going to give you the conclusions.
13
I'm going to start with telling you what we've
14
found, and then we're going to go through, phase by
15
phase, and I'll drill down to some of the examples.
16
So based on our kinase portfolio at
17
Genentech, when we advance a molecule into the
18
clinic, using this lead optimization strategy, what
19
we've determined is that we are advancing highly
20
selective molecules, and that in our hands we
21
believe that 90 percent of the toxicities that
22
we're seeing in the clinic are on-target.
A Matter of Record (301) 890-4188
We don't
87
1 2
see many off-target toxicities. As Richard alluded to, we've learned a lot
3
by advancing oncology molecules, and this is
4
helping us to understand how to better predict the
5
outcome of kinases.
6
think we all have learned, is that some of these
7
toxicities are just not going to work in non-
8
life-threatening conditions.
9
But one of the things that I
Now, let's talk about -- remember our goal.
10
The goal was that we wanted to attrit as much as
11
possible before we got into phase 1, and I believe
12
that we've done that.
13
success or failure, we consider it a success.
14
Whether you see that as
But in getting to Tom's point, we're
15
attriting some molecules based on assumptions that
16
we're making, and we simply have to do that.
17
of them are held based on clinical outcomes, but
18
some of them are just based on decisions not to
19
move forward, and I'll show you those.
20
Some
What we've determined in our portfolio,
21
again largely oncology, is that efficacy still
22
predominates as the number one cause of attrition
A Matter of Record (301) 890-4188
88
1
in the discovery and phase zero periods.
When we
2
do have toxicity-related attrition, 60 percent of
3
the time, it's due to off targets. Again, I've alluded to the non-
4 5
life-threatening condition.
6
look across our entire portfolio out to phase 3,
7
and now we're starting to have some marketed
8
products, toxicity only accounts for about 40
9
percent of the attrition.
10
And overall, when we
It still is
predominately efficacy or business decisions. Going back to our model, we believe right
11 12
now, the model that I've just showed you is one
13
that seems to be working well in our hands, and
14
we'll continue to use that model.
15
time, we'll continue to evaluate what happens in
16
the clinic and bring it back to the preclinical
17
setting to determine how we can better improve
18
those models.
But at the same
So let's go into the details.
19 20
data.
21
showing you today.
22
attrition.
Here's my
This table represents the data set that I'm On the left is the phase of
Parenthetically are the number of
A Matter of Record (301) 890-4188
89
1
either molecules or programs that were attrited.
2
In the discovery phase, we kill programs, not
3
really molecules.
4
phase zero, what I'm referring to are molecules.
But once it gets into
You can see that I've split them out from
5 6
oncology versus non-oncology targets.
7
the final, the last column, is attrition due to
8
toxicity.
9
see by the graphic that it's happening, as I told
10
And then in
Again, it's about 40 percent.
You can
you, in the discovery and phase zero. In discovery, again, in the oncology phase,
11 12
it's efficacy over tox, but in the non-oncology,
13
which is not unusual, it's tox.
14
one determinant of attrition.
15
when we decide, over time, if we can't advance a
16
molecule, and we decide to stop something, it's
17
100 percent termination of the project.
18
stop.
19
druggable.
20
one of our decision points for some of these data.
21 22
That's the number And in our hands,
So we
We determine that the project is not So this is important because this is
In phase zero, attrition is 100 percent toxicity related.
And again, that may make sense
A Matter of Record (301) 890-4188
90
1
because usually we have all the other features
2
built out, and what we're seeing is unexpected
3
toxicity, or what we believe is a non-advanceable
4
molecule.
5
termination, again makes sense because we go back
6
to find backups.
7
However it's resulted in zero project
In phase 1, we have never lost a project to
8
toxicity, but we have terminated for PK efficacy
9
and business reasons, about 25 percent of our
10
portfolio.
11
first -- or end of last year, our first attrition
12
of one of our PI3-kinase inhibitors in phase 3 due
13
to DLT, what we're talking about today, inability
14
to manage efficacy with regard to rash.
15
And recently this year, we've had our
Now, what I'm going to do is I'm going to
16
take you through each phase.
So this is a
17
discovery phase, and what I want to demonstrate to
18
you is the target organ toxicities that we
19
identified, whether we felt they were target
20
related or not.
21
investigational work comes in.
22
in some of this so we can make these decisions, and
And this is where the
A Matter of Record (301) 890-4188
We have to invest
91
1 2
then other considerations. You can see that there's probably just about
3
an equal split of oncology versus non-oncology.
4
the green are two examples, I'll just show you in
5
the next slide, where the decision to stop
6
advancing these molecules was equally tox and
7
efficacy or ability to appropriately target
8
the -- efficacy against the target.
9
In
The take-home from here is that when we do
10
an assessment of the reason for toxicity in this
11
discovery phase, still 50 percent of the toxicities
12
are due to off-target in our estimation based on
13
our evaluation.
14
The other important takeaway is when
15
we -- and this is a dog in the industry, is that we
16
have no good biomarkers still for lung or vascular
17
toxicity.
18
decision-making of why we didn't advance certain
19
molecules or why we terminated certain projects.
20
And so this is part of our
Let me get into some of the actual
21
decisions.
So basically, as I told you, there's a
22
point at which all of this accumulated data, all of
A Matter of Record (301) 890-4188
92
1
our attempts to advance a molecule, are stopped,
2
and we actually will stop the entire program.
3
again may not be unique to Genentech.
This
What I've done here is I've grouped some
4 5
examples.
And the first example, we knew it was
6
off-target toxicity across the scaffolds.
7
it.
8
around it chemically.
We knew
But we didn't have a clear idea of how to get
In addition, we did have formulation issues,
9 10
which were going to severely impact our ability to
11
dose.
12
landscape.
13
that went into us terminating this particular
14
target.
15
And finally, we had a very competitive And so those were the three decisions
For two other molecules, again, we knew they
16
were on-target toxicities, but we actually either
17
had no safety margin, or equally we had no
18
biomarkers.
19
no biomarker to be able to detect it clinically.
20
And so a decision was made to stop those projects,
21
even though they were druggable targets.
22
So if we had a safety margin, we had
The third example, we knew it was an
A Matter of Record (301) 890-4188
93
1
off-target, and again this idea of limited chemical
2
space.
3
get around the IP.
4
landscape, so we stopped the program.
This was an IP issue.
We simply couldn't
And it was a very competitive
Now, in the other two that are below the
5 6
dotted line, you'll see, as I alluded to before, in
7
one case, we identified the target toxicity.
8
was hematopoietic.
9
But there was a real question, based on some
It
We felt we could move forward.
10
clinical information coming in for related targets,
11
of the value of this target.
12
combination of those two things, they terminated
13
the project.
So because of the
Basically, in the last situation, we really
14 15
couldn't differentiate ourselves from our target
16
and the standard of care, which was not in a kinase
17
class, but which was the standard of care, so we
18
terminated the project. Now, let's move into phase zero.
19
Again,
20
what I told you is that we have seen attrition
21
here.
22
you see that, gee, we have the same target
It's 100 percent toxicity related.
A Matter of Record (301) 890-4188
And here
94
1
toxicities, and that makes sense because these were
2
some of the lead molecules that attrited, and we
3
went back to try to identify backups.
4
been in the same chemical space or related chemical
5
space, so we've tended to have similar toxicities
6
in terms of target organs.
We may have
Again, we went back and looked, as I've done
7 8
before, at whether it was target related or not or
9
what were other considerations.
And here we're
10
still struggling with off-target toxicity and
11
trying to get very selective molecules to move
12
forward.
13
and now we have a pancreatic biomarker issue, that
14
I know others of you have had this problem too.
15
So let's look at decision-making around
In addition, we have a biomarker issue,
16
that.
Yes, I'll get into why things were
17
terminated.
18
project was terminated.
19
terminated a molecule, we were successful in going
20
back and in 80 percent of the cases identifying a
21
backup that we could move forward, that was
22
druggable, and we could move forward.
But one important thing is that no So although we may have
A Matter of Record (301) 890-4188
95
1
Here I've grouped some of the factors
2
related to our decision-making.
In the first case,
3
the target organ that we identified in the GLP
4
studies was not part of our discovery target organ
5
group, so we use an abbreviated group.
6
focus on the most important activities.
7
unexpected.
8
GLP studies, because we actually sampled it, we had
9
to go back to discovery, but we have been
We try to It was
And so when we identified it in our
10
successful in moving forward a molecule that's in
11
the clinic now.
12
At Genentech, we try to move fast, and so
13
there are decisions made to advance molecules into
14
phase zero on rodent-only data, so we don't always
15
generate the nonclinical species data.
16
three instances, when we did, we identified a
17
liability that was a significant liability.
18
And in
In one case, we were able to identify a
19
backup, but it was terminated for non-toxicology
20
reasons.
21
target organ toxicity, and again this was one where
22
there was a competitive landscape.
In another instance, we identified a new
A Matter of Record (301) 890-4188
And then
96
1
finally, in a third instance, we were actually able
2
to use a rodent screen to de-risk a particular
3
toxicity and move the molecules forward.
4
In the final instance, we identified the
5
toxicity.
6
identified it.
7
was a clinical decision not to move that forward.
8
And so we went back and identified a backup
9
molecule that de-risked those issues.
10
It was a cardiovascular risk.
We
We had a monitoring plan, but it
So when we summarized the discovery and
11
phase zero experience that we have had, again what
12
we've determined is that we are attriting a large
13
number of molecules due to off-targets, based on
14
our investigational work.
15
they're the big players, cardiovascular, CV, and
16
hematopoietic.
17
These will always be important reasons for
18
attrition.
19
with some idea of what the character is.
20
And you'll see that
And basically, these won't go away.
But at least we'll try to advance them
In my final slide, looking at our portfolio,
21
I've just snapshot nine molecules that were in the
22
clinic from different molecular classes.
A Matter of Record (301) 890-4188
And what
97
1
I want you to focus on, in the second column is
2
in vivo, and then we have an off-target in the
3
third column.
4
The first point is that when we advance
5
molecules, we advance them with little off-target
6
toxicity.
7
don't get everything right, but we are -- as been
8
alluded to before, we are able to predict a lot of
9
the most significant toxicities in the clinic.
10
And then when you look in the red, we
Here's our overall, just the data that we
11
put together here.
12
molecules that I based this on, and we're having
13
that 79 percent attrition rate before it gets into
14
human beings.
15
I'll show you, we had 37
Now, I'll just kind of quickly, because I
16
only have a few minutes left, just go over a little
17
bit about one of the other things that dogs us, and
18
that is, pun intended, what nonclinical species to
19
use, non-rodent species to use.
20
What I have here is some rationale behind
21
how we choose the non-rodent species.
22
we want to make sure we have human metabolite
A Matter of Record (301) 890-4188
One is that
98
1
coverage in either the rodent or non-rodent.
Two,
2
obviously we want to have the right amount of
3
exposure, especially with regard to the non-
4
life-threatening conditions.
5
for us to have some kind of pharmacological
6
activity.
7
Sometimes we simply don't have the reagents to look
8
at the PD markers that we want, especially in the
9
dog.
It's really important
But we also have to consider reagents.
And this is significant for us. Then finally, there are other considerations
10 11
that we look at that, again, I'm sure we're not
12
alone at.
13
there are a lot of considerations that we take into
14
consideration to decide whether we use a monkey or
15
a dog.
16
But I just want to give an image that
Here's our experience based on the number
17
that I've given you today.
I looked at 23 kinases
18
using SBAs over a period of time, and then used our
19
experience of looking at over 48 different
20
molecules.
21
do look at the dog, but we're inching up on the
22
cynomolgus monkey for the reasons that I showed you
And you can see that predominately we
A Matter of Record (301) 890-4188
99
1
on the prior slide.
So we do have a lot of
2
experience across these species. What we've experienced is that in looking at
3 4
over 48 kinases, we've looked at over 18 target
5
classes, mostly in the dog versus cyno.
6
three instances where we've looked at both.
We've had
When we have done that in the small data set
7 8
that we have, we actually saw comparable serious
9
toxicities in the dog and cyno, or we saw other
10
toxicities in the cyno that we didn't see in the
11
dog.
12
lung toxicity.
13
three programs, together we saw that they were
14
comparable, but again it may be other toxicities
15
were identified as well.
16
These included predominately vascular and And again, when we looked at the
Now, one of the specific questions we have
17
within Genentech that our chemists were pushing
18
back on us is the GI toxicity, that dogs were more
19
sensitive.
20
dogs being highly sensitive to GI toxicity.
21 22
So there was this urban legend around
So what we did is we evaluated our small data set, and we only observed GI toxicity where it
A Matter of Record (301) 890-4188
100
1
was unique to the dog in four of -- where it was in
2
the dog in 4 of 11 targets.
3
those actually did translate into human beings.
4
Of those four kinases,
So based on the available data that we had,
5
we did not see any specific sensitivity of a dog
6
versus the human in terms of prediction.
7
dogs are more sensitive, so we also looked at
8
exposure.
9
Sometimes
Now this is one example of exposure.
An
10
important part here is that I can give you one
11
example after another, but what we can't really do
12
well is give you an epidemiological snapshot.
13
this may work for this molecule, but it may not
14
work for another molecule.
15
So
With regard to our PI3-kinase molecules,
16
with regard to GI and lung inflammation in the dog,
17
we not only got fairly good translation, but fairly
18
good translation around risk in terms of exposure.
19
Again, this does not happen all the time, but these
20
are the kinds of assessments that we try to go back
21
and do to try and understand the translatability of
22
our data.
A Matter of Record (301) 890-4188
101
1
I went forward, for the GI to help convince
2
the chemists, I went through the SBAs just for GI
3
toxicity, and the yellow represent non-translatable
4
effects, and the blue translatable across both
5
species.
6
the monkey are pretty equally good at predicting,
7
and I think we'd seen that earlier today in
8
Richard's talk as well, that they actually are
9
fairly reasonable at translating.
10
What I came out with is that the dog and
So when I take us back to our findings
11
again, that in our hands, when we try to go for
12
selectivity, when we try to reduce off-targets, we
13
actually appear to be having good translation.
14
We are going to get toxicity in the clinic,
15
but it appears to be on-target.
16
home here, what I want to leave you with, is how we
17
use the data and what our challenges are.
18
And the final take
So as of today, when we use this integrated
19
data, we use it to make decisions around
20
druggability of a target.
21
entire project using some of these information.
22
use it to identify our lead candidates and our
So we may terminate an
A Matter of Record (301) 890-4188
We
102
1
backups.
And we predominately use it for hazard
2
identification and characterization to enable the
3
starting dose monitoring exclusion criteria.
4
We do the same thing that Tom does.
We'll
5
work with our efficacy, our research colleagues to
6
understand dosing schedule, and we try to translate
7
those things.
8
things we still struggle with is it's hard for us,
9
on a molecule by molecule basis, to predict which
But at the end of the day, the two
10
species is the most sensitive.
And based on our
11
limited TK sampling and the way we design our
12
preclinical studies, it's hard for us to do really
13
good PK/PD modeling.
14
and do additional studies to try to get those,
15
those endpoints.
16
Thank you.
So we have to often go back
And so with that, I will close.
17
(Applause.)
18
DR. JANNE:
The next speaker is Dr. Kluwe
19
from Novartis, and then we'll have the panel
20
discussion.
21 22
Presentation – William Kluwe DR. KLUWE:
Thank you.
A Matter of Record (301) 890-4188
I too would like to
103
1
thank the organizers of the conference for inviting
2
me here.
3
to be able to speak to this group.
It's quite a pleasure and quite an honor
One of the advantages of having a workshop
4 5
is that we not just get to identify what great
6
things we may have done, but we are also able to
7
share some of our experiences, successes and
8
failures, but always keeping in mind the final
9
endpoint here, which is to try to facilitate better
10
identification and management of the drugs that
11
we're introducing into clinical use.
12
So with that, let me begin by starting to
13
build on some of what you heard from the previous
14
speakers here. First of all, we've talked a lot about the
15 16
human kinome, which is nothing more than a
17
collection of enzymes that phosphorylate other
18
enzymes that do wonderful things.
19
ask yourselves, well why all the focus on the human
20
kinome.
21 22
And you might
The answer is because it has been identified as being critical in so many diseases and
A Matter of Record (301) 890-4188
104
1
pathologies that we've recognized in recent years.
2
So it's a rich opportunity for us to be able to use
3
in the process of discovering and developing new
4
drugs.
5
Now, all that being said, you've heard a lot
6
of a word today that you may not commonly hear in
7
your normal conversations, that is "promiscuity."
8
We've dealt with this for a number of years, and I
9
think in the past, we perhaps had a somewhat naive
10
and optimistic notion that we could look at the
11
human kinome, and we could map it out, and we could
12
identify, "toxic kinases and good kinases," and
13
that we could identify those that should be avoided
14
at all costs when we're trying to look for good
15
inhibitors and those that we can focus on.
16
I will throw a little bit of perhaps
17
pessimism into that area, and that is to say that
18
our knowledge here is still incomplete.
19
methodology that we use to look at potencies is not
20
perfect.
21
still a continuing picture.
22
have clinical confirmation of whether or not these
The
It doesn't always translate well.
It's
In many cases we don't
A Matter of Record (301) 890-4188
105
1
are really things to be avoided or not, so just a
2
little bit of caution that we put in there.
3
The other thing to remind ourselves is that
4
if there's been a disappointment with kinase
5
inhibitors in oncology, it's that even while we
6
have found very, very good drugs that quickly
7
suppress tumor growth, oftentimes that
8
growth -- that effect is not a durable one.
9
when we look into asking ourselves why it isn't
10
durable, oftentimes we find out that there's up
11
regulation of another kinase pathway.
12
look back we say, but if I had perhaps a broader
13
range and less selectivity, maybe that wouldn't
14
have happened, and I think we have some indications
15
where we've actually taken that into advantage.
16
And
And when we
So this might be a situation where, dare I
17
say it in public here, but a lack of promiscuity
18
may not always be a bad thing.
19
Why is it then that we are so focused on
20
these as a mechanism of advancement in oncology
21
treatment?
22
there's a lot of things that go on in the tumor
When we look at this, we find out that
A Matter of Record (301) 890-4188
106
1
that are dependent on kinases, and we feel this is
2
a good opportunity for us, if we really want to
3
suppress that tumor growth, this is the area that
4
we need to work in. But as we look at these attributes of the
5 6
tumor that make it so susceptible to kinase
7
inhibition, we also can't lose sight of the fact
8
that we didn't develop a kinase, a human kinome,
9
for the express purpose of growing tumors in the
10
body.
It's also an essential pathway for a number
11
of different functions, physiological functions
12
that go on.
13
when we think about toxicity, we need to understand
14
what these kinases are doing in the normal tissue.
15
Nonetheless, we think there is a mechanism
So we need to always be concerned that
16
here for effective drug treatment because the way a
17
kinome works in the tumor is oftentimes quite
18
different than the way it works in the normal
19
tissue.
20
that you can see why we feel that there are
21
opportunities to exploit kinase inhibition in
22
cancer drug development without running into
And there are a number of reasons here
A Matter of Record (301) 890-4188
107
1
excessive toxicity. We've heard quite a bit about preclinical or
2 3
nonclinical safety testing, how it correlates and
4
what the pathway is for decision-making, a lot of
5
these things.
6
speakers have already aptly covered here.
7
to talk to you instead about the fact of
8
integration and how we look at this data as we move
9
forward.
I don't want to dwell on what other I want
This is a fairly common sequence of events
10 11
that one would see in early drug development and
12
discovery here.
13
all, establish what the tumor mechanism of action
14
is and what the biological relevance of this is for
15
the tumor.
16
It's a very efficient and effective way of looking
17
at it.
18
And that is, we try to, first of
We do this frequently with cell lines.
We then go on and we start looking at these
19
in in vivo situations here.
We try to focus on the
20
dose-response relationship is, what it's
21
relationship is to systemic exposure, metabolism of
22
the compound, treatment durations, types of
A Matter of Record (301) 890-4188
108
1
scheduling.
2
where we have perhaps under-recognized the
3
importance of trying to manage the safety versus
4
the benefit and using schedules as one mechanism to
5
do this.
6
And I think that's one of the areas
Then also, you've heard a lot today about
7
off-target and on-target toxicity, and these are
8
perhaps the less obvious things that we look at,
9
and saying what might be happening that's related
10
directly to the pharmacology of the drug, what
11
might be happening that's not necessarily related
12
to the pharmacology.
13
of the molecule.
14
design the optimum candidate for taking into
15
clinical development?
16
It may be another attribute
Can we understand these and
Then importantly, what we have to do is we
17
have to take all of this information, put it
18
together, make sense out of it in what we call a
19
risk evaluation.
20
thinking not just about what those risks are, but
21
how might we mitigate those risks such that when we
22
move into the clinical setting, we aren't putting
And very importantly, also be
A Matter of Record (301) 890-4188
109
1
patients at undue risk because we didn't think
2
proactively about how best to use that drug.
3
So what you can see here is that we advance
4
on through a series of what we call traditional
5
toxicology studies as well as closely looking at
6
all the pharmacology information and all the
7
in vitro information to try to best understand what
8
that drug candidate does.
9
You can take this same sequence of events
10
and look at it in a somewhat different fashion and
11
say, okay, how do I use that data?
12
hear that I collect all that data, and that's what
13
we spent most of our efforts on, but how do we
14
actually use it as we move a drug forward into
15
clinical development?
16
That's nice to
What we've done here is we've taken these
17
out and said, well, part of that has to do with
18
target selection.
19
decision, is that a target I wish to exploit?
20
if the answer is yes, then what are my criteria for
21
picking the right chemical matter to actually
22
exploit that target?
I do have to make a conscious
A Matter of Record (301) 890-4188
And
110
1
That's where a lot of this information comes
2
in now because we're making value judgments.
3
talked about selectivity, high selectivity, low
4
selectivity, what are the types of off-target
5
effects, what type of pharmacokinetic parameters do
6
we want to design into this drug, so that becomes
7
part of the candidate selection process.
8 9
We
But then, very importantly, it's also the clinical protocol preparation.
Are we using that
10
information to full advantage to come up with a
11
clinical protocol that not only allows us to
12
quickly identify whether the candidate is actually
13
going to be an effective treatment, whether it's
14
going to be an improvement, and how do we minimize
15
the number of patients needed to do this while also
16
trying to maintain them in as safe an environment
17
as possible.
18
Here, I'm going to talk a little bit about,
19
now, some internal data here.
20
advancement over the past several years, and that
21
was with the introduction of the ISH-S9 guidance
22
that simply recognized the fact that cancer is
A Matter of Record (301) 890-4188
We've had a great
111
1
different.
2
not normal volunteers.
3
often failed on other therapies.
4
reticence to be under-dosing them because they do
5
have to derive some benefit, or some hope of
6
benefit, from the early clinical trials.
7
We're beginning our work in patients, These are subjects who have There is a
So we went to this concept of saying, how
8
can we start at what might already be close to a
9
clinically effective dose without putting these
10 11
people at undue risk? Now, what you can see here -- what we've
12
done here is we've just shown the common mechanisms
13
that we use for actually picking a starting
14
clinical dose.
15
other than to say it's a much different mechanism
16
than we would use for a drug in a non-
17
life-threatening situation here.
18
I'm not going to dwell on this
What I've taken here is recent experience
19
from Novartis where we have 14 compounds that have
20
progressed far enough into the clinic where we get
21
some sense as to whether or not ICH-S9 guidance is
22
actually successful.
A Matter of Record (301) 890-4188
112
What I've done here is we've shown the
1 2
number of escalations, starting dose, first cohort,
3
second cohort, which would be the first escalation,
4
and so on and so forth.
5
that we use how to calculate that human dose, let's
6
assume that we're using the non-rodent species
7
here, that initial cohort would be used at
8
one-sixth of what we call a highest non-severely
9
toxic dose, which would be analogous to a maximum
10
If you take the paradigm
tolerated dose in the clinical population. If you go through the first escalation, that
11 12
brings you up to one-third, second escalation,
13
two-thirds, third escalation, you probably are at a
14
risk of exceeding that. Now, that was the theoretical, here is our
15 16
actual experience here.
The number of escalations
17
needed to achieve -- so I'm not going to talk about
18
maximum tolerated dose here because that can be a
19
concept that's difficult to define, oftentimes
20
recognized two years after the fact.
21
instead focus on getting it into a therapeutic
22
range.
A Matter of Record (301) 890-4188
So let's
113
1
What's perhaps most inspiring about that is
2
that the median there is 2, which means, on the
3
first hand, that we are using our patients very
4
effectively in these studies.
5
up another issue, and that is to say that if you go
6
through 2 escalations, you're pushing up against
7
the maximum tolerated dose.
8 9
But it also brings
So we should not be surprised that we see intolerance in the patients.
We should not be
10
surprised that we see toxicity.
11
paradigm that we've chosen to use, and that is to
12
say we should be starting out at close to what is
13
an effective dose.
14
escalate or dose de-escalate shouldn't be viewed as
15
a failure, but rather as a mechanism for us to make
16
efficient use of these patients who volunteer or
17
are put into our clinical studies.
18
This is the
And whether we have to dose
The other thing I want to address here very
19
quickly, and you've heard some of this before, we
20
oftentimes are somewhat negative about our
21
predictability.
22
confuse predictability with susceptibility or
And sometimes that's because we
A Matter of Record (301) 890-4188
114
1
sensitivity.
2
phenomena.
They are two very different
But what we've done here is we've actually
3 4
looked over the kinases that have been approved
5
over the past several years.
6
you, there was an upsweep of those in the last five
7
or six years, so we have quite a bit of experience
8
now.
9
clinical data, let's look first at the common
As Richard showed
And comparing the preclinical data with the
10
adverse events that we oftentimes see emerging
11
fairly quickly.
12
If we look at those, and on the clinical
13
side, this is a broad representation of kinase
14
inhibitors, what we see is a list that quite
15
frankly matches very nicely with the same list that
16
we see from the preclinical studies.
17
see a little arrow there, that denotes that these
18
were common events in both series of studies, and
19
therefore we say it is a good, pretty good,
20
predictability.
21 22
And where you
There are certain things that we see in clinical studies that aren't readily detected in
A Matter of Record (301) 890-4188
115
1
our preclinical studies, and those types of things
2
we recognize.
3
indicates the things that we see uncommonly, but
4
oftentimes are associated with a very specific
5
mechanism, and so as we see dermatotoxicity with
6
the EGFR inhibitors here.
7
Where I've drawn a line here, this
Those are oftentimes seen -- you may have to
8
do a specific type of study, and if you don't look
9
for them specifically you may not be aware of them.
10
But nonetheless, the animal models are actually
11
quite good for either looking directly at what the
12
risk is in a very proactive fashion, or for coming
13
back later and studying perhaps what's going on,
14
could it be affected by a schedule, how is it
15
related to dose, how is it related to
16
pharmacokinetics, and things of that nature.
17
There are some infrequent responses that we
18
see, and has been pointed out here in Tom's talk
19
and some of the other talks today, infrequent
20
occurrences, by their very nature, are going to be
21
very, very difficult to predict in our preclinical
22
studies.
The group sizes are small.
A Matter of Record (301) 890-4188
The animals
116
1
are quite standardized. We're looking for trends.
2
We're not looking
3
for the irregularities.
And perhaps that is why we
4
are sometimes surprised when we see these uncommon
5
effects occurring in the clinic. I've listed some of these here for the
6 7
kinase inhibitors, just speaking generically.
And
8
the one that I think we want to spend a bit more
9
time on -- as we've discussed over the opening
10
session and with some of the subsequent ones and
11
I'm sure we'll hear more this afternoon as well as
12
tomorrow -- the cardiovascular toxicities; things
13
that occur infrequently, thromboembolism, edema.
14
I'm going to spend a little bit of time here on
15
heart failure, or a decrement in left ventricular
16
ejection fraction, and you'll see why in a moment I
17
hope.
18
Let's look at that one about cardiomyopathy
19
and understand, well, what are the comorbidities
20
for cardiomyopathy.
21
any patient population, but they've actually been
22
documented for cardiomyopathy induced by kinase
These would be true in almost
A Matter of Record (301) 890-4188
117
1
inhibitors:
2
anthracycline treatment, and hypertension.
3
their age, experience with
Now just to give you an example of why this
4
is important, a recent publication came out and
5
indicated that the median age at diagnosis of CML
6
is in the early 60's.
7
speaking of is going into a clinical population
8
that we can estimate is going to have these
9
comorbidity risks, and we should not be surprised,
So already what we're
10
then, that if we are looking in normal healthy
11
animals, we didn't necessarily detect that as a
12
risk factor.
13
Now, speaking just as an example with
14
hypertension here, one of the things we'd like to
15
promote is the concept here that you don't
16
necessarily have to predict everything.
17
to manage these things.
18
you need to be aware of what's coming up in the
19
clinical studies.
20
those patient populations might be particularly
21
susceptible, and then think about how you would
22
manage that or how you would address that issue.
You'd like
So when you think of it,
You need to understand what
A Matter of Record (301) 890-4188
118
So just a simple indication here.
1
I'm going
2
to use a MEK inhibitor.
It's going to remain
3
unnamed.
4
trying to be particularly sensitive and not malign
5
any particular compound, so I'm just going to call
6
it MEK001.
It is a real MEK inhibitor, but we are
Common questions that you might ask if
7 8
you're dealing with this, when you know that the
9
drug is associated with heart failure in the
10
clinical population, is ask yourself, well what is
11
the potential for reversibility?
12
oncologist and I'm managing a patient, these are
13
some of the key questions that I would like to
14
know.
15
If I'm a clinical
What is its relationship to hypertension?
16
This is the patient population which is likely
17
prone to hypertension, which may or may not be
18
being effectively managed.
19
Then of course the important one is, suppose
20
the drug works and suppose I then get cardiac
21
failure, and I take the patient off?
22
kinase inhibitors, this is very different than what
A Matter of Record (301) 890-4188
We know with
119
1
we see with anthracyclines, in that the
2
cardiomyopathy -- or I should say the decrease in
3
left ventricular function is in fact reversible.
4
But the drug was working against the tumor.
5
reintroduce the drug to them?
6
highly susceptible?
7
Can I
Are they going to be
So these are good logical questions that can
8
be addressed.
And no, they don't need particularly
9
advanced tools or anything else.
They are highly
10
susceptible to preclinical evaluation, and we could
11
make good use of that.
12
Here was a study that was simply done with
13
relatively small numbers of animals, looking at
14
both normal rats and spontaneous hypertensive rats,
15
short durations of treatment, looking at the effect
16
of rechallenge, so on and so forth.
17
to dwell on the experimental design as much, but in
18
a brief period of time here show you what goes on.
19
I don't want
We can look at these in normal rats, and
20
maybe this tells us why we don't effectively study
21
this in normal rats.
22
comorbidity factor of concern.
They don't incorporate the
A Matter of Record (301) 890-4188
120
1
So here's a study that was done in
2
normotensive Wistar rats with introduction of a MEK
3
inhibitor, one milligram per kilogram daily.
4
you do get a decrement by echocardiography in left
5
ventricular function, but, boy, it's difficult to
6
see, especially with small numbers.
7
animal responds in the same fashion.
8
do start to respond, some of them have such a
9
robust response that you put them into a
10
life-threatening situation.
11
model to work with.
Yes,
Not every And when they
So it's a difficult
12
If we go on then to a more appropriate
13
model, the spontaneously hypertensive rat, the
14
first thing you're going to notice if you look
15
closely at the preceding slide, is these animals
16
already have a reduction in left ventricular
17
ejection fraction.
18
hypertensive, but there's a relationship between
19
hypertension in this model and ventricular
20
function.
21 22
So they're not just
A much different picture now.
It's much
easier to see, and a much more uniform response,
A Matter of Record (301) 890-4188
121
1
that there is a reduction caused by the drug in a
2
relatively short period of time that becomes
3
stabilized that's quite indicative of what you
4
actually see in the clinic.
5
Now remember, your question was to ask,
6
okay, but can I recover from that?
It's
7
answerable.
8
recognized by echocardiography.
9
put on a fairly short reversal period.
In this case, the decrement was The animals were Obviously,
10
you don't let this thing go on for too long.
11
in that sequence, yes, they returned back to a
12
normal stage.
13
But
The next question was, of course, are they
14
going to be more susceptible then if you
15
reintroduce the drug?
16
reintroduced first at a fraction of the dose, and
17
it was no worse than what was seen initially.
18
then it was increased up to the initial level, and
19
again, we see no difference there.
20
And here the drug was
And
So you can answer a lot of these
21
questions -- and I'm not going to suggest that this
22
is a replacement for a clinical trial.
A Matter of Record (301) 890-4188
But when
122
1
you start going into how might I manage these
2
things, what might I try, it's good to have some
3
preclinical information to find out, is this
4
feasible, is it realistic, is there an opportunity
5
here for making an adjustment there?
6
Now, obviously the next question is, it's
7
nice but you can you somehow prevent it?
If
8
hypertension is a precipitating factor for the
9
decrease in left ventricular function induced by
10
some of these kinase inhibitors, could you in fact
11
prevent that with a better control of hypertension?
12
We looked in this model, and this is the
13
spontaneously hypertensive rat.
14
blood pressure can be managed with an ACE
15
inhibitor, but not with a beta blocker.
16
when we went into the study and we looked, in this
17
case, we actually treated the rats first to get the
18
decrement in left ventricular ejection fraction,
19
and then introduced a dose of lisinopril in this
20
case, an ACE inhibitor, sufficient to normalize
21
their blood pressure.
22
The decrease in
And then,
I'm not showing you all of the data here,
A Matter of Record (301) 890-4188
123
1
but, yes, we did normalize the blood pressure back
2
up to what it would normally be.
3
reduce the decrease in left ventricular function.
And it does
You can also do it the other way, and that
4 5
is put them on an ACE inhibitor first, normalize
6
the blood pressure, and then introduce the MEK
7
inhibitor.
8
left ventricular function.
And again, you prevent that decrease in
Let me emphasize again, this is not clinical
9 10
proof of concept, by no means.
It's a short,
11
preclinical, nonclinical experiment just to give
12
you some information about the pharmacology of the
13
drug, how it's working, how you might better manage
14
the types of toxicities that you're seeing in the
15
clinic.
16
That is as a, if you will, sort of the
17
balance to say it isn't necessarily that we're
18
going to find drugs that have no toxicity.
19
likely to trade off one type for another.
20
is a plea for saying let's use this clinical data
21
that we are deriving in a manner that we can go
22
back, look preclinically, and perhaps come up with
A Matter of Record (301) 890-4188
We are But it
124
1
a better way to effectively minimize the toxicities
2
that we're seeing in our patient population without
3
necessarily having to go to a different drug.
4
So just to summarize then, from this brief
5
presentation, and to build on what we heard from
6
some of the previous speakers, kinase inhibition
7
has been a very useful mechanism for developing new
8
drugs in oncology, and we feel that that is still
9
an opportunity awaiting there further definition.
10
The preclinical safety studies are pretty
11
good for accurately projecting the acute risk that
12
you're likely to see, but they're not necessarily
13
the right things if you're thinking about what some
14
of the comorbidities are.
15
attention to these and predict some of
16
them -- predict some of the comorbidities and
17
perhaps customize our toxicology studies to better
18
address those risks.
19
And we have to pay
We perhaps have under-recognized the risk to
20
the cardiovascular system.
I think that may have
21
been a consequence over the past five or six years
22
of focusing on cardiac arrhythmias, QT
A Matter of Record (301) 890-4188
125
1
prolongation, and the emerging difficulties there.
2
But there are other things like contractility and
3
cardiomyopathies that aren't as easily discerned
4
that perhaps we should give more attention to at
5
the present time. Most importantly, that we should be prepared
6 7
for and not be reticent to run some follow-up,
8
nonclinical/preclinical studies to better
9
understand what the risks are in our patient
10
populations as we're developing a drug and use it
11
to manage the risk more appropriately.
12
that, I'd like to thank you for your attention.
13
(Applause.)
14
DR. JANNE:
And with
We'll have our panel discussion
15
now.
We have two moderators, Dr. Palmby from the
16
FDA and Dr. Dambach from Genentech.
17
Panel Discussion
18
Tom Palmby and Donna Dambach DR. PALMBY:
19
Good morning.
My name is Todd
20
Palmby.
I'm one of the pharmacology toxicology
21
team leads within the oncology office at FDA.
22
Thank you all for attending.
Thank you very much
A Matter of Record (301) 890-4188
126
1
for the excellent talks in this morning's session.
2
I think they set the discussion up very well.
3
Now, we're going to have a panel discussion
4
with four speakers from this morning's session as
5
well as some additional panelists.
6
here is that we want an integrated discussion here,
7
so we have representation from multiple disciplines
8
up here.
9
so everyone's aware what representation we have.
10
And the idea
I'll introduce briefly all the panelists
Ultimately, also we'd like this to be a very
11
integrated discussion, so we want the audience
12
participation here.
13
questions, discussion, topics, anything that you'd
14
like to bring up for discussion for the panel,
15
that's what we would like to have here.
16
So please feel free, any
First, the co-moderator for the panel
17
discussion is going to be Donna Dambach.
18
the senior director and head of toxicology at
19
Genentech.
20
Donna is
To go through the rest of the panelists,
21
Dr. Pasi Janne, who you heard from earlier today.
22
He is the scientific director for the Lowe Center
A Matter of Record (301) 890-4188
127
1
for Thoracic Oncology, as well as a senior
2
physician at Dana Farber Cancer Institute and
3
professor of medicine at Harvard Medical School. We'll also have Dr. Richard Brennan, who was
4 5
one of the speakers from this morning.
6
scientific advisor for preclinical safety from
7
Sanofi.
8 9 10
He's a
Dr. Thomas Jones is a scientific officer for toxicology and pathology at Eli Lilly. Dr. William Kluwe is the oncology
11
therapeutic area head for preclinical safety at
12
Novartis Institutes of Biomedical Research at
13
Novartis Pharmaceuticals.
14 15 16 17 18
Dr. Kourosh Parivar is the head of clinical pharmacology in Pfizer. Jose Pinheiro is a senior director in statistics at Janssen Pharmaceuticals. Dr. Alice Shaw is associate professor at the
19
Department of Medicine at Harvard Medical School,
20
and attending physician in a thoracic cancer
21
program at Massachusetts General Hospital.
22
Is Dr. Rubin here?
Did he make it?
A Matter of Record (301) 890-4188
Oh,
128
1 2
okay. Eric Rubin is vice president and therapeutic
3
area head of oncology clinical development in Merck
4
Research Labs.
5
I think I got everyone, correct?
6
We have five goals, and these were presented
7
earlier in Natalie's talk, for this session that we
8
had when we were coming up with this workshop, and
9
particularly this session.
10
One is to discuss the evaluation of a
11
compound including selectivity and potency, primary
12
and secondary pharmacology and toxicology.
13
Two, to discuss the selection of lead
14
compounds, the correlation of animal and human
15
toxicities and predictivity of toxicology studies
16
in identifying clinically relevant DLTs.
17 18 19
Three, to discuss de-risking strategies for molecularly targeted anti-cancer drugs. Four, to discuss approaches to integrating
20
information gleaned from the nonclinical evaluation
21
into the design of phase 1 clinical trials.
22
And finally, five, to discuss the role of
A Matter of Record (301) 890-4188
129
1
industry nonclinical teams during the development
2
process of anti-cancer drugs, including clinical
3
trial design, attribution of toxicities to study
4
drugs, clinical toxicity management, and clinical
5
dose optimization.
6
So I'd like to open it up now for
7
discussion, with the discussion focused around
8
these goals.
9
thought we could start off with, around this
And I guess one of the questions I
10
broader theme encompassing many of these goals, is
11
how is the approach that you've taken to the
12
development of kinase inhibitors changed over the
13
last few years with some of the experience that
14
we've gained clinically with some of the older
15
drugs as we've gained a lot more clinical
16
experience and maybe mechanistic understanding of
17
where some of the toxicities have been coming from
18
with some of those products?
19
How has your approach changed from very
20
early lead selection all the way through that sort
21
of toxicity management during clinical development?
22
So if anyone wants, in particular wants to start.
A Matter of Record (301) 890-4188
130
1 2
Yes? DR. BRENNAN:
One of the things that's
3
definitely changed probably in the last five or six
4
years, and Donna alluded to this as well, what
5
we're willing to do at Sanofi is a very careful,
6
in-depth evaluation of the target that we're
7
addressing, both in terms of that target's biology,
8
what it means to the indication --
9 10
DR. PALMBY: DR. BRENNAN:
Maybe it's on now. -- what it means in terms of
11
the indication, but also what it means in terms of
12
potential undesirable adverse effects.
13
just at the target itself, but at any closely
14
related protein family members for that target.
15
We look not
Additionally, we will go back after a
16
secondary screening where we've identified a
17
high-potency secondary target for some of our
18
compounds that's maybe not addressable with
19
chemistry, and try and understand better the
20
biology of those additional targets.
21 22
This is actually something that is particularly starting to translate into the
A Matter of Record (301) 890-4188
131
1
agency's thinking about the safety of the compounds
2
as we progress into phase 1 clinical trials. We have a compound right now in a phase 1
3 4
trial actually in Germany for rheumatoid arthritis
5
that's targeted against a kinase inhibitor, where
6
there are three family members for this particular
7
target.
8
for the particular target of interest.
9
also another secondary target.
10
We were unable to get target selectivity And there's
So the agency came back and asked us to do a
11
more in-depth analysis of the potential adverse
12
effects from hitting those targets.
13
obviously a program that was initiated before we
14
had the process of target evaluation fully in
15
place, which is something in the last four or five
16
years and something that's actually been revised in
17
the last two years.
18
Donna does.
19
And this is
And it's very close to what
So thinking about the target, doing the
20
systems biology around that target, any information
21
we have from genetic studies is a critical part of
22
even thinking about whether we want to address that
A Matter of Record (301) 890-4188
132
1
target in the first place.
2
DR. PALMBY:
3
DR. KLUWE:
Yes? I think we've become more
4
aggressive at putting into nonclinical study
5
protocols what I would refer to as clinically
6
relevant toxicity monitoring. So the old paradigm was to treat the animals
7 8
for a reasonable period of time, autopsy them, pull
9
out every tissue, do a histopath analysis, identify
10
all the target organs.
11
it helps identify things, but it really doesn't
12
enable you to develop a clinical protocol where you
13
want to focus functionally on certain types of
14
things.
15
That's all great.
I mean,
So although it's a bit of an extra burden to
16
put into our preclinical study protocols more
17
functional monitoring, whether it's cardiovascular
18
or echocardiography, whether it's more frequent
19
ophthalmological examinations, electroretinography,
20
all those types of things, it has pushed us to be a
21
bit more aggressive and do the types of things,
22
which would be more clinically translatable than
A Matter of Record (301) 890-4188
133
1
perhaps what we've done in the past. DR. JANNE:
2
But what about duration of
3
treatment; has that changed in terms of toxicology
4
evaluation? DR. KLUWE:
5
Well, I don't think duration has
6
necessarily been the issue.
We do have a limited
7
life span in the animal, so a lot of times you're
8
going to have complicating factors coming in.
9
it doesn't appear that it's necessarily that.
And
We have the advantage of pressing the dose a
10 11
little bit further, and that can oftentimes
12
compensate for that lack of duration, if you will,
13
in the clinic. DR. DAMBACH:
14
Yes, I'd like to follow up on
15
that.
16
that once we push the dose to determine the
17
spectrum of effects -- once we get to the longer
18
term studies, we really can't push the dose.
19
we actually end up backing off so much that the
20
doses that we're testing are subclinical and of
21
questionable utility in some times.
22
I mean, our experience around duration is
DR. RUBIN:
And
Could I ask our colleagues how
A Matter of Record (301) 890-4188
134
1
you're using whole genome types of approaches in
2
preclinical models to predict toxicity that might
3
be relevant to the clinic, profiling or proteomics
4
and things like that? DR. BRENNAN:
5
Typically, we don't do that
6
for practical and economic reasons at an early
7
stage.
8
observing things preclinically that maybe weren't
9
expected to go back and look at mechanism and
We may use those approaches when we are
10
understand target relationship, trying to identify
11
what the off target may be if it wasn't already
12
recognized.
13
But the productivity of whole genome
14
modeling is limited in some respects.
15
certain things that can be picked out in those
16
early screens, but many things don't appear from
17
those -- typically the things that we've seen with
18
the kinase inhibitors don't have transcriptional
19
fingerprints, correlates if you like, that we'd
20
pick out in that respect.
21
DR. KLUWE:
22
There are
It works very good in sort of
the reverse fashion when you go in and you find
A Matter of Record (301) 890-4188
135
1
something, a clinical phenomenon, and you go back
2
and you know exactly what you're looking for.
3
you see this strong correlation.
4
Then
It doesn't seem to work so well in the
5
forward prediction fashion.
And I think a lot of
6
us are concerned, for that reason, of saying we may
7
be following falsely, we may be misdirecting
8
discovery programs, we may be avoiding what might
9
be very useful molecules because of this concern
10
that it might act like something else that we saw
11
in the clinic.
12
DR. DAMBACH:
I'd like to build on a
13
question for folks around something that Tom
14
brought up, and Bill you brought up as well.
15
this idea of not just hazard identification, but
16
actually managing safety signals and the different
17
experiences folks have had.
18
brought up working with his pharmacology colleagues
19
to do dosing schedules.
It's
For instance, Tom
20
Can anyone give us more comments or ideas
21
about how they're working with their pharmacology
22
colleagues to better manage?
And Bill, you brought
A Matter of Record (301) 890-4188
136
1
up a really great example as well, but I'd like to
2
hear from other folks if they have experience
3
either from the clinical side or the other side
4
working with their nonclinical folks.
5
DR. JONES:
I'll respond to that, Donna.
6
First of all, I want to thank you for being as
7
transparent as you were around some of the
8
discovery phase activities.
9
fair to say that, at least of the companies
I think it's probably
10
represented here today, there's a lot of
11
similarities in the approaches that we take there.
12
I think one of the things that we've
13
emphasized in terms of manageability -- Bill
14
alluded to in his talk, which is reversibility of
15
effects.
16
some of these safety issues, if you stop dosing,
17
you can recover the animal, and hopefully that
18
translates into being able to protect patients.
So making sure that as you press into
19
So working there I think is very important.
20
There was another point that you brought up in your
21
talk, which is some of the areas of concern around
22
management come from just not having sufficient
A Matter of Record (301) 890-4188
137
1
biomarkers in some of the spaces, in the
2
nonclinical models in particular, that we can rely
3
on to be able to track and follow some of those
4
changes.
5
investigation as well, working with discovery
6
colleagues.
7
So those are areas of active
DR. PARIVAR:
If I may add a bit on that
8
topic, Donna?
From a clinical perspective, we are
9
very, very keen to see more tox data looking into
10
dosing schedule.
11
especially -- I'm probably going a bit above just
12
VEGF inhibitors per se.
13
status of oncology drug treatment would be
14
combination therapies and such, looking at which
15
agent comes first and second, so the sequence of
16
that one, and the dosing schedule.
17
Which one of the -- I mean,
Looking at the current
Elaborate a bit more on that one because
18
when just one option is given to us in the clinic,
19
we are left alone, really.
20
emerges.
21
you get, okay, what are we going to do?
22
of drug holiday you should be introducing in order
Meaning that then tox
We get all these crazy ideas coming where
A Matter of Record (301) 890-4188
What type
138
1 2
to avoid the toxicity? These type of exercises -- and Alice and
3
others can comment on that from a clinical
4
perspective of patient treatment -- becomes very,
5
very difficult in a clinical setting.
6
consuming.
7
it becomes -- without having any guidance from our
8
talks, it will be very difficult for us to come
9
with a scientifically-driven decision in regard to
It is time
Patients are not readily available, and
10
which type of paradigm here would be the best
11
paradigm to do in order to avoid that toxicity.
12
Toxicity per se, if you look at safe -- we
13
can't look at safety just in separation, you need
14
to look at [indiscernible] and efficacy of course,
15
two sides of the coin.
16
selective you are, the lower dose, we give better
17
safety, but what about safety [sic - efficacy]?
18
Of course, the more
So that inter-correlation between safety and
19
efficacy becomes very difficult without guidance
20
from preclinical data to make such a decision in
21
the clinical setting.
22
you want to add on to that.
And I don't know if, Alice,
A Matter of Record (301) 890-4188
139
DR. SHAW:
1
Yes.
I would say clinically, we
2
have to deal with the manageability of these AEs
3
all the time, and we do often use different dosing
4
schedules, for example qday versus BID dosing, or
5
3 weeks on/1 week off type of schedule.
6
of experimenting a little bit with that.
I was kind
Obviously, the most common way we have to
7 8
manage the AEs would be dose holding reduction,
9
which often does the trick.
There are other
10
things.
What we'll do is we'll shift the timing of
11
doses, for example a qday dose will shift to q.h.s.
12
in the evening, and even that will help with
13
manageability.
14
things we can do in the clinic to try and offset
15
and deal with these AEs.
So there are definitely different
DR. KLUWE:
16
It would seem that the clinical
17
setting is not the best one to explore options for
18
reducing -DR. PARIVAR:
19 20 21 22
Clinical toxicology, I call
it. DR. KLUWE:
We recognize that.
But the
other side of the coin -- and I think here is where
A Matter of Record (301) 890-4188
140
1
there's a major gap, is when one thinks on trying
2
to manage the benefit-risk ratio, also the question
3
then comes up, so what if we find a way to avoid or
4
minimize or schedule around a toxicity finding, but
5
we're not clear on what the impact is on efficacy? I think here is where the problem is because
6 7
if we're looking at something as simple as a tumor
8
size on a CAT scan, that's probably not a good
9
prognosis of what's actually happening at the
10
tumor.
11
basic pharmacodynamic biomarkers that would tell us
12
what's really going on in the tumor and whether we
13
can afford dosing vacations, whether we can afford
14
different schedules, what the impact is going to be
15
on dose reduction.
16
And what we're lacking are some of the
So that seems to be the gap that we need to
17
address and then we can work more effectively.
18
can look at paradigms that you might consider in
19
the clinic and look nonclinically, and see if that
20
is an effective way of reducing the safety risks.
21
But the information on the tumor itself, the tumor
22
growth, is going to have to come from the clinic.
A Matter of Record (301) 890-4188
We
141
DR. RUBIN:
1
Can I follow up on that a little
2
bit?
3
my perspective, it's sometimes difficult to get my
4
preclinical colleagues to look at alternate
5
schedules with a kinase inhibitor because it seems
6
like there's an underlying assumption that you want
7
100 percent target engagement continuously. So I guess the question is that maybe that's
8 9
Because I do think that in oncology, at least
just the way it is at Merck, but I'd be curious to
10
how much effort tends to go into alternate
11
schedules, to getting into what you're talking
12
about.
13
DR. DAMBACH:
Well, I can tell you from our
14
perspective, we do partner very closely with our
15
pharmacology colleagues, and we determine IC40s,
16
IC50s, IC90s; we try to determine things that are
17
specific for a particular target using different
18
models, and then try to relate those.
19
Dosing schedules, preclinical dosing
20
schedules are a common part of our assessments.
21
And again, getting back to some of the comments
22
that were made earlier, the real question is the
A Matter of Record (301) 890-4188
142
1
translatability of those models and really
2
understanding how -- we know that the tumor bearing
3
mice are not the best models. So we really kind of suffer in trying to put
4 5
forward the best recommendations.
And oftentimes,
6
as a result, we'll suggest that they -- contrary to
7
what Bill said, we'll suggest that it really has to
8
happen in the clinic because we don't really know
9
how to best translate those findings.
So we're
10
willing to try with huge caveats associated with
11
that. DR. JANNE:
12
I had a question to our industry
13
colleagues just about -- one of the things we
14
didn't really talk about was sort of half-life of
15
drugs.
16
inhibitors, for example, out in the clinic, that
17
have vastly different half-lives.
18
decisions obviously go into that.
19
And there are certainly -- take MEK
And different
Could people comment a little bit on is that
20
something that is commonly looked upon in this
21
phase of development and only something you learn
22
in the clinic?
A Matter of Record (301) 890-4188
143
1
DR. DAMBACH:
Again, I can comment and say
2
that we absolutely look at that, not only from a
3
safety perspective but also from a pharmacology
4
perspective.
5
perspective, depending on the characteristic of
6
your molecule, promiscuous, non-promiscuous,
7
target, potency, it really can determine the
8
long-term side effects.
9
I mean, certainly from a safety
Likewise, again getting back to this idea
10
about -- you know, most kinase inhibitors in my
11
experience have tended to have some kind of time
12
over a threshold, and so that's a very important
13
thing that we consider for every molecule, trying
14
to understand what is that pharmacodynamic
15
relationship.
16
DR. PARIVAR:
And again, coming back to
17
half-life, of course we take a look at it.
I guess
18
there could be some over emphasis on the kinetic
19
half-life.
20
looking at the target, 10 over kind of rate,
21
because sometimes even a hit and run type of
22
example would be a short half-life drug.
It's probably better described as
A Matter of Record (301) 890-4188
If your
144
1
target inhibition is so long-lasting, even a short
2
half-life drug could be very, very effective.
3
Sometimes it's probably over-seen, and then
4
early on in development, a short half-life drug
5
could be readily gotten rid of, while it could have
6
had a pretty good successful life in the clinical
7
setting of course. Audience Q&A
8 9 10 11
DR. JANNE: audience.
We have some questions from the
Why don't we start there?
DR. LEIGHTON:
John?
There's some debate going on
12
right now about the utility of recovery groups in
13
your toxicology studies.
14
what I'm hearing is that there's a value considered
15
in having these recovery groups as part of your IND
16
enabling in maybe the chronic studies.
17
Am I not hearing that?
So from the discussion,
So there's value
18
added in determining whether or not you
19
should -- whether the lesions are recoverable,
20
whether you should do dose reduction or dose
21
interruption.
22
DR. JONES:
Certainly from my vantage point,
A Matter of Record (301) 890-4188
145
1
John, it is important, and the earlier the better.
2
We tend to do that with our IND enabling studies as
3
part of characterizing the effects we expect to see
4
again. As Donna alluded to, when you look at the
5 6
last 20 years, I think the one thing that has
7
changed significantly in the industry is how much
8
investment we put into the discovery space. So we actually have a base characterization
9 10
of molecules in the programs, what we're going to
11
see.
12
enabling studies is that question of reversibility.
13
Oftentimes, when you get out to the chronic
And an important question in those IND
14
studies, as has been alluded to, you then push the
15
biology, the pathology so far, I'm not sure it's as
16
important then.
17
everyone does, but that is an emphasis for us.
18
We certainly do it.
DR. KLUWE:
I'm not sure
I think one of the risks with
19
the recoveries is that you may be looking at the
20
wrong thing.
21
looked at, it's not the optimum dose, it's not
22
necessarily the clinically relevant toxicity.
So it's not the optimum time that you
A Matter of Record (301) 890-4188
146
So oftentimes, reversibility becomes kind of
1 2
a checkbox; did you look at it, did you not look at
3
it.
4
proactively about is exactly what am I trying to
5
reverse, and at what stage am I trying to reverse
6
it?
7
human response, and am I focusing on reversibility
8
in that particular model, in that particular
9
environment?.
When in fact what we should be thinking more
And do I have a model that's simulating the
So it may be less an issue of do we or do we
10 11
not do recoveries as much as it is have we focused
12
on the most important type of recovery that would
13
actually benefit the clinical paradigm. DR. PALMBY:
14
That was John Leighton from the
15
FDA.
16
audience, when you ask a question, please just
17
introduce yourself first and give your affiliation.
18
Thank you.
19
Could I just ask that folks from the
DR. KADAMBI:
Vic Kadambi, Blueprint
20
Medicines.
We're trying to develop targeted kinase
21
inhibitors.
22
kind of put it in two phases out here.
I had a question about promiscuity and
A Matter of Record (301) 890-4188
147
1
So most kinase inhibitors tend to be a BCS
2
class II or a IV, if you're lucky maybe close to a
3
I.
4
administer to humans is in the high hundreds of
5
milligrams, other than tofacitinib, which is about
6
10 mgs.
7
So traditionally, the dose that you're going to
So the question really is -- and Donna you
8
alluded to, and most kinase inhibitors require a
9
time over threshold IC90, IC80, if you may.
In the
10
early stages you do an IC50, IC50 comparison, and
11
we know that many kinases need not be inhibited to
12
that level to still have an effect, and so are some
13
ion channels and GPCRs.
14
So I'm really surprised with some of the
15
data you presented where attrition is so low, even
16
though your doses could be high.
17
interrogate or investigate that data early on to
18
understand what is IC90 related to IC20 or IC50 to
19
your target and PK/PD analysis?
20
DR. DAMBACH:
How do we
It's a very good question, and
21
I gave a very high level.
But for each molecule,
22
you have to look at each molecule individually.
A Matter of Record (301) 890-4188
148
1 2
And so we interrogate that very closely. Many times -- we drive everything by
3
starting out with potency and this idea about we
4
think is the time over threshold that we need to
5
achieve, and really understand what we believe to
6
be the appropriate pharmacology, and then we relate
7
the other findings that we have around that
8
pharmacology when we do our risk assessment.
9 10
Is that answering your question? DR. KADAMBI:
It is, but I think the
11
challenge always is if you're trying to deal with a
12
disease or a mutant kinase, your animal models do
13
not have that, and then you're assessing it on IC50
14
or IC90 or some PK concentration.
15
an important point.
16
that, too, in his talk.
17
So that becomes
And I think Bill alluded to
I just want to make a comment on
18
reversibility.
19
predict, from microscopic path, what will reverse.
20
And given enough time -- for example, gonadal
21
effects don't reverse in two weeks.
22
So most often than not, we can
I think John's point is a good one.
A Matter of Record (301) 890-4188
149
1
Reversibility is important.
2
day, knowing what the toxicity is, you know how
3
long it's going to take to reverse.
4
not sufficient.
5
heart, it's going to take more than two weeks to
6
recover.
7
than not, and some companies don't do
8
reversibility.
9
But at the end of the
Two weeks is
If you had some remodeling in the
I think it's more a checkbox, many often
DR. PARIVAR:
As it comes to the attrition
10
numbers that Donna presented, I concur with her
11
numbers.
12
said, Donna, that you had about 25 percent failure
13
rate in phase 1, and I guess that that's pretty
14
much accurate.
15
products through phase 1.
16
any kinetics-related topics early on.
17
As a matter of fact we -- I guess you
We pass about 70 percent of our
DR. PERENTESIS:
Very few of them die of
Hi.
John Perentesis from
18
Cincinnati Children's Center, the COG.
19
quick question -- a great session -- about where
20
we're at in terms of preclinical prediction of
21
extreme AE phenotypes in patients, subsets of
22
patients who'd be exquisitely sensitive to
A Matter of Record (301) 890-4188
I had a
150
1
toxicity.
2
view of this.
3
The hypertension model is a great broad
Where are we at in terms of the other
4
genetically determined -- or I guess it's somewhat
5
speculative -- models for side effects?
6
KR. KLUWE:
I think you have to focus the
7
model that you're going to look at on a relevant
8
question, otherwise you're going to raise an awful
9
lot of extraneous information that's hard to
10
predict in advance, even with kinase inhibitors.
11
Certainly, if you're working with a very
12
specific class with toxicities that are known, you
13
can afford to spend a little bit more time modeling
14
around that particular one.
15
particularly when you have a kinase inhibitor that
16
may have a multitude of activities, to come up with
17
what you think might be a more predictive model.
18
In those instances, it might just be the
It's hard,
19
case that you want a little bit of clinical
20
experience before you jump into one of these
21
alternative models, and then try to make sure
22
that's as relevant as possible.
A Matter of Record (301) 890-4188
151
1
DR. JONES:
I think if there's a message in
2
the presentation I gave, it is that as you look at
3
events that have very low incidence of occurrence
4
in the human population, prediction, using any
5
assay approach, is going to be very challenging.
6
So I think when you see a disconnect between the
7
nonclinical and clinical data, you can ask, is that
8
a problem with the model or is it a problem with
9
the way I use the information from the model?
10
We tend to implicate the model is the
11
problem, so let's go to a spontaneously
12
hypertensive rat.
13
example, Bill, where that was very effective.
14
as we change models, we also get into situations
15
where we have less experience with them, we've got
16
less sense of what the balance of sensitivity and
17
specificity are going to be.
And I think you showed a great But
18
So while we can often show those
19
associations, the question is, are they actually
20
going to play forward and improve the overall
21
predictivity.
22
of the accumulation of false positive information,
And what's going to happen in terms
A Matter of Record (301) 890-4188
152
1
which can also damage patients by keeping molecules
2
from entering into the clinic that could possibly
3
deliver benefit. So I think we've got to be very deliberate
4 5
in how we think about the performance of these
6
models and think about how we're going to try to
7
improve our protection of patients in the future. DR. DAMBACH:
8 9
And I would echo those.
need to be very prudent.
We
And often, to say what
10
Tom said, we need clinical information so that we
11
can have a hypothesis around what we're trying to
12
evaluate.
13
would never get any drugs to the market in that
14
case.
15
To do it just as part of a screening, we
DR. PERENTESIS:
And I wasn't looking at it
16
as excluding.
17
the advent of more accessible genomics, there may
18
be ability to predict clinically silent drug
19
metabolism or other types of risk factors that may
20
exclude small populations from exposure to a drug,
21
but in a sense save a good drug.
22
I think in the next five years, with
DR. DAMBACH:
One of the things -- I'd like
A Matter of Record (301) 890-4188
153
1
to be a little provocative for my clinical
2
colleagues, and it may address this in the long
3
term.
4
see from a preclinical safety perspective is I
5
would love for you to take samples from all your
6
cohorts so that if something comes up, we may be
7
able to use these new types of approaches, like
8
IPS, to actually interrogate a controlled group.
But one of the things that I would like to
9
Because what we get is largely empirical,
10
but if there was some mechanism by which we could
11
have these samples and ask a hypothesis driven
12
question, if it's hepatotoxicity, let's make
13
hepatocytes; let's use the cells from the patients
14
where we saw the effect, where we know we can't
15
re-dose them, in the context of their other cohorts
16
in that study, and try to interrogate what it was
17
about that person.
18
I think that kind of leads to a more
19
personalized toxicology assessment, but it may
20
uncover some of the things and some of the actual
21
approaches that I think you're alluding to.
22
DR. PERENTESIS:
Thank you.
A Matter of Record (301) 890-4188
154
1
DR. RUBIN:
Just one quick comment on that.
2
I think we have the samples generally because we're
3
collecting them for other reasons.
4
issues around consent for things that aren't
5
prespecified, just complicated.
6
DR. DAMBACH:
It gets into
So it may be a situation where
7
we can add language around the consent to
8
investigate their toxicity or a group toxicity
9
that's identified.
I know we do it for
10
pharmacogenomics.
I know we do it for looking at
11
tumor samples.
12
something for us to consider how we would put that
13
language in the informed consent and work together
14
to try to elucidate those very unique toxicities,
15
which we don't predict very well preclinically.
But I'll just put it out there as
16
DR. PERENTESIS:
17
DR. RATAIN:
Thank you.
Mark Ratain, University of
18
Chicago.
I want to ask a question about one of the
19
agency's favorite topics, QT prolongation, which is
20
particularly relevant to kinase inhibitors given
21
that it appears more and more that this is
22
non-target toxicity rather than off-target,
A Matter of Record (301) 890-4188
155
1
specifically around PI3K alpha. I was just wondering what companies are
2 3
doing from a screening, from an optimization.
4
unless you're directly targeting PI3K alpha, are
5
you trying to optimize that out of your screen?
6
And also, what is being done for drugs that are
7
being directly targeted PI3K alpha? DR. DAMBACH:
8 9
that.
And
I've had some experience with
What I can tell you is that we are very
10
aware, and you saw the paradigm that we use.
11
very actively use all the tools in our toolbox to
12
try to assess not only electrophysiological but
13
also other causes for -- translational causes that
14
could result in changes in receptor trafficking,
15
things like that, as well as different formats to
16
look at the hematopoietic.
17
So we try to build in and reevaluate.
We
There
18
are new technologies that are coming onboard to
19
make it easier for us to actually incorporate
20
telemetered endpoints into our long-term studies.
21
And I think that's really important for kinase
22
inhibitors as well because it's not just the EFis,
A Matter of Record (301) 890-4188
156
1
right?
It's the functional effects, those we don't
2
catch in a dedicated study. We really need to incorporate those in our
3 4
repeat-dose studies because it takes -- as you even
5
saw in the spontaneous model, it takes days for
6
some of these things to happen. So we're aware of those things and try to be
7 8
mindful in incorporating those in our endpoints,
9
not only for that target but for other targets as
10
well. DR. RATAIN:
11 12
alpha?
13
surprised.
So nobody cares about PI3K
I guess that's what I'm hearing.
14
(Laughter.)
15
DR. WANG:
Diane Wang from Pfizer.
I'm
I have a
16
question about the combination therapies.
17
know, combination therapies have been used more
18
often than not in the drug development right now.
19
And I'm wondering, based on your experience, how
20
the combinability of these agents -- whether it was
21
another chemotherapy or another target therapy, how
22
preclinically this combinability has been evaluated
A Matter of Record (301) 890-4188
As we
157
1
in terms of both sequence of dosing or the ratio of
2
the drug, whether we combine them with the standard
3
of care or we can look at the combination itself,
4
using each of the agents as a primary agent for the
5
combination. DR. DAMBACH:
6 7
Does anyone want to take that
on?
8
(No response.)
9
DR. DAMBACH:
I think we heard initially
10
that there's a struggle on the clinical side, if I
11
understood your question correctly, to understand
12
how best to approach the combination dosing.
13
there's a request for more preclinical information.
14
And
I think on the preclinical side, we struggle
15
with the relevance of some of our models and
16
informing the real sequence.
17
identify the hazard.
18
pharmacology and look for what we presume will be
19
synergistic effects.
I mean, again we can
We can look at the
20
I don't know how good we've been at really
21
answering those questions that you ask, like which
22
will come on the pattern of expression.
A Matter of Record (301) 890-4188
We've
158
1
relied largely on clinical information from our
2
shop.
3
had on the panel.
4
I don't know of other experiences folks have
DR. JONES:
I think this has been a very
5
challenging area just to -- again, when you look at
6
the challenges of predicting from the nonclinical
7
models in single agent.
8
area that some of the early discovery pharmacology
9
models, if there's a real problem in terms of the
So again, it has been an
10
synergistic interaction that's going to be adverse,
11
that usually pops up in those models, and we never
12
go forward.
13
In terms of what value more nonclinical
14
toxicology can provide getting ready I think is one
15
that we've struggled with finding the right place
16
for that.
17
DR. WANG:
Yes.
I think this was our
18
experience, too.
So mostly we normally see the
19
evaluation of efficacy endpoints, see how they
20
combine, whether the synergy or relative effect.
21
But for the toxicology perspective, we rarely see
22
the combination, especially when they have
A Matter of Record (301) 890-4188
159
1
overlapping toxicities, and how a combination will
2
affect the overall safety endpoint at a preclinical
3
stage.
4
DR. VENKATAKRISHNAN:
Hello.
This is
5
Karthik Venkatakrishnan from Takeda Oncology
6
clinical pharmacology.
7
DR. JONES:
You may need to speak up.
8
Actually, the acoustics are a little funny.
9
hard to --
10
DR. VENKATAKRISHNAN:
Okay.
Hi.
It's
My name is
11
Karthik Venkatakrishnan from clinical pharmacology
12
at Takeda Oncology.
13
morning, very nice multi-institution cross sector
14
perspectives.
15
Thank you for the session this
My question is around how we can use the
16
nonclinical information that emerges from both the
17
safety assessments and the pharmacology studies and
18
understanding of the biology of the target in
19
guiding the phase 1 entry into humans plan as to
20
dose schedule decisions, whether to go proactively
21
with more than one dosing schedule, perhaps
22
intermittent as well as continuous dosing, or to
A Matter of Record (301) 890-4188
160
1
respond to clinical toxicities that emerge in the
2
early clinical experience and then modify the
3
program.
4
This perhaps will bleed a little bit into
5
the later conversations that we'll have during the
6
program as well, but I'd be specifically interested
7
in understanding the panel's views on how to
8
integrate nonclinical information to guide that
9
decision.
10
DR. KLUWE:
I think part of the problem is
11
that we tend to fall in love with our models,
12
particularly on the preclinical side.
13
those models on an implanted tumor, there's little
14
question that suppressing the pathway to the
15
maximum, with the longest duration, and so on and
16
so forth, always looks like it's the best regimen.
17
And then all we do in safety is we say what are the
18
consequences of doing that?
19
So with
I think we need to think more in terms of
20
what an in situ tumor situation is like and see if
21
we can model that, what might be a little
22
different.
And the schedule that gives us the
A Matter of Record (301) 890-4188
161
1
optimal suppression in that type of a model might
2
lead us into an opportunity for also suppressing
3
some of the adverse events that we see.
4
frankly, that's a difficult task to put before us
5
preclinically, and for that reason, it hasn't been
6
exploited as much as it perhaps could be.
7
DR. DAMBACH:
But quite
I think along those lines, what
8
we can do -- especially, you have to remember, some of
9
these are novel molecules, and we don't have the
10
clinical experience.
11
what we see preclinically.
12
about whether we think that if this adverse event
13
happened in a human being, could we dose.
14
And so we're relying heavily on And we make an assessment
So sometimes those are the decisions made
15
around alternative dosing schedules.
16
through the attempts to identify, again, using an area
17
over threshold, how long is it over that threshold
18
that we think we'll see efficacy or downstream
19
effects, and we will try to interrogate those
20
endpoints.
21 22
So we'll go
But at the end of the day, if it's your first molecule, or your second molecule, sometimes the
A Matter of Record (301) 890-4188
162
1
knowledge that's taken forward into the clinic is that
2
okay, we'll watch for that, but we're going to go full
3
bore, again, based on the cytotoxicity type of
4
approach, see if we see the AE, then back off and use
5
that information, or cycle back to the preclinical
6
folks with, okay, we saw this AE, and try to start to
7
tease out the PK/PD relationships. I mean, that tends to be the experience
8 9
we've had where we identify the risk.
We may do
10
some work around dose schedule.
But it doesn't
11
necessarily impede our program moving forward.
12
They'll still go aggressively and try to see what
13
the AEs are in the clinic.
14
DR. JONES:
The example I used around this
15
was a case in point where the -- it was actually
16
the severity and the nature of the nonclinical
17
toxicity that created such concern going forward
18
that it sort of drove a lot of that nonclinical
19
work.
20
that what you're referring to, that your hand
21
essentially gets forced in that regard as opposed
22
to believing, a priori, that the nonclinical
So I guess my question to you, Donna, is
A Matter of Record (301) 890-4188
163
1
schedule will translate much better than -DR. DAMBACH:
2
Yes, you articulated that very
3
well.
4
if you have to rely on the nonclinical data.
5
then do you have to generate all this information
6
about where you think the efficacy will take place,
7
where you think the toxicity will take place.
8 9
That's essentially what happens, especially
So your model is exactly spot on.
So
And I
think that is a very common scenario that we're
10
presented with, especially, again, when we have a
11
new molecule or a new target, and we don't really
12
know much about it.
13
DR. BURNS NAAS:
Leigh Ann Burns Naas,
14
Gilead Sciences.
15
question and I think pursue that a little bit
16
further, not so much from the efficacy point of
17
view, because I agree with Tom.
18
going to see profound combination toxicity in your
19
efficacy studies, I think that's going to be pretty
20
obvious, but you may or may not know what that is.
21 22
I want to go back to Diane's
I think if you're
From a regulatory perspective -- and I don't want to speak for any of the regulators in the
A Matter of Record (301) 890-4188
164
1
audience.
But not specifically for oncology, but
2
for other therapeutic areas, there's sort of a
3
heightened awareness or heightened concern around
4
combination safety when you put these drugs
5
together.
6
us to test a combination tox to support clinical
7
development, whether it's early or late.
8
obviously if we have a cause for concern.
But from S9, there's no requirement for
We all evaluate that.
9
Although,
We all consider
10
potential for overlapping tox.
11
potential for PD interaction.
12
certain things, for example the cardiovascular
13
hemodynamic risk, or LEF risk, you can investigate
14
that.
15
We all consider But if you know
But what if you're not -- what if it's not
16
an obvious overlapping toxicity?
17
an obvious overlap in PD?
18
somewhere down here they're connected?
19
What if it's not
Or it's PD distant, but
I guess what I'm starting to consider, at
20
least myself, is whether or not we actually should
21
be doing combination toxicology testing for
22
oncology, specifically for targeted oncology
A Matter of Record (301) 890-4188
165
1
agents.
And I'd be interested in hearing the
2
panel's thoughts on whether you have started to
3
consider that yourselves or not. DR. KLUWE:
4
I think it comes down to let's
5
not dwell on regulatory requirements, let's try to
6
use the knowledge that we have and ask ourselves,
7
is there good reason to do a combination study.
8
And if there is a good reason to do a combination
9
study, what would the design of that study be?
10
What would the impact of it be?
How would I use
11
that information?
12
other one, then we get into this one about, well,
13
somebody else is interested in it.
Because if we fall into the
But as a sponsor of the drug, if I think
14 15
there's a concern, or if I think there's a risk, it
16
doesn't make much difference if there's a
17
requirement to do it or not.
18
it.
19
then be able to express that in terms of what we
20
think the impact would be on the clinical design of
21
such a combination.
22
We should simply do
I mean, it would be silly not to do it, and
DR. BURNS NAAS:
I completely agree with
A Matter of Record (301) 890-4188
166
1
you, but what if there's not an obvious cause for
2
concern?
3
concern, even when we do a combination testing in
4
other therapeutic areas.
5
get, but should we be more proactive in actually
6
doing this, specifically for targeted agents in
7
oncology? DR. KLUWE:
8 9
There's not always an obvious cause for
So the obvious ones I
I think if we wait for a
disaster to occur, and then say we should be more
10
proactive, it's a little bit problematic.
11
again, I think we should look at this, particularly
12
now because we're coming up with not generic
13
agents.
14
These are not cytotoxic.
There
And you say if I
15
double the cytotoxic, I double the cytotoxicity.
16
These are targeted, so we're pretty sure we know
17
what systems they're affected.
18
say is there either a clear risk for combination,
19
is there a clear no-risk for that combination, or
20
is there an area of ambiguity that it would behoove
21
us to spend a little time looking at?
22
necessarily in a prescripted type of study, but in
A Matter of Record (301) 890-4188
And we can look and
Not
167
1 2
a very highly focused and very directed study. DR. BURNS NAAS:
So from a clinical point of
3
view, then, have there been times, clinically,
4
where it has been observed that two agents that you
5
might not expect would have -- that have surprised
6
you, I guess is the best way of putting it.
7
put two together, and you see something you didn't
8
necessarily expect to see, or that may not have
9
been predicted based on the preclinical evaluation
10 11
You
of the two agents independently. DR. RUBIN:
Just being a medical oncologist,
12
I can say there probably are -- but people may want
13
to comment.
14
good example off the top of my head.
15
I don't know.
I can't come up with a
I will say that sometimes, perhaps
16
conversely, I've been surprised at how
17
mechanistically we'll get to a combination where
18
there is the potential for overlapping toxicity,
19
and yet we still manage to get it in without the
20
combination being not developable.
21
probably other people in the audience who might
22
want to comment on that.
A Matter of Record (301) 890-4188
There's
168
1
DR. PARIVAR:
I guess it can go both
2
directions.
3
safety, I would say, yes, I've seen examples
4
that -- especially now, I mean we're going above
5
two agents.
6
together.
7
Specifically, if it comes to hepatic
We're going to look at three agents
The two agents together -- take pancreatic
8
cancer for instance.
You give them Abraxane,
9
fairly safe, manageable.
And then you insert a
10
third agent into the picture, and then suddenly I
11
mean the liver is so compromised anyway because it
12
has METs.
13
So the question is, then, how much reservoir
14
do you have left, and probably it's different from
15
patient to patient.
16
inside there; suddenly your LFTs start kind of
17
going off the roof.
18
You introduce a third agent
So these are the exercises which could
19
potentially have been predicted preclinically if
20
they were tested together rather than us finding it
21
clinically.
22
will predict it and not happens, but sometimes you
It could be like Eric says that you
A Matter of Record (301) 890-4188
169
1
would be expecting it, and it happens. I'm all for it.
2
I do believe that's it's a
3
new era of drug development in multiple agents.
4
The time of single agent therapy probably has done
5
its course.
6
DR. RUBIN:
But those were known monotherapy
7
toxicities, right?
8
was being an example of something where, out of the
9
blue, some new AE came out of a combination that
If I understood, the question
10
you wouldn't have predicted from either
11
monotherapy. DR. PARIVAR:
12
Exactly.
In that case, while
13
rare, it could very well happen because it could be
14
synergism, that different level that you may not
15
know.
16
than doing it preclinically.
17
kind of a standard panel in that case.
18
Then there's no other way of predicting it
MR. AYERS:
So it could be some
My name is Dan Ayers.
19
Vanderbilt.
20
preclinical work and the clinical work.
I'm from
I sense a disconnect between the
21
(Laughter.)
22
MR. AYERS:
And I think that's really
A Matter of Record (301) 890-4188
170
1
becoming a major issue, from my standpoint.
2
Extension trials, extensions on top of the normal
3
dose escalation in phase 1, have increased sample
4
size from about an average of 10 to an average of
5
50 in the last 15 years.
6
have increased to 38 percent of phase 1 clinical
7
trials.
8 9
The extension cohorts
I can give you the citations. So the impact of this is that the cost of
development seem to be going up.
The risk for
10
patients may be going up as well if this disconnect
11
can't be brought together.
12
thought that there was a big disconnect between the
13
clinical and the preclinical work.
14
I just noticed that you
Do you feel that this -- so these are
15
industry, generally industry sponsored kinds of
16
extension cohorts.
17
industry responding to this disconnect with larger
18
extension cohorts in phase 1?
19
endpoints in these are preliminary efficacy and
20
then finding more toxicity.
21
DR. PARIVAR:
22
Is this disconnect?
Is
The two primary
Do you want to go first?
a clinical point of view, you may have a point
A Matter of Record (301) 890-4188
From
171
1
there.
2
The rate of happenings are different.
3
pace and dynamic environment is much more rapid.
4
Things are happening a lot, lot faster in the
5
clinical setting.
6
days.
7
reactive on what we find and we change course.
The clinical
We learn a lot faster these
We do things differently.
We're a bit more
I guess that the ship in the preclinical
8 9
But I don't feel there's a huge disconnect.
setting is much bigger, and for it to change
10
course, it takes a bit longer time.
11
say there's a huge disconnect.
12
data to our clinical scientists, and they
13
interrogate that one, and then when adapted, they
14
change.
15
But I wouldn't
We feedback our
But yes, I mean, a bit more kind of
16
interactive play would be, of course needless to
17
say, desirable.
18
thing for oncology, and my practicing colleagues
19
can comment on that one, is this whole notion of
20
combination therapies, which has been just
21
flourishing in the past couple of years.
22
we have immuno onc [ph] added to the picture, too.
That's at least my -- the biggest
A Matter of Record (301) 890-4188
And now
172
1
And I guess that there probably is where our
2
preclinical setting needs to be changing a bit more
3
aggressively, looking into that one.
4
I don't know.
5
DR. SHAW:
Alice, if you want to.
Well, I guess don't view having
6
these larger expansions as posing more risk to
7
patients.
8
these drugs quicker to get a better handle on
9
toxicity and to get patients access to the drugs,
I feel it's a way for us to develop
10
and for us to learn about efficacy quicker.
11
don't really see that as posing any extra risk to
12
the patients. DR. PARIVAR:
13
No, neither do I.
And I
I don't
14
believe expansion cohorts, per se, are any risk at
15
all.
16
the dynamic safety and kinetics, and enables us to
17
do the PK/PD modeling and simulations that Donna
18
and others, they were asking for.
19
positives as a matter of fact.
20
It actually enables our teams to learn about
DR. DAMBACH:
So those are all
I would agree.
I'd like to
21
know more about the specific area of disconnect.
22
think that we struggle with combination in the
A Matter of Record (301) 890-4188
I
173
1
preclinical space right now.
2
true.
3
similar way, where we're really looking at the
4
question and asking, is there value in doing a
5
nonclinical assay, whether it's in vitro or
6
in vivo?
7
ask each time, but I think that is a big struggle
8
for us.
9
That is absolutely
We're all approaching it probably in a
You know that's, I think, the question we
But with regard to the extended studies, we
10
do the same thing.
11
use it immediately, if there's something that
12
emerges.
13
preclinical space to ask the question about is
14
there anything we can do to inform the clinicians
15
about how to deal with this AE?
16
We get the information, and we
So we'll do the best we can in the
So I don't know if there was another
17
disconnect you had that maybe you felt we were
18
discussing, or maybe you can articulate in
19
particular.
20
MR. AYERS:
21
DR. PINHEIRO:
22
[Inaudible - off mic.] Maybe I can comment, being
the statistician between the nonclinical and the
A Matter of Record (301) 890-4188
174
1
clinical folks here, so I can try to straighten out
2
the bridge.
3
(Laughter.)
4
DR. PINHEIRO:
I think it's a very good
5
question, and I think a question actually to the
6
panel myself is that, how much do Bayesian models
7
get used in the context of trying to connect, let's
8
say, the model that you develop on the preclinical,
9
the PK/PG, or whatever model you have, and the
10
models that will be used on the clinical decisions
11
for phase 1 an even further? So one way that could maybe be made explicit
12 13
is by incorporating, let's say, prior information
14
coming from their own clinical information that you
15
have, but put that into the actual model that you
16
use.
17
and how much you believe in your own clinical data
18
can more informative or less informative priors
19
being used.
20
Some of the parameters that you have in there
I know that Stuart Bailey is going to be
21
talking about that in the next session, so there'll
22
be some discussion.
It's like a plug for what's to
A Matter of Record (301) 890-4188
175
1
come.
2
proposal.
3
But I think it's a very interesting
I don't know how much that's being used.
It
4
definitely can be used.
5
uncomfortable with that because they've made it
6
real explicit, and sometimes it may be how much do
7
I believe that, really, to bring that into my model
8
for the clinical data, rather than, okay, well I'll
9
look at evidence of toxicity so it's like a binary
10 11
Some people may feel like
outcome as well. It doesn't look too good, so therefore don't
12
develop your work, be careful.
13
develop it.
14
belief you have, I think it will be the answer to
15
what you're asking.
16
You're going to
But if you can quantify how much of a
So, yes.
We can leverage that connection,
17
even use other information that we can get, but I
18
don't know how much has that been used or how much
19
would you feel comfortable doing that, I guess, on
20
both sides.
21 22
DR. PARIVAR:
We do use Bayesian models.
Across the board, we do a lot of model-based
A Matter of Record (301) 890-4188
176
1
meta-analyses, and then if the agent, per se, is
2
not first in class, we do use MBTI, which is a kind
3
of targeted kind of the model-based escalation
4
method for which we have data, toxicology data.
5
But if it's an agent which is first in
6
class, a bit more difficult to really rely on
7
model-driven kind of dose escalation upwards.
8
Models are being used as much as we can, but they
9
have their own limitations.
10
But again, coming back to the size of
11
expansion cohorts, I believe there are other
12
driving factors for the size of expansion cohorts.
13
In addition to safety, we are using expansion
14
cohorts very, very effectively to get at more
15
signals of efficacy in the enriched population.
16
And hopefully if you have that one, if you have the
17
safety data, you have the exposure response models,
18
that could be a prelude to a very shorter kind of
19
registration path.
20
So there are regulatory strategy activities
21
which are going into deciding how you design your
22
escalation part, plus the dose expansion and a
A Matter of Record (301) 890-4188
177
1
pivotal study after that for oncology setting,
2
because it's very, very different than a
3
non-oncology drug development pattern.
4
DR. JONES:
I think around the Bayesian
5
models and having good establishment of prior
6
beliefs is really part of the call I think some of
7
us are trying to make in terms of better connecting
8
clinical data flow to the nonclinical data flow,
9
which was alluded by one of the speakers, the
10
opportunities that come from larger sharing of
11
information and safety data across the industry, so
12
that we can begin to get a better picture of what
13
some of the true priors should be around, some of
14
the human outcomes that we're trying to predict.
15
Again, I think it's important to realize
16
that when you look at what little data that exists
17
around some of the correlation or concordance
18
information between clinical, nonclinical, a lot of
19
that is built around target organ level
20
concordance.
21
change in an animal, cardiomyopathy in humans is
22
concordant.
In other words, a blood pressure
A Matter of Record (301) 890-4188
178
These are not direct translations of one-
1 2
for-one of effects in the animals to effects in
3
humans.
4
much more resolved data to be able to approach the
5
kind of real application of a Bayesian view of this
6
as we can.
7
And I think we need to get much better and
From that then, I think there's tremendous
8
power in thinking about how you can stage assays to
9
actually magnify the predictive capability of some
10 11
of the nonclinical approaches that we take. DR. KLUWE:
So it's an interesting concept,
12
and maybe it's something that we could follow up
13
on, with the help of statisticians, to actually
14
say -- we tend to take our models on the
15
preclinical side, and we look at group dynamics.
16
So there's a dose, and it's associated with an
17
effect or lack of effect.
18
information on an individual animal basis.
Yet, we collect all the
19
We could probably go back and deconstruct
20
that, and actually start looking and seeing could
21
we have interpreted that data in a different manner
22
and modeled it differently if we were looking at
A Matter of Record (301) 890-4188
179
1
all of those as independent variables rather than
2
grouping them all together and taking a mean or a
3
median. DR. PALMBY:
4
We're actually just about out
5
of time.
We probably have enough time for one more
6
question, so those who have other questions, please
7
save them for the afternoon session.
8
another panel, then, where you can ask questions. DR. DE ALWIS:
9
There will be
Dinesh De Alwis, Merck.
10
Actually, my question leads on immediately after
11
the statement that was just made by one of the
12
panelists.
13
variability.
14
preclinical species in very homogenous species, and
15
then the key issue in man of course is the
16
heterogeneity of response and the variability that
17
we see.
18
And it's actually regarding So we do these studies in the
Are there circumstances where, based on
19
potential covariates and information that you know
20
about the target or the class of compounds, you
21
could actually do some studies where you could get
22
a handle of that variability?
A Matter of Record (301) 890-4188
Are there ways?
Are
180
1
there examples of that?
2
DR. JONES:
I'm going to go with yes.
3
(Laughter.)
4
DR. PARIVAR:
I would like to -- I concur
5
with the notion, I guess, that structurally this is
6
the way it is.
7
not carry the sort of variability that we see
8
across our clinical setting and patients coming
9
with different age, different comorbidities,
An animal species preclinically do
10
different liver, organ dysfunctions, and such.
11
we collect samples throughout our phase 2 and 3
12
studies, and we look at covariates, and there comes
13
the population analysis, PK/PD analysis.
14
And
Because of lack of such, I'm also very keen
15
to understand that if you can generate any formal
16
data where you can give us some lead where you get
17
to the degree of variability from a preclinical
18
setting, that would be great, but I'm kind of
19
doubtful.
20
So I take it you're a yes, but could you
21
elaborate on that?
22
DR. JONES:
I think it's a great question.
A Matter of Record (301) 890-4188
181
1
It's something that we struggle with a lot trying
2
to understand variability.
3
perspective of in non-oncology indications, having
4
a finding in a single animal and a single dose
5
group, result in a clinical hold, and trying to be
6
able to manage that in the context of what is the
7
variability around a finding.
8 9
Oftentimes from the
We are working with small group sizes of the animals.
Even in the case of the rodent species,
10
when you get to the non-rodents, you're working
11
with much smaller groups.
12
we have to always be aware of.
13
I think it's something
It's always easier to see the role of
14
variability in retrospective analysis than it is in
15
a prospective analysis.
16
do to understand that?
17
sharing across the industry where we begin to get
18
better sense of variability in a broader population
19
of these animal species, perhaps we can.
20
think that's as far as I can go with that.
21 22
DR. PALMBY:
And is there more we can Perhaps again, through data
Okay.
But I
Well, I think that's the
conclusion of the first session.
A Matter of Record (301) 890-4188
We'll reconvene I
182
1
believe at 12:45, after lunch.
2
the panelists and all the speakers from this
3
morning's session.
4
(Applause.
5
(Whereupon, at 11:47 a.m., a lunch recess
6
I want to thank all
It was an excellent discussion. )
was taken.)
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
A Matter of Record (301) 890-4188
183
1
A F T E R N O O N
(12:50 p.m.)
2 3
S E S S I O N
DR. NIE:
Good afternoon, everyone.
4
Dr. Eric Rubin and I would
5
the afternoon session, Dose Finding of Small
6
Molecular Oncology Drugs.
7
Eric, so I will get started.
8 9
like to welcome you to
I'm still waiting for
The agenda, it's a little bit -- it has some minor mistakes because we switched the order.
The
10
first speaker will start at 12:45, then go to 1:00.
11
Then Laura will talk about 15 minutes.
12
break at 1:45 to come back at 1:00 [sic], then
13
everything remains the same.
14 15 16
We take a
Our first speaker is Dr. Nitin Mehrotra. Presentation – Nitin Mehrotra DR. MEHROTRA:
Good afternoon, everyone.
17
name is Nitin Mehrotra.
18
Division of Pharmacometrics, OCP, at the FDA.
19
you can see from the title of my slide, I will be
20
talking about Dose Selection for Small Molecule
21
Oncology Drugs, their present state.
22
I am a team leader in As
I will provide some of the recent examples
A Matter of Record (301) 890-4188
My
184
1
showing rational dose selection, and then we'll end
2
with some key considerations, particularly
3
highlighting the role pharmacometrics can play in
4
tackling the issue of dose selection. This slide depicts some of the emerging
5 6
trends in oncology.
What you can see here are some
7
of the postmarketing requirements and postmarketing
8
commitments, PMRs and PMCs, that have been issued
9
in the last five years, indicating that we are
10
trying to tackle the issue around dosing
11
postmarketing, rather than pre-marketing, which
12
might not be the best and efficient way to do. Having said that, we have seen in the last
13 14
three to four years several examples where a dose
15
lower than the maximum tolerated dose was evaluated
16
and got approved, indicating a paradigm shift in
17
selection of dosing at least for small molecule
18
drugs.
19
Now, if you talk about the current state of
20
dose finding in oncology, particularly focusing on
21
small molecules, maximum tolerated dose paradigm
22
still is the most used way to select the dose.
A Matter of Record (301) 890-4188
185
1
When we understand the mechanism of action of the
2
drug, the target inhibition data or the early
3
biomarker data is not critically evaluated to
4
inform does in phase 1 or phase 2. We all know that phase 2 trials rarely
5 6
evaluate more than one dose level, and there is
7
limited evaluation of various dosing regimens.
8
also, the characterization of PK and PD and
9
exposure response in early clinical trials is not
10
consistent across INDs, based on our recent
11
experience.
And
12
What is the question we want to answer?
13
What is the key question in front of all of us?
14
we have the right dose?
15
the most logical way to approach this question is
16
to do dose-ranging trials.
Do
In my personal opinion,
You will hear from several speakers in this
17 18
session and sessions tomorrow, what are the
19
possibilities of conducting dose-ranging, good
20
dose-ranging trials?
21
that can be utilized to conduct good dose-ranging
22
trials?
What are the characteristics
What are the pitfalls?
A Matter of Record (301) 890-4188
What are the
186
1 2
challenges? From my viewpoint, there are three
3
particular things, which I believe should be
4
utilized in terms of dose selection in the context
5
of oncology.
6
Model-based dose selection, which integrates
7
the knowledge that you have from various pieces to
8
develop a quantitative framework, which can be very
9
efficient.
10
Secondly, exposure response of efficacy and
11
safety should be conducted throughout clinical
12
development programs.
13
Thirdly, we should invest in disease models
14
where applicable to understand the tumor response
15
outcome relationship such that we could project the
16
likelihood of success of various phase 2 dosing
17
regimens or doses based on the tumor data.
18
Let me shift gears and provide you with a
19
couple of examples where the sponsor utilized the
20
information around PK, IC90, and exposure response
21
to select the dose.
22
This is an example for idelalasib.
A Matter of Record (301) 890-4188
For the
187
1
ease of pronunciation, I will call it idela moving
2
forward.
3
dose lower than MTD was granted.
4
And this is a case where approval of a
Just a little bit of background.
Idela is a
5
specific inhibitor for PI3K-delta with an EC90 of
6
125 nanogram per mL.
7
eight hours.
8
dose-ranging trial that evaluated both QD and BID
9
dosing regimens up to 350 mg BID.
It has a half-life of around
Sponsor conducted a dose-finding,
The important
10
thing to note is the MTD was not reached by 350 mg
11
BID, and then the proposed dosing regimen for all
12
the indications was 150 mg BID.
13 14 15
Now, the key question for us was, is 150 mg BID the right dose? Apart from efficacy and safety trial data,
16
which form the primary evidence of efficacy and
17
safety of 150 mg BID dosing regimen in NHL and CLL
18
population, there were two key pieces of
19
information that were utilized to support the
20
dosing regimen; first from a PK perspective, second
21
from an exposure response perspective.
22
Let us see those one by one.
A Matter of Record (301) 890-4188
PK
188
1
justification for 150 mg BID dosing regimen, shown
2
here is a plot comparing steady state, trough
3
concentrations of various dosing regimens that was
4
evaluated in the trial.
5
regimens; on the right you see 5 BID dosing
6
regimens.
7
idela.
8 9
On the left, you see 2 QD
The red dotted line shows the EC90 for
There are two things which can be clearly seen here.
First, the QD regimen, as expected,
10
produces less or lower C-trough concentrations
11
compared to BID dosing regimen.
12
150 mg BID dose produces exposures, for the most
13
part, which are higher than the EC90 for idela.
And second, the
14
Now since there was limited PK data
15
available at each dose level, what was done to add
16
more confidence to it -- so there was a population
17
PK model available, and PK simulations were
18
conducted to support the 150 mg BID dosing regimen.
19
What you see here is the cumulative
20
distribution of steady-state C-trough for various
21
BID dosing regimen.
22
150 mg BID dose, indicating that given the
The solid green line is for
A Matter of Record (301) 890-4188
189
1
variability, you expect around 3 percent of the
2
patients to have steady-state trough levels below
3
EC90.
4
proportion would increase to 11 and 37 percent,
5
respectively.
And with lower dosing regimens of 150, the
What was the evidence from an exposure
6 7
response perspective?
The sponsor conducted a
8
univariate exposure response analysis.
9
divided the data they had available in NHL patients
10
into four groups based on their steady-state trough
11
concentrations and plotted them against best
12
reduction from baseline and percent.
They
13
What you can see here is that as the
14
exposure increases from Q1 to Q4, there is increase
15
in reduction in tumor size, which plateaus after
16
second quartile.
17
dosing regimen fall in the plateau region of the
18
curve.
19
And the exposures of 150 mg BID
This information tells us two things.
20
First, if you go lower than 150 BID, you probably
21
may experience lack of efficacy, or if you go
22
higher than 150 mg BID, there may not be much
A Matter of Record (301) 890-4188
190
1
improvement on efficacy side.
2
So in summary, sponsor utilized PK
3
information to justify why the BID dosing regimen
4
is needed from a target engagement perspective, and
5
then also utilized the clinical information
6
collected to say why 150 mg BID dosing is
7
appropriate. Let me very briefly talk about a very
8 9
different example for axitinib, which is approved
10
for advanced renal cell carcinoma, where the
11
sponsor utilized dose individualization in their
12
drug development program. In session 3, Dr. Yazdi will talk in detail
13 14
about this example, but my intent here is to show
15
how the dose individualization concept was
16
weavened [ph] in the drug development program,
17
which allowed for labeling as a dose titration
18
study.
19
Just a little bit of background.
Axitinib
20
is inhibitor of VEGFR receptors 1, 2, and 3.
21
dosing regimen, which is approved for axitinib, is
22
5 mg BID as a starting dose, which can be increased
A Matter of Record (301) 890-4188
The
191
1
to 7, and up to 10 mg BID based on tolerability. The doses can also be de-escalated down to
2 3
2 mg BID based on tolerability as well.
Important
4
thing to note here is 5 mg BID is the MTD.
5
though the dose is higher than 5 mg BID approved on
6
the label, the sponsor identified 5 mg BID as the
7
MTD.
So even
8
The titration scheme was prospectively
9
evaluated in both phase 2 and phase 3 clinical
10
trials.
The primary adverse events of interest
11
were hypertension, proteinuria, diarrhea, and
12
fatigue.
13
the proposed titration-based dosing regimen
14
appropriate?
So the key question in this case was, is
15
To understand this, first we are to
16
understand what is the relationship between
17
exposures and adverse events.
18
done, and it was found out that there is an
19
exposure-dependent increase in hypertension and
20
proteinuria.
21 22
The analysis was
What you see on the left-hand side is the relationship between axitinib daily AUC and
A Matter of Record (301) 890-4188
192
1
probability of hypertension; on the right-hand
2
side, there is a similar relationship for
3
probability of proteinuria. What you can see from this relationship is
4 5
there is an exposure-dependent increase, and the
6
patients who have lower exposures at 5 mg BID dose
7
level are likely to experience less adverse events.
8
So somehow, if we can identify the patients in the
9
lower exposure group, we can only increase dose in
10
those patients to match the exposures in the other
11
group.
12
What the applicant showed was that dose
13
titration based on tolerability produces similar
14
exposures across doses.
15
the box plot for patients prior to dose titration,
16
stratified by their highest BID dosing received.
17
Again, all these three groups are the exposures at
18
5 mg BID level.
19
And what is shown here is
For this group of patients who had larger,
20
higher distribution of exposures, they were not the
21
candidates for up titration, probably because of
22
adverse events like hypertension or proteinuria.
A Matter of Record (301) 890-4188
193
1
However, when the exposures were seen in the
2
patients who eventually were up titrated, it was
3
readily apparent that these were the patients who
4
had higher clearance and lower exposures such that
5
if doses were increased in these patients, the
6
exposures would match to the exposures observed
7
with 5 mg BID in the other subgroup, indirectly
8
saying the exposures were not an overshoot [ph].
9
In summary, the high individual variability
10
for this drug was realized early in the clinical
11
development program.
12
utilized dose titration as their strategy, and
13
eventually was approved on the label.
14
imagine from a patient perspective, probably it's
15
the best thing, because you are optimizing efficacy
16
and safety at an individual level.
The sponsor initiated and
You can
17
Let me just provide you some of blinded
18
examples from IND space, which we have recently
19
seen, which show the promise of learn and apply as
20
the drug development moves on from phase 1 to
21
phase 2 to phase 3.
22
In the first example, the sponsor was taking
A Matter of Record (301) 890-4188
194
1
fixed dosing regimen forward for the phase 3 trial.
2
That was their plan.
3
trial data for this drug X, it was clear that there
4
were higher discontinuation and dose reductions in
5
patients who had lower body weight.
When they saw the phase 2
When this was tied up to PK, what was seen
6 7
was it was primarily the higher exposure in low
8
body weight patients which was driving higher
9
discontinuation rate.
So prospectively, a decision
10
was made that body weight-based dosing should be
11
done in the phase 3 trial, and the dosing that was
12
proposed in the phase 2 trial was stratified by
13
body weight. In a second example, there was a PK/PD
14 15
model-based dose selection for phase 3, which we
16
rarely see in the area of oncology.
17
sponsor did a phase 2 trial with two doses.
18
developed the PK/PD model based on the PD
19
biomarker.
20
found out that efficacy was driven by PD biomarker
21
level.
22
that phase 3 doses should be stratified based on
In this case, They
Once the analysis was done, it was
Sponsor did several simulations to show
A Matter of Record (301) 890-4188
195
1
baseline PD biomarker.
2
The important thing to note here is the
3
phase 2 trial was not done based on baseline PD
4
count.
It was done regardless of baseline PD
5
count.
So in a way, you can imagine this PK/PD
6
simulation-based exercise alleviated the need of
7
any additional trial.
8 9
Let me just summarize our experience in the context of dose selection from a regulatory
10
perspective.
11
a chronological manner?
12
What kind of evidence is generated in
What I have here is the six pieces of
13
information row-wise, which we think should be
14
collected during the clinical drug development
15
program:
16
almost always done; modeling and simulation to
17
design trials; assessing efficacy of lower doses,
18
alternate regimens; conducting exposure response
19
for efficacy and safety at the IND level;
20
covariate-based dosing in the registration trials;
21
and submission of exposure response for efficacy
22
and safety in the NDA or BLA submissions.
MTD, determination of MTD, which is
A Matter of Record (301) 890-4188
196
What I have in four columns are four
1 2
hypothetical scenarios, scenarios 1 to 4, just to
3
iterate where we are in terms of dose selection and
4
where ideally we would like to be. For example, in scenario 1, MTD is
5 6
determined.
We see NDA/BLA.
There are exposure
7
response analyses done to justify the dosing
8
regimen, but the other pieces in between are
9
missing.
10
In scenario 2, which we are seeing often
11
now, is the utilization of exposure and response
12
for both efficacy and safety endpoints at IND
13
stages.
14
modeling and simulation more and more to design
15
trial, rather than empirical approach, and to
16
assess efficacy of lower doses or alternate dosing
17
regimens.
18
What we should be aiming for is to utilize
We should even take the science a step
19
forward where instead of one dose fixing all
20
strategy, we utilize covariate-based dosing
21
prospectively in the clinical trial to get adequate
22
safety and efficacy information in the
A Matter of Record (301) 890-4188
197
1
subpopulation.
2
somewhere here.
3
here.
4
Again, to reiterate, we are Ideally, we should be somewhere
Just to sum up the dose selection future
5
consideration sort of wish list, I think
6
dose-ranging trials are probably the best way to
7
answer this question, and you'll hear from some of
8
the speakers today and tomorrow on what kind of
9
trials are feasible, possible.
10
Pharmacometric exposure response models have
11
shown promise.
12
justification should be included in all phases of
13
drug development.
14
quantitative analysis, for example, multivariate
15
analysis to adjust for confounding factors, or
16
modeling the time course of adverse events to
17
identify an optimal dosing regimen.
18
disease models, as I previously mentioned, wherever
19
applicable, should be considered.
20
Exposure response-based
There is room for innovative,
Then using
Hopefully, we can reach a stage where
21
integrated dose-response models along with modeling
22
and simulation is very routinely used to design
A Matter of Record (301) 890-4188
198
1
trials, and then building dose individualization
2
concept as part of the clinical drug development
3
program is a smart way to approach dosing and
4
eventually labeling.
5
Just to sum up, like with any other
6
therapeutic area, I believe dose selection is an
7
important issue for oncology as well.
8
response analysis along with other quantitative
9
analysis should be routinely conducted across drug
10 11 12
development program. I would close here and thank you for all your attention.
13
(Applause.)
14
DR. NIE:
15 16 17
And exposure
Our second speaker is Dr. Dinesh
de Alwis from Merck. Presentation – Dinesh De Alwis DR. DE ALWIS:
Good afternoon.
I realize
18
I've got the post-lunch session, so I realize
19
there's a lot of PK/PD going on in the audience at
20
the moment.
21
(Laughter.)
22
The title of my talk is Optimal Dosing for
A Matter of Record (301) 890-4188
199
1
Targeted Therapies in Oncology, and I'm actually
2
going to be very brave here and actually talk about
3
two drug development case studies as leading by
4
example.
5
if I don't get the facts absolutely correct,
6
because I do realize there are people who are more
7
intimately involved with these particular cases
8
than I am.
9
So please bear with me and be kind to me
I realize I'm the only one who actually has
10
a disclosure statement, but this is my disclosure
11
statement.
12
review of the literature, and this was from the
13
years 2000 to 2013.
14
where we had an MTD that had been published and
15
what the approved dose was, subsequently.
16
based on this, worked out this diagram here, which
17
actually is very interesting I think.
18
So what we did was actually we did a
And we looked at all drugs
So we had 77 compounds in total.
And then
This is
19
both large and small molecules.
20
was actually small molecules.
21
was actually two-thirds of the compounds are
22
actually approved at doses lower than the MTD, and
A Matter of Record (301) 890-4188
Majority of this And what we found
200
1
roughly a third are approved at half that, of half
2
of MTD.
3
MTD.
4
even higher than the MTD.
5
And only a third of drugs are equal to the
And in fact, in a few cases, they're actually
I thought this was an interesting finding.
6
So clearly, MTD is not always the approved dose,
7
which is a good thing.
8
it took to get there.
9
But it depends on how long
The next slide is actually another
10
interesting review, and this is by authors in 2010,
11
where they actually looked at 201 phase 1 trials,
12
and they looked at 119 cytotoxics and 82
13
non-cytotoxic studies.
14
that only in about 50 percent of the time in
15
non-cytotoxic trials, they could actually even
16
identify a maximum tolerate dose.
17
significant issue if MTD is a paradigm that you
18
want to adopt.
19
And what they found was
So this is a
Of this 50 percent, only 30 percent report
20
an objectively quantifiable clinical toxicity.
And
21
what do I mean?
22
basically have a subjective toxicity such as pain,
Because the rest of the time, you
A Matter of Record (301) 890-4188
201
1
fatigue, or other related side effects that can be
2
objectively quantified.
3
So it's kind of a given that as the process
4
of drug development goes on, that MTD changes due
5
to the subjectivity of the assessment.
6
have an assessment clearly where even establishing
7
MTD is a problem.
8 9
So now we
The next is another paper, and this is from the Clinical Cancer Research in 2010, and this was
10
from 24 trials treating 683 patients from 2004 to
11
2008 at MD Anderson Cancer Center.
12
What you have here was in order to compare
13
these 24 trials across, they normalized the
14
patients into four groups:
15
a high dose that excluded doses above MTD, and a
16
high dose that included all patients.
17
look at the graph on the left, while phase 1 was
18
never really set up to assess efficacy -- they did
19
that just to understand this.
20
a low dose, a mid-dose,
And if you
So they looked at the number of complete
21
responders, partial responders, and stable disease,
22
and what they found was actually there wasn't a
A Matter of Record (301) 890-4188
202
1
statistical difference between the four groups.
2
However, if you look at the graph on the right,
3
they actually did find a statistical significance
4
in terms of the percentage of patients that were
5
off trial for toxicity.
6
They looked at 3 months, 6 months, and 12
7
months, and they found a statistical significance
8
between the low dose and the high dose.
9
remember, there was no difference in terms of at
And
10
least the efficacy that they assess between all
11
those four dose groups.
12
So what this really tells you is that doses
13
reaching MTD increase toxicity without necessarily
14
improving response in phase 1 oncology patients.
15
We did do a lot of review.
As you can see,
16
I've been looking at a lot of papers over the last
17
few weeks and months with respect to this talk.
18
And I looked at the literature in terms of phase 1
19
design, and we've spent an inordinate amount of
20
time, over the last two decades, on getting to a
21
DLT and establishing an MTD.
22
designs; continual reassessment methods, modified
So we have 3 plus 3
A Matter of Record (301) 890-4188
203
1
CRM, accelerated titration.
2
not an exhaustive list here.
And remember, this is There's many more.
But what this is all focused on is getting
3 4
to that DLT, and then subsequently an MTD.
So it's
5
very much in the paradigm of cytotoxics.
6
think what we heard this morning, and I think the
7
general theme of this meeting is, where with
8
cytotoxics, toxicity was possibly a good thing
9
there, anti-proliferative toxicity related to
And I
10
efficacy, this is not the case with targeted
11
agents.
12
round hole.
13
And it's like fitting a square peg in a
What we need to do more of is actually this
14
diagram on the right, which is exploring the
15
biologically effective dose range, and I mean
16
exploring a minimal effective dose, a biological
17
effective dose, and a maximum tolerated dose as
18
well, so really understand the response surface.
19
Understand the heterogeneity of response.
20
Understand the subpopulations that might be
21
responding better than -- overall, you might not
22
have a response, but if you look at a particular
A Matter of Record (301) 890-4188
204
1
subpopulation, you have this response.
2
really exploring the right-hand corner of this
3
slide.
4
So it's
In terms of what we should be doing in terms
5
of a phase 1B/2 trial, the design of these -- I
6
think Nitin has touched on some of these
7
concepts -- but we should be looking -- and if you
8
go back to the slide here, if you're going to
9
explore this properly, we should actually be doing
10
this in a randomized setting -- and this is in
11
order to avoid potential bias -- and evaluate more
12
than one dose.
13
Clearly, that's a common theme.
And look at
14
surrogate endpoints to make early go/no-go
15
decisions based on safety, efficacy, and where
16
possible, pharmacodynamic marker in order to
17
understand the response better.
18
We really need to look for factors
19
contributing to individualized response, and this
20
can be demographics or tumor response, et cetera,
21
and establish the proof of concept.
22
Now, this tends to be an overused word, but
A Matter of Record (301) 890-4188
205
1
what I mean by this is really good pharmacology, so
2
has the drug reached the site of action?
3
of mechanism been established?
4
was a hypothesis that a certain level of inhibition
5
was required for efficacy, but have you seen that
6
in this study?
7
Has proof
At the start, there
Has efficacy been established in the
8
targeted patient population?
And if possible, is
9
it actually better than standard of care, because
10
that's really important.
11
importantly, has an adequate therapeutic range been
12
established?
13
small molecule oncology drugs, this involves
14
chronic oral dosing.
15
And then finally and most
And this is difficult.
Given it's
The first example I'm going to talk
16
about -- actually Tom Jones talked about this
17
earlier in the morning, and someone told me that
18
repetition is the mother of all learning, so I hope
19
the second time around it sounds better.
20
This is the TGF-beta compound, the
21
transforming growth factor; it's a small molecule
22
type 1 receptor antagonist.
And here, since I'm a
A Matter of Record (301) 890-4188
206
1
PK/PD scientist -- I've got much more PK/PD than
2
Tom Jones did, so anyway.
3
the -- I think Tom outlined the scenario.
4
there was a significant issue with cardiac
5
toxicity, cardiac valvulopathy.
6
related tox.
7
So here you have Clearly,
This was targeted
So clearly, the team had to come up with a
8
way of dosing this compound that was it even
9
possible to manipulate this target effectively and
10
test the paradigm in the clinic with a significant
11
margin of safety to do that safely.
12
A number of preclinical experiments are done
13
way before they even thought of going to the
14
clinic, and I think that's the difference with this
15
program, in terms of both efficacy and tox studies,
16
to really understand this.
17
There were several xenograft experiments
18
done, and they linked the pharmacokinetics to this
19
piece-MAD phosphorylation, which is a biomarker,
20
and which in turn was linked to efficacy.
21
were able to really understand that you only needed
22
actually 30 percent target inhibition over
A Matter of Record (301) 890-4188
And they
207
1 2
24 hours. This is very important because the previous
3
companies that have gone ahead with manipulating
4
this target with a small molecule had gone for
5
maximal target inhibition, and obviously there was
6
no margin of safety.
7
Here, 30 percent of inhibition was required
8
for efficacy, but more importantly, they explored
9
schedule dependency.
So based on the PK/PD, they
10
were able to identify that there was a significant
11
signal transduction event post-piece-MAD
12
inhibition, which meant actually that you could
13
give two weeks on, two weeks off dosing, and get
14
the same efficacy in the preclinical species as
15
chronic oral dosing.
16
Now, this doesn't seem like much, but there
17
was a significant increase in the margin of safety.
18
So this helped, really, the program feel confident
19
in taking this forward to the clinic safely.
20
they decided the dose range was 160 to
21
360 -- actually it wasn't 300, it was 360 -- total
22
daily dose that they wanted to explore in the
A Matter of Record (301) 890-4188
So
208
1
clinic, and they took this forward into the clinic.
2
The other thing I just wanted to tell you
3
was the phase 1 study, the objective of the phase
4
was not to establish an MTD.
5
outlined as not an objective in the study.
6
Actually, that was
The objective of the study was there was a
7
dose range they wanted to explore, this what they
8
wanted to do, and they were going to look at this.
9
And if there was no efficacy in that patient
10
population, then potentially they would have to
11
stop the development of this compound.
12
So they went in -- and this is actually in
13
all glioblastoma patients.
14
which was a dose escalation part, and part B, which
15
was a dose confirmation part.
16
glioblastoma patients.
17
There was a part A,
Both of them had
This is from the escalation part, and this
18
is the famous waterfall plot I'm sure many of you
19
are familiar with.
20
baseline.
21
are a significant number of patients here with
22
clearly a benefit here in terms of reduction of
And this has changed from
And what you can clearly see is there
A Matter of Record (301) 890-4188
209
1
tumor size and gone on from several cycles.
2
this was further confirmed with the part B in
3
primary glioblastoma patients.
4
And
So what happened here was that the predicted
5
biological effective dose was between 160 and 360,
6
but based on the pharmacokinetic data, they adapted
7
that to 160 to 300.
8
range, they were able to establish that efficacy
9
was seen, and this compound now is in phase 2/3.
10
So in fact, in that dose
The next example is crizotinib, which I
11
think many of you are very familiar, and this
12
compound clearly was actually developed at the
13
time, back in 2006, if memory serves me well, as a
14
MET inhibitor in phase 1.
15
So they had a standard dose escalation part,
16
and they started at 50 milligrams QD, and they went
17
up to 300 milligrams BID.
18
dose escalation, establish MTD in the solid tumors,
19
et cetera.
20
And it was a standard
During the course of this escalation, they
21
found 2 patients who had remarkable efficacy, 2
22
non-small cell lung cancer patients.
A Matter of Record (301) 890-4188
And at a
210
1
similar time frame, as luck would have it in 2007,
2
ALK was recognized as a molecular target in
3
non-small cell lung cancer.
4
into being, that was an "Ah ha" moment, and those
5
at Pfizer I think did two things.
6
So the science came
One, clearly really looked at this data very
7
carefully.
8
experiments but also amend the phase 1 to look at a
9
particular group of patients, and I'll come to that
10 11
They decided to do some preclinical
in a moment. So they looked at this -- first the
12
preclinical data was generated, and what you see
13
here was in fact -- this is ALK inhibition and
14
tumor growth inhibition, and as you can see, it's a
15
beautiful correlation.
16
dreams of PK/PD scientists are made of, very easy
17
to estimate and an excellent example of good
18
correlation.
19
This is really what I guess
Then they looked at the MET inhibition, and
20
actually it wasn't so great.
So you needed
21
something like a maximal sort of MET inhibition to
22
get around 50 percent reduction in tumor growth.
A Matter of Record (301) 890-4188
211
1
So in fact, it was clear, looking at this, that it
2
was a wonderful ALK inhibitor, and likely the
3
efficacy is driven by ALK inhibition and not MET
4
inhibition.
5
As I said, they amended that phase 1 trial,
6
and they did a whole, completely different
7
approach.
8
changing the standard paradigm.
9
was in fact screen 1500 patients, 1500 non-small
And this shows you the power of actually And what they did
10
cell lung cancer patients, to get something like 82
11
ALK positive patients, because the occurrence was
12
about 5 percent of these ALK positive tumors.
13
they did a study -- and this is published in the
14
New England Journal of Medicine -- and got close to
15
60 percent response rate and 72 percent PFS at
16
6 months.
17
And
Now, what was interesting with all this,
18
remember that MTD that was selected was
19
250 milligram BID, and it so happened that, the
20
preclinical information as well, you had 75 percent
21
inhibition in ALK around that dose.
22
So here the science and the kind of more
A Matter of Record (301) 890-4188
212
1
traditional with the amendment to -- based on the
2
data that emerged from the clinic, really came to a
3
good place to really get an optimal outcome for
4
patients.
5
really using that information as you go along.
6
So this is an excellent example of
Next, we talked a bit about
7
pharmacodynamics, but I think one of the things
8
that we have at our fingertips, but we don't tend
9
to use that much in the manner I think we should
10 11
doing, is changing tumor size. Here, you have an example on the vertical
12
axis, change in tumor size, resist classification,
13
and response classification.
14
a highly continuous set of data to an ordered
15
categorical to a categorical data, so you're losing
16
information as you go down this spectrum.
17
So you're going from
Here you have 5 patients or 5 example
18
outcomes, and what you can see here is if you look
19
at patients B and D is, if you look at this, you
20
have a 15 percent increase and a 25 percent
21
shrinkage in tumor based on a change in tumor size.
22
But here the patient is considered stable disease,
A Matter of Record (301) 890-4188
213
1
and to me that doesn't seem right.
I mean, a 15
2
percent increase and a 25 percent shrinkage in
3
tumor are two different things.
4
information by using resist classification.
So you're losing
Similarly, if you look at patients C and D,
5 6
you have a 25 percent shrinkage and a 35 percent
7
shrinkage, but in fact, one is called stable
8
disease and one is called a partial response. The same is true here.
9
You hear both are
10
just considered responder, but one is 35 percent
11
shrinkage and one is 100 percent shrinkage in
12
tumor.
13
by using a categorical or ordered categorical
14
scale.
15
So you're fundamentally losing information
This is not a novel concept by the way.
As
16
far back as 1981, Lavin showed that actually a
17
sample size required for a continuous endpoint can
18
be something like 50 percent less than needed for a
19
dichotomous endpoint for randomized two-arm study.
20
So we should really be using this.
21
information is there.
22
in our early trials.
This
This is routinely collected
A Matter of Record (301) 890-4188
214
1
How can we use change in tumor size?
It
2
enables early assessment of response, so we can
3
take it week 9, week 18, week 27, based on some
4
understanding of the particular class of compounds.
5
It could be that we need to look at week 27 versus
6
week 9, but in some cases week 9 may be
7
appropriate, or we should be using all of it, 9,
8
18, and 27 scans, and incorporating this in a
9
longitudinal model.
10
The rest is proof of concept.
We can
11
actually demonstrate exposure response using this.
12
And at the minimum, the drug should not exhibit
13
tumor growth.
14
just see tumor growth for a compound, really, I
15
don't think you've got a compound on your hands.
16
So I think this can be used to actually
I think we can all agree that if you
17
establish that, the ability to select
18
subpopulations that respond due to increased power
19
compared to RR.
20
you can actually start looking at covariates.
21
can start looking at initial tumor size, tumor
22
burden, biomarker status, ECOG status, et cetera.
So here, based on a smaller study,
A Matter of Record (301) 890-4188
You
215
1
And we can have the ability to identify optimal
2
dose and indication. Finally, we can also link to survival and
3 4
PFS.
And I think there's an individual here in the
5
audience who played a very critical role in doing
6
that within the FDA, and particularly for a
7
non-small cell lung cancer model.
8
been other indications like colorectal and others,
9
where they have been able to show change in tumor
10
size to survival.
11
to make early go/no-go decisions.
12
But there have
So basically, we could use this
There is a critique of tumor size, and the
13
critique of it is that you can't ignore new
14
lesions, dropout, and death.
15
challenging to mixing continuous tumor size and
16
event data.
17
approaches, and one of them is a probabilistic
18
Bayesian method to estimate the probability of
19
response.
20
lesions, dropout, and death with continuous tumor
21
size information.
22
So there's a
But there are some proposed
So this integrates information about new
We at Merck, we've carried out some
A Matter of Record (301) 890-4188
216
1
simulations rerunning past trials and show that the
2
probability of response gives the same power with
3
something like 25 to 50 percent fewer subjects in
4
phase 1 and phase 2 trials, or substantially more
5
power with the same number of subjects. In conclusion, I think it's time that we
6 7
have a paradigm change for oncology dose-finding.
8
And MTD is sometimes useful, and as I showed, only
9
sometimes possible.
So clearly, we need to look at
10
other factors and really explore the surface.
We
11
need to explore the minimal efficacious dose, the
12
biological effective dose, and the maximum
13
tolerated dose. We need to use all available information.
14 15
This can be toxicity, preclinical, preclinical and
16
clinical PK/PD information, and really incorporate
17
this.
18
for that class of compounds.
19
the competitive space and inform the design.
20
we can this in a Bayesian manner or even a more
21
traditional manner.
22
It can be for your particular compound or Use information on
Tumor size change aligns with the
A Matter of Record (301) 890-4188
And
217
1
therapeutic goal of anticancer treatment.
2
we should use it.
3
including tumor size is going to be really critical
4
for designing efficient and rapid studies.
5
the paradigm of where -- when you're going to a
6
phase 1 -- and I think many of you who are in
7
industry will know that.
8 9
I think
Use of continuous endpoints
And in
If you going to phase 1 and you haven't designed it properly, and you get negative data,
10
unfortunately, it becomes very difficult to change
11
governance or change people's thinking in terms of
12
actually this is a good drug, let's explore it
13
further.
14
first time if possible.
15
use all possible information.
16
So we need to do this -- get it right the And to do that, we need to
I think the key objective, or at least I
17
hope so, is to really determine the most
18
efficacious dose regimen or the most rapidly
19
registerable, if that is indeed a word, and
20
indication, and the most responsive patients.
21 22
Finally, I'd like to acknowledge several of my colleagues who provided significant input into
A Matter of Record (301) 890-4188
218
1
some of this material.
So thank you.
2
(Applause.)
3
DR. RUBIN:
4
I'm Eric Rubin from Merck.
Thanks, Dinesh. I'll be
5
introducing the subsequent speakers.
We're going
6
to have three talks from statisticians, so we'll
7
have a statistical view of dose-finding approaches
8
in oncology.
9
Fernandes from the FDA, and she'll be speaking
The first will be from Dr. Laura
10
about innovations in dose-finding clinical trial
11
designs.
12 13
Presentation – Laura Fernandes DR. FERNANDES:
Good afternoon, and welcome
14
to my talk on Innovations in Dose Finding Clinical
15
Trial Designs.
16
introduction.
17
FDA in Oncology Products and the Office of
18
Biostatistics.
19
Thank you, Eric, for the I'm a statistician working at the
This is the FDA disclaimer, and basically it
20
means that I'm held accountable for the thoughts
21
presented in these slides, and I have no financial
22
disclosures to report.
A Matter of Record (301) 890-4188
219
1
The outline of my talk is as follows.
After
2
a brief introduction on dose-finding trials in
3
oncology, I will present some of the methods used
4
in such clinical trials, the gaps in current
5
methods, and the kinds of innovations we hope to
6
see in the future.
7
My talk will focus on the aspects of
8
biostatistics as opposed to pharmacometrics views
9
presented by Nitin and Dinesh in the earlier talks,
10
and kind of set the stage for later speakers in the
11
session.
12
At this start of the dose-finding trial, we
13
have developed a well-characterized small molecule.
14
We have completed nonclinical PK/PD modeling and
15
have also completed nonclinical trials in animals.
16
We are now interested in determining the best dose
17
level to be administered to treat patients
18
effectively and safely.
19
So given that we have a potential study
20
drug, the correct dose level has to be determined
21
from a set of predefined dose levels.
22
three main goals of a dose-finding trial in
A Matter of Record (301) 890-4188
There are
220
1
oncology.
2
the MTD.
3
the patients.
4
patients at biologically active dose levels and
5
gaining experience at the recommended dose level.
6
Goal 1 is to obtain a good estimate of Goal 2 is to limit the number of DLTs in And Goal 3 aims at treating the
Traditionally, phase 1 cytotoxic drug trials
7
evaluated only the MTD and safety in healthy
8
subjects, but with the recent trend of inclusion of
9
cancer patients in this phase of testing, we now
10
see trials monitoring efficacy, for example, in
11
terms of response rates.
12
Common features of the typical dose-finding
13
clinical trial are based on cytotoxic drugs.
14
trials are small, non-randomized, and sequential in
15
nature, wherein patients are unaware of the dose
16
level that they will be receiving.
17
These
Patients are assigned to a dose level based
18
on the responses of earlier patients in the trial.
19
Since cytotoxic drugs are administered in cycles,
20
typically 21- or 28-day cycles, most of the
21
dose-finding trials and methods are based on such
22
fixed dosing cycles.
Dose-finding clinical trial
A Matter of Record (301) 890-4188
221
1
designs can be broadly classified into algorithmic
2
and model-based designs. Moving on to algorithmic designs, the 3 plus
3 4
3 design is the most commonly used algorithmic
5
design and is an example of the general A plus B
6
design.
7
particular dose level before recruiting another
8
cohort of 3 patients at either the same dose level
9
or at the next dose level, depending upon the
10 11
Patients are treated in cohorts of 3 at a
observed DLTs. There are many variations of this design.
12
One of the main properties of these
13
algorithmic-based designs is that they are easy to
14
follow and execute.
15
mathematical framework and a statistical model, and
16
do not allow the flexibility to treat patients at
17
the best available dose level.
18
On the flip side, they lack a
In addition, they use information only from
19
the current cohort of patients and ignore the
20
information from earlier patients in the trial
21
while making dose assignments for the new cohort of
22
patients, thereby losing out on efficiency.
A Matter of Record (301) 890-4188
222
1
In recent years, we have seen examples of
2
model-based approaches to dose-finding trials.
3
These methods typically have three components.
4
They are driven by a statistical model that
5
describes the dose toxicity relationship, which is
6
usually a non-decreasing function of toxicity with
7
increasing levels of the drug.
8 9
These methods also employ Bayesian methodology for the estimation of the parameters of
10
the model.
11
current beliefs about the dose toxicity
12
relationship in the model.
13
case, the prior that is used says that the toxicity
14
will be around 0.2, so it's a 20 percent chance of
15
probability at this dose level.
16
Priors are set up that incorporate our
For example, in this
Finally, we have a framework to use all the
17
data available from the patients in the study at
18
that point.
19
estimated and used to draw inference about the
20
probability of toxicity on each of the dose levels.
21 22
The posterior distribution is
Using all the available information helps in obtaining precise estimates of the MTD and in
A Matter of Record (301) 890-4188
223
1
assigning the safest and highest dose level to the
2
next patient in the study.
3
Some examples of the model-based design
4
include the CRM, modified version of the CRM with
5
the time to event component, accelerated titration
6
designs, EWOC, and mTPI.
7
As mentioned briefly in the earlier two
8
slides, the model-based methods can be viewed as
9
designs with memory and allow the MTD estimation to
10
be precise and accurate, owing to the fact that all
11
the data from all the patients until that point is
12
considered in obtaining the parameter estimates;
13
hence, sometimes described as designs with memory.
14
As a result of this, there are fewer cases of
15
dose-limiting toxicities, and a higher percentage
16
of patients are treated at and or near the MTD.
17
All the model-based designs are known to be
18
efficient.
In a recent publication, based on
19
citation database search index of phase 1 trials in
20
cancer between 1991 and 2006, only 20 of the 1235
21
trials, a mere 1.6 percent, followed a model-based
22
dose-finding design.
A Matter of Record (301) 890-4188
224
1
Let's move on to some successful examples of
2
using model-based designs.
3
2014 by Dr. Alice Shaw and colleagues for ceritinib
4
is an example of a model-based trial design.
5
clinical trial for ALK rearranged non-small cell
6
lung cancer included a dose escalation phase
7
followed by an expansion phase in which all the
8
patients received treatment at the maximum dose
9
established in the dose escalation phase.
10
This study published in
This
Dose escalation was guided by means of a
11
2 parameter Bayesian logistic regression model with
12
the use of the principle of escalation with
13
overdose control.
14
ceritinib dose on day 1, followed by a 3-day
15
pharmacokinetic evaluation period and subsequent
16
daily oral dosing for the remainder of the cycle.
17
Treatment included a single
Daily dosing of ceritinib was continued in
18
21-day cycles.
The starting dose was 50 milligrams
19
daily on the basis of nonclinical safety data.
20
Priors were set up on the parameters in the
21
Bayesian model based on the nonclinical data, and
22
the responses from only the first cycle were used
A Matter of Record (301) 890-4188
225
1
to fit the Bayesian model. This table shows the number of patients that
2 3
were treated at each of the 9 dose levels.
4
is a doubling of the dose levels followed by
5
100-milligram increments.
6
were treated in the first phase.
7
phase, a total of 114 patients received treatment
8
with the MTD of ceritinib that had been established
9
in the first phase, which was 750 milligrams once
10
There
A total of 39 patients For the expansion
daily. It should be noted that 74 percent of
11 12
patients required at least one dose reduction or
13
interruption due to ARs in the second half of the
14
study.
15
only the data from the first cycle was considered
16
and the long-term use of the drug was ignored while
17
establishing the MTD.
18
This can be attributed to the fact that
This slide basically summarizes what we see
19
lacking in the current field of dose-finding
20
clinical trials, trials that do not use all the
21
information from each patient and all the patients
22
in the trial; trial designs that do incorporate the
A Matter of Record (301) 890-4188
226
1
rich source of information in nonclinical and
2
historical data; a lack of monitoring of patients
3
for longer duration and for long-term safety and
4
delayed toxicities; small molecule dose-finding
5
trials that do not mimic the longer duration of
6
drug usage beyond the 21- or 28-day cycle and have
7
frequent dose reductions in later stages of
8
clinical testing. This brings me to our wish list of what we
9 10
would like to see in future clinical trial designs.
11
There are many possible solutions, and I have
12
listed only a subset of the current options in
13
publications as examples, so don't feel bad if your
14
publication did not make it to the slides.
15
Most dose-finding studies include data only
16
from the first cycle for each of the patients, when
17
in reality the study drug is administered over
18
multiple days and multiple cycles.
19
could be administered over multiple cycles, and the
20
patient could be allowed to escalate the dose
21
level.
22
Study drug
Obtaining data from multiple cycles could
A Matter of Record (301) 890-4188
227
1
also aid in arriving at a better estimate of the
2
effective dose and prevent dose reductions due to
3
safety in future trials.
4
methods include intra-patient dose escalation
5
model, extensions to the CRM for repeated measures,
6
cumulative hazard model for schedules, or the
7
Markov model for repeated cycles.
Some of the possible
8
Secondly, some of the safety signals and
9
toxicities are observed beyond the 21-day cycle.
10
For example, in a pooled analysis of clinical
11
trials involving 9 kinase inhibitors of VEGFR2, the
12
median time until the first incidence of grade 3 or
13
greater hemorrhagic event was 78 days.
14
observing the toxicity in a 21- or 28-day cycle is
15
not adequate, and we need to have better study
16
design approaches in identifying such toxicities.
17
We need to think about how we can
Hence,
18
incorporate long-term effects of the study drug in
19
the study design, possibly using the cumulative
20
hazard model for schedules or some extensions to
21
the TITE-CRM model for repeated measures.
22
Moving on, how do we study dose combinations
A Matter of Record (301) 890-4188
228
1
with more than two study drugs tested
2
simultaneously, possibly using the 2-D dose-finding
3
trial model or the two agent dose-finding model.
4
Incorporating data from nonclinical and other
5
clinical studies in the trial design would help us
6
obtain more from our trial.
7
through using priors in a Bayesian model, and we
8
need to find ways to assess these priors possibly
9
through modeling and simulations.
This could be done
I'm going to say this again.
10
We definitely
11
need to think beyond the 21- or 28-day fixed cycle
12
in the context of non-cytotoxic study drugs that
13
are administered on a continuous, daily, long-term
14
basis.
15
Finally, there is a need to evaluate
16
Bayesian models when using different kinds of prior
17
information and the validation of the software that
18
is used to draw inference.
19
undertake evaluation of safety and efficacy through
20
a formal model design, and the bivariate continual
21
reassessment method could be possibly used for this
22
purpose.
Finally, we need to
A Matter of Record (301) 890-4188
229
1
In a recent paper by Mandrekar and
2
colleagues in 2010, the author concluded that
3
pragmatic challenges with implementing a
4
model-based design outweigh the scientific reasons.
5
Some of the challenges faced are lack of
6
familiarity with the design, fear of the black box
7
design, fear of regulatory acceptance, and more
8
importantly, resistance to change and unwilling to
9
be the first to try a new approach.
10
I would like to say that the FDA is open to
11
looking at novel approaches to dose-finding
12
clinical trials.
13
rise up to the challenge and pave the way for
14
innovations in dose-finding trials developed in a
15
clinical paradigm through concerted efforts from
16
experts in various disciplines, then the acceptance
17
of these designs may be quicker and easier.
18 19
I'm sure if statisticians would
Some of my references, and thank you for your attention.
20
(Applause.)
21
DR. RUBIN:
22
Thanks a lot.
So we're running
about 10 minutes ahead of time, so I've been told
A Matter of Record (301) 890-4188
230
1
that I should relay information that we're going to
2
have a 10 minute break.
3
back by five after 2, rather than 2:10, that would
4
be great.
Thank you.
(Whereupon, at 1:49 p.m., a recess was
5 6
So if people could come
taken.) DR. RUBIN:
7
So I think we'll go ahead and
8
try to keep on track with our 2:05 restart here in
9
our session on Design of Dose Finding Studies.
And
10
as I mentioned, we have two additional statistical
11
perspectives.
12
Bailey, who's the global head of Early Clinical
13
Biostatistics at Novartis Oncology, and he'll be
14
talking about best practices of adaptive
15
dose-finding studies.
The first will be from Dr. Stuart
Stuart?
Presentation – Stuart Bailey
16
DR. BAILEY:
17
So good afternoon.
I'd like to
18
thank the organizers for inviting me to speak
19
today.
20
having a disclosure information slide, so he's not
21
the only one.
22
shareholder.
I'm going to make Dinesh happy by also
I work for Novartis, and I'm a
A Matter of Record (301) 890-4188
231
I'm going to talk to you today about best
1 2
practices for adaptive design.
To give you a
3
little background, within Novartis, about 2003, we
4
began changing the paradigm away from 3 plus 3
5
designs into what was modified continual
6
reassessment method, and Laura introduced those to
7
us earlier on.
8
around those from a modeling perspective, and in
9
2005, 2006, we actually changed to a slightly more
And we had significant challenges
10
flexible approach.
The concept was still the same,
11
but we changed the modeling approach. Since 2006, we've run over 150 phase 1
12 13
trials that utilize adaptive Bayesian phase 1
14
designs.
15
35 of those are actually combinations of either two
16
or more drugs, both combinations of new chemical
17
entities, as well as combinations with registered
18
partners as well.
19
from 10 years of my experience within Novartis, and
20
I'm certainly happy to hear feedback from others as
21
well.
22
Somewhere in the region of around 30 to
So I'm giving you a perspective
I think starting point for any development
A Matter of Record (301) 890-4188
232
1
of a drug, whether you use an adaptive design or
2
not, is clearly you have to understand your drug.
3
We've heard this morning clearly the concept of
4
bringing the information from preclinical into the
5
design of your trial is essential.
6
understanding what kind of adverse events I'm going
7
to see, but understanding the pathway I'm actually
8
going to hit, the mechanism, whether a drug by its
9
own nature doesn't need to go beyond a certain dose
10 11
That's not just
level in order to have its best activity. There are some drugs, particularly in the
12
immuno space, not necessarily the small molecule
13
space, but in the immunize space, where going
14
beyond a certain level this drug starts to inhibit
15
its own activity.
16
to reach MTD.
17
And therefore, you may not need
So the concept of more is better is a
18
challenging one for me.
19
definitely indications or situations where more
20
definitely is better, but it shouldn't be that we
21
go in with that as our only assumption.
22
There are certainly
In any design, we have to maintain the
A Matter of Record (301) 890-4188
233
1
protection of patients.
We just heard the concept
2
of the phase 1 trial for ceritinib.
3
show you an example of another phase 1 trial.
4
matter what we base our endpoints on for the
5
escalation, there has to be an underlying principle
6
for the protection of the patient safety.
7
phase 1 trial; that's what our studies are there
8
for.
I'm going to No
It's a
9
So while I will tell you that we shouldn't
10
be targeting MTDs, we shouldn't be considering DLT
11
as a primary endpoint, actually, we should be using
12
them in order to protect the patients and maintain
13
a level of security around the doses we will study
14
from the other endpoints we're going to use.
15
It's critical to incorporate preclinical or
16
historical information, not just preclinical
17
information, but we may in the combination space,
18
once we incorporate what we know about the single
19
agents, the dose toxicity relationship, we may have
20
preclinical information there about drug-drug
21
interaction.
22
When we talk about interaction of two
A Matter of Record (301) 890-4188
234
1
compounds when you're combining them together, it's
2
not just PK drug-drug interaction.
3
actually the safety interaction.
4
pharmacokinetic interaction at all, but the actual
5
safety overlaps to a point that you see increases
6
in frequency or grading of adverse events
7
irrespective of any DDI from the drugs themselves.
8 9
There's You may see no
Preclinical information about the mechanism of action will give us a lot of information about
10
the kind of biomarkers that we need to collect and
11
also the timing of the collection, and it may lead
12
us into upfront an idea around patient selection.
13
But again, patient selection is something that
14
during the development of the study, you may start
15
to change your idea about which patients you need
16
to recruit.
17
certain copy number, but as you start to go through
18
the trial, you may need to expand that or enclose
19
that patient population.
20
You may look for patients with a
So any design that you want to use has to
21
incorporate information but also be flexible enough
22
to adapt to the information that you're actually
A Matter of Record (301) 890-4188
235
1
obtaining on the study.
2
Communication, in terms of best practices,
3
the one learning I've taken from 10 years of doing
4
this, is communication is the absolute key thing.
5
This is not just as a statistician.
6
communicate statistical concepts to non-
7
statisticians.
8
statisticians, to regulators.
I have to
I have to communicate them to
The communication of what I can consider to
9 10
be a simple concept, such as uncertainty, can be
11
very difficult for someone whose mind is not
12
necessarily in the area of uncertainty or to
13
quantifying uncertainty.
14
6 patients.
15
not.
16
I've seen two DLTs in
I know the drug's too toxic.
Maybe
The same is said that as a statistician, I
17
can't solve everything.
And as the person
18
implementing a design, I need to implement a design
19
that sets boundaries and thresholds on where we can
20
go and where we can't go based on risk.
21
are other factors beyond purely one simple model in
22
terms of how you actually make a decision.
A Matter of Record (301) 890-4188
But there
236
1
So simple language such as a model makes a
2
recommendation is a language we've tried to get
3
away from.
4
identify potential doses, and all other information
5
and the expertise of investigators, clinicians
6
treating patients, will actually make a
7
recommendation from that set of doses.
8
you examples of that as we move forward.
A model is going to be in place to
I'll give
9
So if you think in the simplest case in the
10
traditional designs, we take -- and we don't ignore
11
preclinical information.
12
information.
13
where we think we're going to get to.
14
classical case was to do a modified Fibonacci
15
approach to selecting dose levels through a DLT
16
algorithm; 3 plus 3 was the standard approach and
17
will determine an MTD.
18
We take preclinical
We identify starting dose.
We know
And the
Our first attempts were, still with the
19
mindset of MTD, to change the algorithm in place,
20
to introduce something more flexible, a Bayesian
21
logistic regression model, and there are references
22
in the end for the design; but to take all of the
A Matter of Record (301) 890-4188
237
1
historical information to incorporate it into a
2
modeling environment, to make more efficient use of
3
the dose toxicity relationship between dose and
4
DLT, still with the concept of identifying an MTD.
5
Then, based on the information we collect in
6
the escalation, to select a dose for the
7
expansion -- it may be the MTD, it may be a lower
8
dose -- and then subsequently determine the
9
recommended phase 2 dose.
And the key there is the
10
word "a."
11
And that, I can tell you, is something that I would
12
strongly push our regulatory friends and other
13
industry colleagues and the investigators to come
14
back to companies to say, "Don't assume it should
15
only be one dose.
16
assessing multiple dose levels."
17
We select "a" dose to do an expansion.
You should be open-minded to
So why a Bayesian environment?
The Bayesian
18
approach, as has been nicely shown earlier on, is a
19
way for us to formally incorporate information, be
20
it information from preclinical, information from
21
other patients.
22
knowledge and assess a risk, in this case to future
It allows us to quantify our
A Matter of Record (301) 890-4188
238
1
patients for experiencing DLTs. The reference I mentioned is here.
2
The
3
paper, the original paper, was Neuenschwander
4
et al.
5
changes to the standard CRM approach.
6
really focus on the appropriateness of a model.
7
The CRM originally was a one parameter statistical
8
model.
9
dose to select, and then optimize the allocation of
10 11
They key from this paper, there's two One was to
And it was really designed to tell you the
patients to doses around the region of the MTD. It saw a number of cases where you would
12
have overdoses.
13
had then staggered escalation restrictions on
14
increases.
15
to move away from the optimization of that model
16
using a one parameter approach.
17
Modifications to it meant that you
And as you start to do that, you start
The model doesn't allow effective modeling
18
of the dose DLT relationship, so although
19
simulation show long-term characteristics of one
20
and two parameter models are relatively similar,
21
actually a significant number of studies, both
22
within our company and other companies, have then
A Matter of Record (301) 890-4188
239
1
called us up to say how do we fix our trial?
We've
2
used a one parameter model; struggled with the
3
issues that they were seeing recommendations from a
4
model that were inconsistent with data, seeing DLTs
5
on trials, and recommendations to store double
6
doses. So it's absolutely critical to have the
7 8
appropriate model.
The second thing is the mindset
9
that the model is not there to make a
10
recommendation.
That was the two differences from
11
the standard approach. Clearly, the approach allows you to then
12 13
have flexibility to incorporate additional patients
14
at varying dose levels.
15
protect the patients within the trial, but as we
16
start to see that data, we start to open ourselves
17
up to other questions based on the information
18
we've seen.
Our primary goal is to
So as we're starting to observe some
19 20
activity, whilst we continue escalation, we may
21
want to incorporate additional patients at lower
22
doses.
And this is perfectly allowable within the
A Matter of Record (301) 890-4188
240
1
Bayesian approach.
We can still then incorporate
2
their toxicity information into the modeling at
3
subsequent stages in order to better estimate the
4
dose DLT relationship, but also to get really the
5
data we need to answer the other more important
6
questions. It allows us also to re-escalate, and that's
7 8
a critical piece that's missing from algorithmic
9
designs.
And I'll come to an example.
When I give
10
the clinical example, I'll show you where we
11
actually run into a situation where re-escalation
12
was quite beneficial. Clearly, it allows you to explore what most
13 14
people call unplanned doses.
Unplanned doses, at
15
the beginning of a trial we used to prespecify the
16
dose levels.
17
did was prespecify a dose range of interest, and
18
any of the doses within there are doses that we
19
should be considering on the trial, obviously with
20
an understanding of difference between
21
pharmacokinetic properties between two different
22
doses.
The reality is that what we really
You don't want two doses that are almost
A Matter of Record (301) 890-4188
241
1 2
exactly the same. But we should have the flexibility within
3
our trial, based on potentially tablet sizes as the
4
one limiting factor, to select whichever dose falls
5
within the range that we feel is appropriate.
6
We also are able to look at adapting the
7
schedules or formulations based on what we've seen
8
in the preclinical studies.
9
have some assumptions about the impacts of those
10 11
That requires us to
changes. Having an understanding of the underlying
12
properties of different formulations or preclinical
13
evidence about the impact of changing schedules on
14
the DLT relationship, you can actually incorporate
15
that in the modeling framework and use data from
16
one schedule and the assumption about the change in
17
schedule to predict the doses for the new schedule,
18
meaning that we're actually protecting patients
19
coming into that new schedule upfront.
20
Now, I was commenting on the need for
21
communication being one of the challenging things.
22
When you come to the concept of uncertainty, even
A Matter of Record (301) 890-4188
242
1
for statisticians this can be sometimes a
2
challenging concept.
3
communicate these kind of complex ideas in pretty
4
simple and effective ways.
5
doing anything is drawing a picture and using that
6
to translate from statistical language into
7
something else.
8
So we need to be able to
And my favorite way of
So for a statistician, for example, the
9
graph that you see here, we see the median for the
10
prior distribution for logistic parameters for the
11
preclinical data.
12
gives us a median DLT, dose DLT relationship, with
13
dotted lines showing us a 95 percent credible
14
interval.
15
people who aren't statisticians and probably to
16
some who are.
That's the solid red lines that
And that's quite confusing to most
17
For the clinicians, it's a simple language
18
to say what you see here is our estimated dose DLT
19
relationship based on our animal data and our level
20
of uncertainty.
21 22
To come to some of the earlier comments about, for example, positive predictive values or
A Matter of Record (301) 890-4188
243
1
negative predictive values, on the left-hand side,
2
you'll see an uncertainty boundary, these dotted
3
lines, where we're saying we pretty much know
4
nothing about the dose DLT relationship. On the right-hand side, you'll see quite a
5 6
different looking curve.
At dose levels below 300,
7
we've got some level of information, and this can
8
come from a number of different ways.
9
represent, for example, preclinical variability.
This could
We may have two or three different species
10 11
tested.
All of them tell us that doses below 300
12
look to be safe.
13
predictive value is fairly strong.
14
certainty about those doses.
And we know our negative So we have more
15
As we move above, we start to understand
16
that higher dose levels, some of the species are
17
telling us that the DLT rates are predicting an MTD
18
somewhere in the range of 400 milligrams, others
19
may be higher.
20
uncertainty there, and this could be reflective of
21
that.
22
So we've got clearly higher
It's also reflective of a situation where we
A Matter of Record (301) 890-4188
244
1
may have studied this drug in a non-oncology
2
indication and had some safety data at low doses,
3
and you can see that incorporated.
4
high doses, we still have no information.
5
are different ways to be able to get this.
And yet, at So there
But what we like to do is to explain to the
6 7
medical counterparts that what this means is that
8
although this solid red line is a best guess, the
9
reality can be very, very different.
So a simple
10
way of showing this is by plotting some random
11
samples of dose toxicity relationships from this
12
relationship.
13
So taking the different distributions, you
14
can see that every time a line is created, this is
15
a random sample, a random sample from the
16
parameters that underlie the dose toxicity
17
relationship, and show possible dose toxicity
18
relationships from these prior distributions. Very, very different relationships.
19
Here,
20
you see very flat, safe, everything looks to be
21
fine.
22
parameter model is that whilst our initial
Here, very steep.
The issue with the one
A Matter of Record (301) 890-4188
245
1
relationship was somewhere around here, you would
2
never be able to have something that goes lower and
3
come above.
4
to reflect a true dose toxicity relationship.
5
So it's under parameterized to be able
You can see as you start to generate this,
6
you start to understand more about the
7
distribution.
8
means for a specific dose level, if we take the
9
second distribution, the informative one, and look
Translating this into what this
10
at 300 milligrams, every time one of these curves
11
crosses 300 milligrams, we can actually estimate
12
the probability of DLT.
13
generate, we start to understand the likelihood
14
that the DLT rate is a specific value, and we start
15
to understand the uncertainty about this.
And as this starts to
16
So these simple graphics can help a
17
clinician or a non-statistician to truly understand
18
what we mean by uncertainty for DLT rates.
19
what we are then able to do is we're able to take
20
this distribution for the probability of DLT and
21
split it.
22
a traditional definition of 33 percent DLT rate,
We now split it into two areas.
A Matter of Record (301) 890-4188
And
We use
246
1
and we look at the chance that given our
2
information to date that the true DLT rate lies
3
above this value.
4
Now in this situation for 300 milligrams,
5
the chance of over toxicity is 34 percent.
6
relying on the translation of information, so this
7
is the best guess that we have for the chance of
8
overdose.
9
overdose control principle from Babb et al., we are
10
This is
Now, based on the escalation with
happy to set a boundary of 25 percent of risk.
11
We have simulated this in many, many
12
situations and shown that this can control the risk
13
to expose patients to overdoses on trials.
14
this boundary is a fairly well-understood boundary
15
now, but this is now considered too high.
16
would be for us a dose level that you would not
17
consider to use for patients in the next cohort.
18
And
So this
With a different set of data at a lower
19
dose -- you see here we've got a much lower chance
20
of overdose at just 9 percent.
This one would be a
21
dose that could be considered.
So I'll come to the
22
"could be" in a moment.
What we end up doing is
A Matter of Record (301) 890-4188
247
1
for every single dose level, making an assessment
2
of this chance of overdose and identifying doses
3
that satisfy this EWOC criteria. Now, as I said, traditionally we used
4 5
language, and as a statistician, when I use
6
language "the model recommends the following
7
doses," my assumption is that we then use our other
8
information to choose from them. Other interpretations of the word
9 10
"recommendation" is the model is telling me what to
11
do.
12
of dose escalation meetings say, "I want to meet
13
this Dr. BLRM because he's telling me to do this."
14
And I said, he's not here actually.
15
And I've actually heard an investigator in one
It's a --
But clearly, the one thing that this model
16
is telling us is these doses are not acceptable.
17
We are not going there.
18
though, we can now start to use our other
19
information in order to select the most appropriate
20
dose for use.
21 22
From this range of doses
So what we're in essence looking at now is not a paradigm where we escalate through to MTD.
A Matter of Record (301) 890-4188
248
1
And I can't agree more with Dinesh's conclusion
2
earlier on that we should start to move away from
3
this mindset, that our goal is MTD.
4
presentation, goal number 1 I would say should only
5
be there if necessary.
6
And to Laura's
We have a model that allows us to deal with
7
the DLT.
What we need to have is better ways to
8
assess adverse events in general, and there are
9
additional models that can be put into place.
This
10
model actually can be adjusted in multiple ways to
11
look at bi-cycle events.
12
We have examples of studies where we
13
incorporated three cycle toxicity because of late
14
onset neutropenia.
15
you can do in conjunction with them all.
16
start to just assess time to onset of adverse
17
events, time to dose reductions, and then look at
18
other risk boundaries for them, conditional on
19
satisfying the primary DLT model.
20
You have also other things that So you
You also generally look at tolerability, so
21
we need to assess the relevant dose intensity.
I
22
mentioned already interruption for reductions.
And
A Matter of Record (301) 890-4188
249
1
we start to assess for this dose range, where in
2
this dose range do we sit in comparison to our
3
preclinical data?
4
or the Cmax that we expected?
5
that is telling us that we've got patients being
6
exposed above the IC90 for such an extended period
7
of time that they're having adverse events that we
8
could actually change the frequency of the dosing?
9
And we can actually use this data to adjust
10 11
Are we in the range of the AUC Is there PK data
schedules on the study. One key thing, which has poorly been done in
12
most studies in the past and even for us in an
13
adaptive setting, we were recently in a case where
14
we're actually getting live pharmacodynamic
15
data -- is to be able to get this data in, not just
16
to be able to say I'm in the range of dosing I
17
expect to be in, but at that range I'm having the
18
effect I expect it to have.
19
This is a critical part of operationalizing
20
a design that has generally been ignored.
21
Pharmacodynamic data tends to be retrospectively
22
analyzed, and I plead everyone to push to
A Matter of Record (301) 890-4188
250
1
have -- if you're in a trial, to take the samples,
2
to ship them, to have them analyzed.
3
in a company designing the trials, to implore that
4
these are done on a proactive fashion, because this
5
with the pharmacokinetic data is the key for us to
6
be able to assess where we are and what changes
7
need to be made. Of course, we have efficacy data.
8 9
And if you're
My
background, my PhD was in bivariate safety and
10
efficacy modeling, and I thought I was going to
11
come into industry and change the world with a
12
wonderful bivariate design.
13
than the ones designed.
It was slightly better
The reality was efficacy comes in much, much
14 15
later.
It comes in at maybe week 6, but is one
16
observation of a reduction enough?
17
see multiple observations?
18
confirmation?
19
think of the immuno setting where activity is
20
likely to come much later, the ability to make
21
decisions driven by a bivariate model looking at
22
safety and efficacy, taken at the same time point,
Do you need to
Do you need
So it's quite possible.
A Matter of Record (301) 890-4188
And if you
251
1
is unlikely to be the case.
2
ability to explore a range, and then to go back
3
once we start to see the efficacy, to then adjust
4
and add patients as we're going.
5
We need to have the
So again, best practices, insurance of the
6
availability of data.
7
real time because it's the only way that we can
8
actually review both the acute and the chronic
9
treatment effects and make dosing decisions about
10 11
We have to have this data in
our recommended dose at every step. We should allow the design to impact the
12
on-study decisions.
13
start to come into the operational side of
14
prioritization in your questions.
15
thing that we look to do is change the dose.
16
As I said, here is where you
So the first
At some point when we realize the dose, we
17
can't change the dose, we start to look at
18
schedule.
19
formulation.
20
the dose to a point where you can no longer take
21
it, that actually it's the schedule you need to
22
change earlier.
Then we start to look at potentially It may be that rather than pushing
A Matter of Record (301) 890-4188
252
1
You need the ability or the flexibility
2
within your design, and this is one of the key
3
learnings for us as well about communication with
4
regulators, to very clearly and explicitly within
5
your protocols state how you intend to do that.
6
It may be that you -- I mean, clearly at the
7
start of the trial, you don't have the data to tell
8
you what the change would be.
9
understanding from your preclinical study.
You have an But
10
what you do have is an ability to state how you
11
will use the data you have at the time in order to
12
make a change, and that's not about writing an
13
algorithm to say, "At this point when I see this, I
14
will do this."
15
data with this modeling approach, we will use it to
16
determine a dose, and this will be our approach to
17
then choose a dose from that information."
It's about saying, "When I get this
18
It's quite fair for some regulators to come
19
out and say, "Well, when you've got that data, show
20
it to us, and then we'll be happy to take it
21
forward."
22
doing this many times that we will gain that
But it's only through the experience of
A Matter of Record (301) 890-4188
253
1
acceptance, and actually, yes, it will be an
2
appropriate way to deal with those changes and not
3
have to question it.
4
not something that's simply ad hoc.
It is a preplanned analysis;
Dose expansion, as I already mentioned,
5 6
should not always be restricted to a single dose,
7
though, there may be times where the data really
8
tells you you need to go with this.
9
afterwards about designs that look at multi-dose
10
phase 2 trials.
11
as a standard.
12
been a challenge to implement.
Jose will talk
And outside oncology, this is done And yet, inside oncology, this has
The other thing, as much as a statistician
13 14
can be -- that to really use all of this
15
information to classify where not to go, the
16
strongest part of any study is the team working to
17
identify doses together.
18
work with a fantastic group of medics,
19
pharmacologists, pharmacokineticists, biomarker
20
specialists, some are in the room in order to
21
identify doses that don't just rely on this DLT
22
model.
I'm very fortunate to
A Matter of Record (301) 890-4188
254
1
We need experienced investigators.
Alice
2
Shaw was here earlier on, on the end of the panel
3
session, with the ceritinib trial.
4
vast network now of investigators like Alice who
5
have worked with these designs, who understand the
6
designs, and who can give us feedback about how to
7
change those designs because we're not saying that
8
everything is perfect right now.
9
evolve.
And we have a
We still need to
We still need to understand how to better
10
change things.
So that's also something that is a
11
best practices, to not assume that you've reached
12
the pinnacle. So I mentioned about the review of all study
13 14
data.
To give you an example, we are talking about
15
having the pharmacokinetic data available to us at
16
the time point that we need it, to make a decision
17
about the next dose level, to be able to present
18
actively within a cohort, pharmacodynamic or
19
laboratory data; not just to say there was an
20
adverse event, but actually to look at time
21
profiles of changes in specific values related to
22
the drug from our preclinical data; to understand
A Matter of Record (301) 890-4188
255
1
that at a specific time point, there may be a need
2
to analyze a large set of data in order to not
3
restrict ourselves based on DLT, but to restrict
4
ourselves based on something else.
5
An example is QT data, to be able to predict
6
the point that we do not want to go beyond this
7
value, even if from a DLT we've not yet seen DLTs.
8
Of course, we've got our primary model.
9
we've got to the point where we say, well, we now
And once
10
know the range of doses to look at, we start to
11
understand the efficacy data in that neighborhood
12
of those doses, and not just where we're seeing
13
responses, but in who we're seeing the responses.
14
So actually, being able to bring in our
15
genetic data, and we're now seeing genetic data
16
coming in anywhere from two weeks to four weeks
17
after samples are being taken.
18
allowing us to say, in specifically the patients
19
that we think we should have seen activity, are we
20
seeing it?
21 22
And this is
If we're seeing it in patients who are not the patients that we thought we would see it in,
A Matter of Record (301) 890-4188
256
1
why are we seeing it there?
And we can go back to
2
our preclinical teams and start to try to
3
understand this additional activity. So it is a paradigm shift away from DLT.
4
Of
5
course we have the DLT model to identify this, but
6
we have to use this other information.
7
there's a responsibility for very strong
8
statistical experience here, but there is a
9
significant level of responsibility beyond that in
10
So clearly,
terms of the additional data. So quickly, I'm going to go through a case
11 12
study.
This was an Hsp90 inhibitor, AUY922.
This
13
is published data.
14
targeted to be given once weekly.
15
from a pharmacokinetic standpoint is the drug
16
actually clears from the body within a day, but
17
from our preclinical data, we knew that the drug
18
remained in the tumor for over a week.
19
able to give this on a once-weekly basis.
It was an IV compound that was Key thing here
So we were
We had preclinical data from the rat and the
20 21
dog.
Albeit that they looked close, there's
22
actually quite a difference between 20 and 28 when
A Matter of Record (301) 890-4188
257
1
you consider the values, so preclinical uncertainty
2
in terms of where we should be in the dose range
3
and the exposure.
4
We also had -- to the earlier points about
5
the observation of a single event in a single
6
animal -- a QT signal in one dog that led us to put
7
the drug internally on hold.
8
analyzing sinode [ph] data, and we actually ended
9
up with exposures far higher than this where we
We had a year of
10
were able to show it was safe, but we still started
11
at 2 milligrams.
12
We centered our knowledge around these two
13
species.
14
start to escalate 2 to 4 to 7 so forth.
15
wanted to do was to be able to accelerate through,
16
because our interpretation of the sinode data was
17
to go much higher.
18
If you use a modified Fibonacci, you What we
But if we were to see any grade 2 events in
19
two or more patients that were treatment related,
20
then we would start to reduce that, at least
21
protecting us, even if DLT could drive us forwards.
22
So we saw no DLTs up to 16 milligrams.
A Matter of Record (301) 890-4188
We
258
1
escalated to 22 milligrams, and we saw one DLT in 4
2
patients expanded, and 5 patients with no DLTs.
3
Escalated to 28 and everything was fine.
4
to 40.
5
milligrams and 2 DLTs.
6
mindset, we've exceeded MTD, 2 DLTs, we should be
7
declaring 28 as our recommended dose.
We moved
We actually ended up with 7 patients at 40 So kind of traditional
8
What we actually saw was that the BRM when
9
analyzed the full set of data actually would allow
10
you to stay at 14 and we assessed some lower doses
11
again with expansion at 20 and 28.
12
going back to 40 milligrams, we saw no additional
13
DLT in another 6 patients.
14
out of 13 patients.
15
safety profile, we actually consider we can move
16
forwards.
17
And then when
So we suddenly had 2
And based on the overall
So this is the final study data.
We ended
18
up at 70 milligrams.
We had 3 DLTs in 24 patients.
19
So about 12 percent DLT rate.
20
patients were recruited actually at 40 and
21
54 milligrams as well, and this allowed us not just
22
to assess efficacy across these dose ranges, but we
A Matter of Record (301) 890-4188
The additional
259
1
were able to start to assess some on-target
2
toxicity relationships with ocular events, with
3
diarrhea. But at the end of the expansion, we actually
4 5
determined 70 milligrams to be the recommended
6
dose.
7
threshold.
8
the dose that the MTD would have determined under a
9
3 plus 3, and we subsequently showed it was both
So this was supported from an overdose It was two-and-a-half times higher than
10
safe, but it also had some activity in patients in
11
lung cancer in multiple different mutation types.
12
So clearly, we ended up at a dose level that was
13
safe, more effective, than a more traditional
14
design. In terms of a proof of concept, we actually
15 16
used biomarker data.
So for the target biomarker,
17
it was Hsp90, we don't have an assay for but we
18
knew that was directly related to Hsp70
19
upregulation. We had down regulation of HER2 in patient
20 21
scans.
We had partial response from the CT.
22
had PET responses.
We
And we saw a dose response in
A Matter of Record (301) 890-4188
260
1
terms of clinical activity.
2
used to determine that 70 milligrams was the active
3
dose or the recommended dose, even though we could
4
have gone higher based on DLTs alone. So clearly, the key is to use all the
5 6
available information.
7
forwards to the summary.
I'm going to just skip
Summary is you need to have statisticians
8 9
So all of the data was
with strong communication skills and everyone else
10
with open-mindedness to those skills.
11
your questions.
12
that you can adapt within a clinical trial and it's
13
almost impossible to adapt them -- it's definitely
14
impossible to adapt them all at once, but it's
15
almost impossible to consider them all upfront.
As I said, there's so many things
You should be very clear.
16
Prioritize
Not everything
17
needs to be or can be answered in a single trial.
18
So if we go back to the recommended dose, exposure
19
response relationships, yes, we want to be able to
20
use this in order to be able to analyze, but it may
21
require us to assess this outside the trial as
22
well.
A Matter of Record (301) 890-4188
261
General PK/PD modeling and relationship back
1 2
to our preclinical data, certain signal detection,
3
hepatic impairment, PK, QTC, food effect,
4
formulation, can you do these whilst you're ongoing
5
with the trial that then come back into the trial
6
to impact where you move forwards. The case of ceritinib would have been one
7 8
where I think everybody would have liked to have
9
made more use of the food effect data that we had,
10
and Dan Howard here will talk more about this
11
tomorrow about we ended up with decisions.
12
the ability to understand subpopulations, and then
13
other things:
14
antibodies, that's a very important thing.
cost of goods.
Again,
If you're working in
So be very explicit about the decisions that
15 16
you can take in a trial and the ones that you
17
can't.
18
regulators understand.
19
question, it's not necessarily that they're not
20
agreeing with the trial.
21
want to understand how you intend to do this, and
22
we see this a lot with Japan as well.
And as a company, be willing to help the When a regulator asks a
It may just be that they
A Matter of Record (301) 890-4188
262
Most critical of all is to ensure that you
1 2
have the data.
If you don't have data, you're back
3
to finding an MTD.
4
time.
So thank you very much for your
5
(Applause.)
6
DR. RUBIN:
7
The next speaker is Dr. Jose Pinheiro from
8
Janssen, and he's head of the statistical modeling
9
department there, and he'll also be speaking on
10 11 12
Thanks, Stuart.
best practices of adaptive dose-finding studies. Presentation – Jose Pinheiro DR. PINHEIRO:
Well, I'm going to go before
13
because you're going to notice that I have the same
14
little girl here as Stuart has in his slides.
15
That's coincidental, but not entirely.
16
because this is joint work that I have done with
17
two former colleagues from Novartis, from the time
18
that I was there.
19
Bjorn Bornkamp.
20
little girl from Novartis.
21
(Laughter.)
22
DR. PINHEIRO:
It's
So that's with Frank Bretz and So that's I guess it's the default
So the style of the slides
A Matter of Record (301) 890-4188
263
1
were slides that were used at the time that we had
2
to present that I'm leveraging a lot of slides here
3
from.
4
So I'd like to thank the organizers for the
5
opportunity to be here today.
6
workshop, very informative, at least until now.
7
I hope not to screw it up too much.
8
(Laughter.)
9
DR. PINHEIRO:
It's been a great So
So I do have the disclaimer,
10
so I guess statisticians and pharmacometricians in
11
this session here are -- so I'm an employee of
12
Janssen, and I do still have stocks from Novartis,
13
from the time that I was there.
14
I'm going to be changing somewhat the type
15
of dose finding that we have been talking thus far.
16
I'm going to be talking about dose finding in the
17
context of selection for confirmatory studies, and
18
not about the traditional oncology way of having
19
one dose sometimes even without an active control
20
going in to it, but rather, I'm going to be talking
21
about, let's say, dose finding outside the oncology
22
paradigm.
And I think probably that's what I was
A Matter of Record (301) 890-4188
264
1 2
invited to come here today. A lot of my work has been in that area.
3
Let's say dose-finding methods, model-based methods
4
for dose finding and so forth.
5
intention is that perhaps, as we saw in many
6
presentations leading to this one, maybe we should
7
not only be concerned about getting the MTD in
8
oncology studies.
9
sense as a paradigm for cytotoxic drugs, even
10 11
And I think the
Perhaps that makes a lot of
there, not entirely sure. But definitely for non-cytotoxic or for
12
cases of personalized medicine, perhaps an
13
understanding of the dose-response relationship or
14
even better, the dose-exposure response
15
relationship, is a way to go.
16
In that kind of a framework, you don't
17
necessarily want to go to the maximum dose that you
18
can get before you reach a certain amount of
19
toxicity.
20
efficacy side, and maybe I can go with a much
21
smaller dose and have about the same efficacy.
22
why submit patients from much higher than necessary
But maybe you'll say, let me look at the
A Matter of Record (301) 890-4188
So
265
1
dose? So that's sort of the intention of my
2 3
presentation.
I'm going to be talking about a way
4
of doing dose selection that perhaps can be used in
5
the future or more frequently in the context of
6
oncology studies.
7
talk about what are the issues that we face in
8
these other kind of applications, because it's not
9
that outside of oncology everything is great within
And in the process, I'll also
10
dose finding, the best way that's possible, and
11
everything works perfectly; far from there. There are issues and issues that perhaps
12 13
would be of concern in the context of oncology
14
phase 2 studies if this path is gone, and maybe
15
that's a trap that can be avoided. So I'm going to be talking about the
16 17
motivation.
I'm going to be talking about this
18
issue between pairwise comparisons versus modeling,
19
which is going to become clear when I talk about
20
it.
21
dose, phase 2 study, so that's not an issue.
22
if you have multiple doses, yeah, it does come in.
It doesn't have to be the context of a single
A Matter of Record (301) 890-4188
But
266
I'll mention one specific matter, this
1 2
MCP-Mod approach that was developed together with
3
my former colleagues from Novartis.
4
it's going to be very brief on that one. All right.
5
But again,
So in the context of dose
6
finding, like more broadly across the different
7
therapeutic areas and across the industry, and with
8
regulatory agencies as well, there's an
9
understanding that we also do improper dose
10
selection because of improper understanding of
11
dose-response relationships, both for efficacy and
12
safety. We just don't know.
13
So people just try to
14
go very quickly -- as is the case also in oncology;
15
that's no different -- and move fast -- we don't
16
understand exactly how the relationship between
17
dose in efficacy and safety occurs. So there are major consequences to that that
18 19
increase the costs of development by a lot.
20
Oftentimes, you have to repeat a phase 2 study
21
because it wasn't conclusive the first one that you
22
got.
It may delay regulatory approval or even deny
A Matter of Record (301) 890-4188
267
1
it altogether.
You just got the wrong dose.
Many drugs have been investigated that
2 3
there's a strong suspicion that, hmm, maybe if we
4
had gone with a smaller dose, who knows, that
5
safety outcome would not have been seen, and the
6
efficacy would be about the same. Many cases of post-approval changes in
7 8
dose -- I mean that's rather common I guess -- in
9
the context of oncology as well, and definitely
10
throughout other therapeutic areas, often lead to
11
reductions, which from an interesting point of view
12
has consequences on revenue. So the price of the drugs has to do with the
13 14
dose.
15
like major consequences on the bottom line.
16
reason probably to consider carefully from the
17
decision-makers in industry.
18
So reducing it after approval definitely has A
Some common reasons for that is that more
19
often than not, people tend to think about phase 2
20
studies as if they were a mini phase 3 study.
21
they try to say that, okay, if I can do this study
22
here, looking like I would, a phase 3, and I'm
A Matter of Record (301) 890-4188
So
268
1
successful, I can perhaps use this one as one of
2
the pivotal studies and just have a second one that
3
I can complement that, and I can save two years in
4
development or even more. So that's very, very tempting, but
5 6
unfortunately doesn't tend to work very well,
7
there's a probability of that working or not.
8
again, the regulatory experience, the feedback that
9
we get from people who see that all the time is
10
that it's a very risky proposition.
11
industry point of view, it's a very tempting
12
proposition.
13
hopefully at the panel discussion as well.
But
But from an
So more on that later on, and
As a consequence of this view that you're
14 15
trying to go with something that, okay, let me make
16
this just with a kind of grade of control of type 1
17
error rate that I would have in a phase 3, is that
18
you tend to go with fewer doses in a smaller dose
19
range.
20
I'll look at those two doses, see what
21
happens, if I can show that one of them beats
22
placebo or beats the active control in a
A Matter of Record (301) 890-4188
269
1
statistically significant way, maybe that's going
2
to be after FDA, after CMA, and everything is going
3
to be great.
4
unfortunately, as I said.
5
And it doesn't work that way,
This has been recognized by regulators, by
6
people in the industry, by a lot of people.
Just
7
like last December, there was a workshop that some
8
of us here in the audience attended at EMA that was
9
exactly focused on that, but in the context of more
10
traditional, let's say, dose selection, not just
11
specific on the oncologists, as this one is being,
12
but rather, okay, what are the issues?
13
try to address that?
14
emphasis in trying to push forward an idea of
15
model-based understanding of dose-exposure response
16
and of benefit-risk understanding as well.
17
that's a topic for a different discussion.
18
How can we
And there was a strong
But
In the context of phase 2 studies, typically
19
we want to address two things.
20
proof-of-concept that you just want to say is there
21
any relationship between dose and response?
22
just going to say, okay, if I change the dose here
A Matter of Record (301) 890-4188
One is like this
We're
270
1
within this range, what do I see with my response?
2
What's happening over there?
3
would be the selection of the dose.
4
establish that, then which dose should I bring
5
forward to my confirmatory stage?
And the second one So if I can
In order to do that, statisticians for the
6 7
most part have been responsible for two main
8
approaches that they -- well, actually one of the
9
main approaches with the pairwise comparison that
10
falls into this mini phase 3 view of the world.
11
So they say that, okay, here's what I'm
12
going to do.
I'm going to have like multiple
13
doses.
14
doses to placebo, and I want to see which ones I
15
can establish significant p-value.
16
if this is significantly better than placebo and so
17
forth.
I'm going to be comparing each of those
I want to see
18
Implicit in that kind of thinking is that
19
the smallest one that's statistically significant
20
better than placebo, is the one that I'm going to
21
take forward, hoping that you're going to reproduce
22
that result in phase 3, and that's it.
A Matter of Record (301) 890-4188
271
1
The modeling approach, on the other hand,
2
tries to actually understand the dose-response
3
relationship, and based on that, make a decision
4
that, okay, if my target is to have that level of
5
efficacy, or for that matter, as Stuart just
6
presented, that level of toxicity, this is the dose
7
that I should bring forward.
8 9
Just to illustrate what that is -- I agree with Stuart that a graph speaks a lot more than
10
just trying to put words in there -- these are the
11
two views of the same data.
12
side panel over there, you have this multiple
13
comparisons approach.
So on the left-hand
14
So suppose that I have 4 active doses, and
15
I'm showing you already the comparison to placebo.
16
So on the pairwise comparison approach, I'm looking
17
at confidence intervals, the difference to placebo,
18
and what I care about is if it's above zero,
19
meaning I'm statistically better than placebo in
20
that sense.
21 22
The notion of a clinical relevant outcome is typically not there at the design stage, but can be
A Matter of Record (301) 890-4188
272
1
brought in an ad hoc manner.
2
I'm saying is that, okay, suppose that the
3
horizontal red line targets that.
4
least having that effect for this to be clinically
5
meaningful.
6
not intended it should be anything real.
7
to illustrate the point.
8 9
So in this case, what
You should be at
Again, this is made up there.
It's It's just
In that case you'd be asking yourself, okay, D2 -- D1 definitely doesn't make it because it's
10
not statistically significant.
11
do, but which one should I choose?
12
The confidence interval includes the clinically
13
relevant threshold.
14
point estimate is above.
15
D2, D3, and D4 all I should go D2?
D3 also includes, but the
Should I go with D3?
Should I go with D2?
16
It's not an easy decision to be made.
17
in real life, people would bring in safety and
18
other considerations.
19
Of course,
But just to move to a different paradigm, on
20
the right-hand side, you have a modeling approach
21
of the same data.
22
taking the doses as unrelated, categorical
So in this case, instead of just
A Matter of Record (301) 890-4188
273
1
variables in there, you just say that, well there's
2
a continuum between them, and therefore I can
3
establish a model that tells me what is the
4
relationship; as I increase the dose, what happens
5
to my difference to placebo. In that framework, you can for example say
6 7
that, hmm, perhaps it's not D2 nor D3; it's
8
something in between that's the one that I would
9
like to get, that would be the minimum effective
10
dose, the smallest dose per dosing effect that
11
you'd like to have, which is estimated by that red
12
dot in the plot. A very important element of that different
13 14
mindset of doing things is that then you can talk
15
about a precision of that estimate; say, okay, my
16
guess is that that point in between the 2 and the 3
17
is the right one.
How confident am I about that?
You see in the horizontal, like in the
18 19
X-axis in there, that you have a fairly wide range
20
of potential doses that could have that same
21
effect.
22
about your estimate of the dose, which you cannot
So you bring the notion of uncertainty
A Matter of Record (301) 890-4188
274
1
do on the left-hand panel.
So that's just an
2
illustration of the two aspects of it. The way that I'm talking, of course, it
3 4
sounds like I'm favoring the model-based approach,
5
and I do, but I recognize that it's not that
6
simple.
7
the CRM.
8
guess of trials between a certain long period in
9
there, used the CRM.
We just had the statistic about the use of Laura was talking about 1.2 percent, I
The same thing -- not maybe as dramatic, but
10 11
by and large, people tend to use the pairwise
12
comparison approach in phase 2B much, much more
13
than a model-based approach is used.
14
model-based approach becomes like case studies in
15
presentations like this one, because they are rare
16
still.
Typically, a
17
But there are reasons for that, too.
18
not just people don't like them or something of the
19
sort, but they're hard in a way to do because,
20
oftentimes, not knowing beforehand what to expect
21
in terms of the dose response or the dose-exposure
22
response makes people feel kind of iffy, and say,
A Matter of Record (301) 890-4188
It's
275
1
well, okay, I have to write something in my
2
protocol.
3
What if that relationship is not the one that I
4
anticipate it to be?
5
going to be your primary analysis, typically you do
6
have to specify it beforehand.
7
I'm going to be sending this to FDA.
And mind you, if this is
So again, talking about history -- Stuart
8
talk about his PhD when he went to work in
9
industry.
When I went to work at Novartis, I came
10
with some modeling background.
11
statistician, I said, "Hey, I'm going to be really
12
making a difference here using modeling left and
13
right," and it wasn't that simple because people
14
would push back saying that, "Well, your modeling
15
ideas work very well, but we don't know what the
16
right model is, so let's stick to the pairwise
17
comparison approach."
18
And being a
So it took some time to see that, but how
19
can we try to get out of this situation?
20
of this MCP-Mod method that I'm going to be briefly
21
presenting shortly, has to do with that.
22
recognize that there is model uncertainty typically
A Matter of Record (301) 890-4188
The idea
It's to
276
1
when we come at this stage of development. Oftentimes because of what we base our
2 3
initial selection or understanding -- it's based on
4
nonclinical data, preclinical data.
5
biomarkers.
6
And it's oftentimes only at this point that we get
7
the really clinically meaningful endpoint, if we
8
are lucky, that we want to use to make a decision
9
to go to phase 3.
10 11
It's based on
It's based on a different population.
And people just say, what can be
done? So recognizing this model uncertainty was an
12
important element, at least for us, to be more
13
successful pushing this kind of an approach
14
forward.
15
combination of multiple comparison procedures and
16
modeling, is exactly an attempt at that.
17
the only one by any means; there are other methods
18
that use similar ideas.
19
presentation here is much with the intent of saying
20
the model-based methods are the way to go, and we
21
should be pushing them and using and becoming more
22
familiarized and comfortable with them moving
So this MCP-Mod, which stands for a
It's not
And I think my
A Matter of Record (301) 890-4188
277
1 2
forward. I will present to you very briefly the idea
3
of MCP-Mod, what it is.
4
big-stage kind of thing.
5
the trial design stage.
6
that, all right, I cannot avoid model uncertainty.
7
It's part of life.
8
That's the reality of it.
9
And it's like a two So that's the first part, The key idea is to say
I don't know what the model is.
I can try to either go with a very flexible
10
model that accommodates any kind of pattern that
11
people can throw at me.
12
that, you typically have to go with lots of
13
parameters in a model, and from a statistical or a
14
modeling point of view that may be a waste if your
15
relationship is a simple one.
16
But of course, if you do
So what the method proposes is this.
Well,
17
maybe we don't know exactly what the relationship
18
is, but it's not that we don't know anything.
19
Maybe we know some potential shapes that can
20
explain that.
21
thinking about a prior in Bayesian statistics.
22
Any flavor.
It's very similar to
So you may not know what your parameter
A Matter of Record (301) 890-4188
278
1
value is, which you don't of course, but you may
2
have an idea about the distribution of the
3
parameter, even before you look at the data.
4
course, if you know a lot about the distribution,
5
you may have an informative prior. If you know close to nothing about the
6 7
distribution, you're going to have a
8
non-informative prior.
9
you try somehow to characterize your knowledge
10
Of
But at the end of the day,
about the parameter through a distribution. In here, what we try to do is to
11 12
characterize our knowledge or ignorance about the
13
truth underlying dose-response shape through a set
14
of candidate models or candidate dose-response
15
shapes.
16
going to throw in and vice versa, making it small.
The less we know, the more shapes we're
17
So at the end of the day now, what we do is
18
that we replace the idea of pairwise comparisons to
19
placebo.
20
tests.
21
of the dose response.
22
trend for, let's say, a quadratic shape in the dose
Instead, we look at different trend We're going to look at trend for linearity We're going to look at a
A Matter of Record (301) 890-4188
279
1
response.
And each of the shapes that we choose is
2
going to have an associated test.
3
So the multiple comparison aspect of this
4
methodology has to do with the fact that you have
5
multiple trend tests, meaning that I'm going to be
6
trying to capture different signals of dose
7
response in my data.
8
the proof of concept part of the goal of this.
And that would account for
9
So if I can establish that there's a
10
significant dose-response relationship, even
11
without knowing which dose is significantly better,
12
that would tell me that, okay, there is something
13
going on here.
14
trying to see which dose or doses should I bring
15
forward.
16
Now it's worthwhile digging in and
So the second stage then would be exactly,
17
trying to bring modeling into the picture, and
18
through modeling, try to estimate what dose you
19
should bring forward.
20
The way that it would occur is that we would
21
design a study with that goal in mind, like
22
multiple models, multiple tests, take into account
A Matter of Record (301) 890-4188
280
1
the multiplicity that's inherent to that, and some
2
specification of how I'm going to be selecting the
3
dose based on modeling.
4
Once we have data, we now test for the
5
presence of dose response.
6
verified, we fit models to the data.
7
If that can be
We choose the best model or we use some
8
average of those models, which can also be done.
9
And we estimate a dose to bring forward together
10
with some notion of precision about that dose,
11
which is very important, because always, I would
12
say -- I haven't come across a case that's been
13
otherwise -- we're going to be designing a study
14
like this one, based on the principle of
15
identifying a dose response because the sample
16
sizes are feasible, 200, 300 patients, and we
17
evaluate what kind of precision do we get in
18
selecting the dose.
19
Even though our main goal is to choose a
20
dose or a couple of doses that we are confident
21
about moving toward, we typically cannot afford to
22
do that in the context of the traditional phase 2.
A Matter of Record (301) 890-4188
281
1
So MCP-1 is useful in my mind because it
2
does combine ideas that are from the two fields,
3
like the idea of taking into account testing for
4
establishing proof of concept, correcting for
5
multiplicity.
6
pay in terms of uncertainty.
7
in the same methodology, some idea about
8
dose-response estimation.
9
So the less you know, the more you But you also can use
So it's intended for phase 2, even though
10
there are some variations of it, and I'm not going
11
to have time to talk about it all, that could be
12
used in the context of a confirmatory study in the
13
sense of really controlling type-1 error rate in
14
the level needed to make it phase 3 able.
15
handle like a lot of types of events, but let me
16
just move forward and look at an example.
17
It can
Now, as I mentioned, typically in oncology,
18
we don't use multiple doses in phase 2.
19
Interesting, Stuart just mentioned about a study
20
that I was involved long time ago when I was still
21
there, and he was mentioning that it was an
22
oncology study in which we weren't thinking
A Matter of Record (301) 890-4188
282
1
about -- actually it was designed using this
2
MCP-Mod approach, but apparently it didn't go
3
through, for different reasons; not because of the
4
methodology.
5
But we don't have that many examples in
6
oncology.
We have a lot of other examples in other
7
areas, so I'm going to just present the key ideas,
8
show you something in the area of respiratory
9
diseases, because in there that was used.
So I
10
just want to go all the way through so that it
11
makes some of the concepts that I mentioned before
12
more applied than before.
13
This is a study in which to establish that
14
there was a dose response, to select a dose for the
15
confirmatory study and understand about the
16
dose-response relationship.
17
respiratory disease, and there were 4 doses being
18
used, from 0.25 up to 2, with a doubling each time.
19
It's in a chronic
The primary endpoint was a trough forced
20
expiratory volume in one second.
21
and see how much air you have in your lungs, two
22
weeks after being in the study.
A Matter of Record (301) 890-4188
So you just blow
And we want to
283
1
look at the 24 hour trough because that's intended
2
to be a once daily drug.
3
this case that we say by the end of the effect of
4
the drug, are you still okay?
5
So it was critical in
There was establishment of the minimum
6
effective dose should be one producing at least
7
120 mL improvement in FEV1 compared to placebo.
8
that's kind of the idea.
9
worse than an active control.
10
So
And it should not be
One of the interesting aspects of this
11
particular program is that there was an intention
12
of making that a global development, so we want to
13
do this together with Japan.
14
approval, and then go to Japan afterwards, but to
15
say that, okay, we're going to be including
16
Japanese patients in the study and what's going to
17
be enough for us show that the dose will be the
18
same or not the same.
19
So not wait, get
So there was an interest in there of
20
comparing dose-response relationships, and that's
21
actually what made this approach, MCP-Mod, we use
22
in this study.
A Matter of Record (301) 890-4188
284
1
So it was an agreement to use 25 percent of
2
Japanese patients in the study with the
3
understanding that at the end, we would be giving
4
estimates for the doses in the two populations,
5
mean Japanese and non-Japanese, but also looking at
6
confidence intervals to see how much overlap or
7
non-overlap there would be.
8 9
All right.
There were five candidate
models, so that was the level of uncertainty that
10
the clinical team, together with some previous
11
development that was done in this area, came to at
12
the end.
13
forth, a lot of iterations in terms of coming up to
14
those five.
15
Of course there was a lot of back and
But what you can see in there is those
16
shapes assume you're going to converge to a maximum
17
possible improvement.
18
is it quick or is it less quick?
19
shape in there, some kind of an S-shaped curve.
20
And as is often the case, people say that it might
21
be that for very high doses, I'm going to be seeing
22
the effect going down, and maybe there's no
But the difference is that
A Matter of Record (301) 890-4188
There's a linear
285
1
biological, no pharmacological sense to that. I have heard a lot of my PK/PD colleagues
2 3
say that that shouldn't be the case.
4
oftentimes talking to clinicians, they say that,
5
well, just to be on the safe side, let's include
6
that in there.
7
sense.
So it's very empirical, as you can
Three types of analysis were done in this
8 9
But
one.
There was one for the pooled.
And it's based
10
on the pooled that we selected what the
11
dose-response shape would be.
12
those Emax models that you saw, the one that
13
converts to an acetone; should it be cleaner?
14
So there was an agreement with health
Should it be one of
15
authorities in Japan that, okay, we're going to do
16
that based on the pool, but then we're going to
17
apply this model to the Japanese and the
18
non-Japanese.
19
evaluate the dose-response shape only for Japanese
20
because there would not be enough patients in
21
there.
22
data.
So we're not going to be able to
So the full MCP-Mod was used for the pooled
A Matter of Record (301) 890-4188
286
Results.
1
We got a very significant
2
dose-response signal, the usual p-value less than
3
10 to the minus 4.
4
significant.
5
dose-response relationship.
6
overpowered because there was a concern about also
7
being able to detect something for the Japanese
8
patients.
So they're very, very
No doubt that there was a The study actually was
So we knew going into this that if there was
9 10
a dose response of what we were expecting to be,
11
that establishing that significantly would not be a
12
problem at all.
13
Emax.
14
acetone [indiscernible], the one that you saw
15
before, so that kind of confirm their expectations.
16
The best fitting model was the
That's the one that converges to an
The estimated minimum effective dose, the
17
one producing 120 mL was 0.9.
18
one of the doses.
19
had some experience with it.
20
And 1 milligram was
So that was something; okay, we
Now, what about the precision?
As I
21
mentioned before, when we were discussing that with
22
regulators, they said that, okay, not only are we
A Matter of Record (301) 890-4188
287
1
going to be giving you a p-value in a test, but
2
we're also going to be able to give you an estimate
3
of this minimum effective dose, and also some idea
4
of precision around it.
5
interesting part came in about the modeling.
6
And that's where the more
The way that was done was using a method
7
called Bootstrap, which essentially what you do is
8
that you get your sample that you have and you
9
select patients with replacement formula as if
10
you're running the trial again.
11
let's say, 200 patients, and I say let me select
12
200 patients again, but I'm going to allow some
13
patients to be selected twice and again.
14
So we get our,
That gives you an idea that what happens, if
15
I were to apply, the whole methodology now, get a
16
new estimate of the MED, and I keep doing that over
17
and over and over; that would allow me a fair
18
assessment of the variability in that estimate and
19
also take into account model uncertainty, because
20
each time that I resample my data, I may end up
21
with a different model.
22
I'm taking to account all of the different sources
So there's a notion that
A Matter of Record (301) 890-4188
288
1
of variability in my response. So what we did is that we then applied
2 3
MCP-Mod to the entire data and kept going.
And at
4
the end, we got confidence intervals and also
5
estimates for the dose response with those bands
6
around it to talk about uncertainty. So what you see there is on the left-hand
7 8
side is the estimated dose response for the pooled
9
data.
And if you hard enough in there, you're
10
going to see that towards the highest dose, there's
11
a little bit of a banding down, which of course
12
would not be possible with a pure Emax model, but
13
sometimes a quadratic model was chosen.
14
throws some slight normal authenticity towards the
15
very end.
16
And that
You also see how accurate those estimates
17
are, so these are the bands that you see around it.
18
And then you see the Japanese dose response in the
19
middle and the non-Japanese on the right-hand side.
20
And you can see that the Japanese dose response is
21
likely moved to the left, meaning that lower doses
22
produce a higher response than compared to what we
A Matter of Record (301) 890-4188
289
1
see in the non-Japanese, which again is very much
2
aligned with what the PMJ thinks. They say that, well, the way that we select
3 4
doses is by cutting the American dose in half.
5
that's our dose selection, which is of course not
6
quite.
7
doses, which is very much consistent here.
8
this is kind of the depressing part of all of that,
9
which is the reality of it.
10
But there's this idea that they need lower But
So I mentioned that our p-value was like 10
11
to the minus 4.
12
of view, forget it.
13
when we come now to the precision of that dose.
14
So
So from a purely statistical point We know the dose response, but
That's what you see in there, the blue dots
15
are the median values of the estimates in the
16
Bootstrap.
17
resampled, and the blue dots is the one that I get
18
as my median for that, which they very much agree
19
with my initial estimate of 0.9 -- I have now
20
0.84 -- and our estimate for Japanese is 0.6, and
21
then for non-Japanese, is 0.9.
22
So I did like a thousand Bootstrap,
But what matters now is what you get as a
A Matter of Record (301) 890-4188
290
1
confidence interval for that, which is very wide.
2
So that highlights the fact that the way that we
3
design phase 2 studies, with multiple doses mind
4
you, so we have multiple doses in there, is aimed
5
entirely at detecting statistical significance of a
6
dose response, which basically comes from the fact
7
that we tend to look at those pairwise comparisons.
8
That's how we powered those studies.
9
If we were, however, mainly interested in
10
getting more precise estimates -- so if I were to
11
make those confidence intervals narrower than they
12
are, the sample sizes that would be needed would be
13
much, much larger than what we currently use.
14
maybe at the panel, we can discuss a little bit of
15
how can we get out of this situation.
16
So
Bottom line here is that there was a lot of
17
overlap, maybe not terribly like something to be
18
proud of, because they were very wide.
19
least it was not enough to cause too much concerns.
20
But the message here is that, yes, we don't get
21
enough precision, even under highly statistically
22
significant situations.
A Matter of Record (301) 890-4188
But at
291
What we can also do, and that was done in
1 2
this trial that's based on real data, is they show
3
account of safety as well.
4
how we do it.
5
left-hand side, it's the analysis for the efficacy
6
endpoint.
7
happened to the active control, which is that
8
horizontal line, and also what happened on the
9
safety, which is given by the right-hand side
In real life, that's
So what you have now is on the
And I'm also bringing in there what
10
panel.
So there was a safety concern, and that's
11
kind of the response that we got in it. Of course, on the safety side, we don't have
12 13
this night convergence to an acetone because that's
14
now how it happened.
15
tends to be exponential; it increases a lot with
16
dose.
If anything, sometimes it
17
But based on those considerations, they say
18
that with 1 milligram, we seem to be somewhat above
19
what the active control is in efficacy, but we are
20
definitely below in terms what would be a safety
21
concern.
22
forward.
So that's good enough for us to move
A Matter of Record (301) 890-4188
292
We meet the clinically relevant criteria of
1 2
having the 120.
We are slightly but about
3
equivalent to the active control, but better in
4
terms of safety than what we are.
5
was to go with the 1 milligram.
So the decision
So wrapping up, first of all, the idea of
6 7
dose selection should not be one of testing
8
hypothesis, which for the most part is intended
9
only to work as a phase 3 trial if we are very
10
lucky, and I hear that, until today, a lot through
11
my career.
12
really nice wouldn't it?
13
very rarely I would say, we are that lucky, so we
14
just have to go and really learn about what's going
15
on.
16
But if we are lucky, that would be But more often than not,
From what I understand from listening to
17
regulators and coming to meetings like this, is
18
that the days may be behind us when it was enough
19
to get a dose that had efficacy, and the safety
20
wasn't too bad.
21
for us to establishing what is a dose for which you
22
don't have enough efficacy or that at least you see
So more and more, there's a push
A Matter of Record (301) 890-4188
293
1
that your dose response is going down?
2
just you are entirely on the plateau of the dose
3
response.
4
discussed.
5
It's not
But again, it's something to be
But it should be seen as an estimation
6
problem.
So if you have a target, either it's
7
something to achieve a certain effect or to
8
achieve, let's say, 90 percent of the maximum
9
effect, like an 80/90, whatever it is the
10
criteria -- but there should be one -- it should be
11
a problem of estimation, not just of hypothesis
12
testing.
13
So model-based methods are considerably
14
better in that regard than hypothesis testing
15
methods because with a model base, you can actually
16
establish your target, estimate it, and give some
17
idea of a precision for that.
18
It goes without saying that to do proper
19
modeling, we should go with wider dose ranges than
20
we typically do with a larger number of doses.
21
we should be taking to account both efficacy and
22
safety when we do that.
A Matter of Record (301) 890-4188
And
294
With 60 seconds to spare, thank you very
1 2
much.
3
(Applause.)
4
DR. RUBIN:
5
So the panel speaker for this session will
Thanks, Jose.
6
be Dr. Amit Roy, who's head of pharmacometrics at
7
Bristol-Myers Squibb.
8
slight change from the title on the program.
9
be talking about dasatinib dose optimization.
10 11 12
And he's going to give a He'll
Thanks, Amit. Presentation – Amit Roy DR. ROY:
Thanks Eric, and thanks you very
13
much to the organization for inviting me to talk
14
about our experiences with dasatinib dose
15
optimization, and as my title says, from phase 1
16
through phase 3.
17
As you've heard from a number of the
18
previous speakers, the importance of dose finding
19
in phase 2 -- and I'll be talking a bit about dose
20
finding and dose optimization in phase 3.
21
like to make the case that in some cases, that
22
might be the most appropriate thing to do.
A Matter of Record (301) 890-4188
And I'd
295
1
So just a brief overview of dasatinib.
2
Dasatinib is a potent inhibitor of 5 oncogenic
3
tyrosine kinases; BCR-ABL being the one that's
4
probably most well-known given its indication for
5
chronic myelogenous leukemia.
6
kinase, c-KIT, PDGF beta receptor, and ephrin
7
receptor kinases.
8 9
The others are SRC
The indications currently that dasatinib is indicated for is newly diagnosed
10
Philadelphia-positive CML in chronic phase, as well
11
as adults with chronic, accelerated, and
12
blast-phase Philadelphia-positive CML, as well as
13
for Philadelphia-positive ALL.
14
So why did I go through all of this?
The
15
first indication that's mentioned over here was
16
actually approved later on.
17
are the initially approved indications for
18
dasatinib.
19
indication for dasatinib was with a dose of
20
70-milligram BID, which was subsequently changed to
21
a dose of 100-milligram QD.
22
The second two bullets
And actually, the initially approved
This is the story of why that happened and
A Matter of Record (301) 890-4188
296
1
why it may be the most appropriate thing to do in
2
this circumstance with the intent of sort of
3
balancing speed with rigor.
4
So the situation was that dasatinib
5
was -- the phase 1 study was initiated in CML back
6
in late 2003, and this is sort of coming on the
7
heels of the success of imatinib, which was
8
completely a paradigm shifting molecule that was
9
developed.
10
But still, despite the amazing effectiveness
11
of imatinib, there was resistance that developed in
12
many patients.
13
unmet medical need.
14
with dasatinib in CML, given that it also was an
15
inhibitor of BCR-ABL kinase.
And so there was still a higher So a phase 1 study was started
16
Then soon after the -- I'll talk a little
17
bit about the phase 1 study and subsequent steps.
18
But basically, the overall timeline for the
19
development was that subsequent to the phase 1
20
study, there was a series of unsuccess [sic] with
21
that study in showing efficacy against CML.
22
Several phase 2 studies were initiated on
A Matter of Record (301) 890-4188
297
1
the basis of which the drug was approved.
2
note at the bottom in red is the approvals, major
3
regulatory milestones that were met; on the top in
4
green are the key clinical studies that were in the
5
development program.
6
So I'll
It was basically a two-year time between the
7
initiation of the phase 1 study and the application
8
for regulatory approval.
9
was targeted at meeting an unmet medical need, get
10
to patients early, with admittedly a dose that was
11
suboptimal.
12
actually initiated, as you can see in the bottom
13
tree-green areas, a phase 3 study,
14
dose-optimization study, even before the drug was
15
approved.
16
So again, a program that
And recognizing that, the company
So coming back onto the phase 1 study, the
17
initial dose selection for the phase 1 study was
18
based on both toxicology data as well as the
19
preclinical PK/PD data.
20
concentration time profiles at two different doses
21
in a mouse model, along with the phospho BCR-ABL
22
inhibition shown on the left-hand side arrow bar.
And this is showing
A Matter of Record (301) 890-4188
298
1
So what you can see with the decreasing
2
concentrations, initially, you get a rapid decrease
3
in inhibition of BCR-ABL, and over time that sort
4
of goes back up again with the decreasing
5
concentrations.
6
parallel is the concentration.
7
There's a bit of a lag, but almost
So it was felt that we needed to have
8
inhibition of BCR-ABL over essentially the entire
9
dosing interval, and that's kind of been the
10
paradigm for target therapies, and one that we'll
11
get into a little bit later on.
12
In the phase 1 study, it was a multiple dose
13
escalation study in patients with chronic phase
14
CML, initially, which was later expanded to include
15
both accelerated phase CML, as well as blast-phase
16
CML, and Philadelphia-positive ALL patients who are
17
resistant or intolerant to imatinib.
18
The initial dose escalation part of the
19
study was a traditional 3 plus 3 design started
20
with 15 milligram QD dose given 5 days on and
21
2 days off.
22
chosen, again, based upon preclinical tox studies
And that particular schedule was
A Matter of Record (301) 890-4188
299
1
showing high levels of possible GI toxicity.
2
so a drug holiday was actually built into the
3
phase 1 study.
4
the way up to 240 milligram given once every day.
5
The study also allowed for within subject dose
6
escalation, after cycle 1 for lack of efficacy.
7
And
The plan dose escalations go all
Soon after the study was initiated and after
8
the first few patients came in and we had PK data
9
from the first few patients, it was noticed that
10
the half-life in patients and subjects was quite a
11
bit shorter than had been expected.
12
upon that, a BID schedule was incorporated,
13
because, remember, what we were aiming to do is
14
actually maintain the level of BCR-ABL inhibition
15
over the entire dosing period.
16
And so based
Then based upon again data from mainly the
17
more advanced patients, the accelerated phase CML
18
patients, it was noticed that with the 2-day drug
19
holiday, there was a bit of a rebound in the blast
20
counts in peripheral circulation, so we went to a
21
BID continuous dosing.
22
basically the idea being we were looking at the
And as you can see,
A Matter of Record (301) 890-4188
300
1
data and adjusting the study based upon what the
2
data was telling us. The efficacy endpoint that were assessed
3 4
were complete hematological response, major
5
hematological response, and major cytogenic
6
response.
7
from peripheral blood.
8
the major response also allows for a non-return of
9
thrombocytopenia because some of the adverse
And the first two are actually coming The main difference is that
10
events, the hematological adverse events, may be in
11
some cases because the drug is actually working.
12
And the cytogenic response is actually a bone
13
marrow biomarker assessment. So based upon the results of this study,
14 15
which was published in the New England Journal of
16
Medicine, a 70-milligram BID dose was chosen as
17
shown over here.
18
only for the chronic phase CML patient populations.
This is showing the key results
Initially, as shown, it was QD dosing.
19 20
There were only a few subjects actually dosed with
21
QD dosing before we actually moved to the BID
22
dosing.
But based upon the very encouraging
A Matter of Record (301) 890-4188
301
1
efficacy, it was decided to go ahead with the
2
70-milligram BID dose for the phase 2 studies.
3
There was no MT that was reached.
4
that was actually administered was 120-milligram
5
BID.
The highest dose
6
So actually on the basis of these
7
encouraging phase 1 results, in very short order,
8
there were five phase 2 studies that were
9
initiated, all with 70-milligram BID doses with
10
slightly different inpatient populations:
11
phase, accelerated phase, myeloid blast phases, as
12
well as lymphoid and blast phase, as well as
13
Philadelphia-positive ALL.
14
that was a randomized study against imatinib,
15
high-dose imatinib, because these are all studies
16
in which patients were either intolerant or
17
resistant to imatinib.
18
chronic
And there's one study
So as result of these studies, the efficacy
19
was actually demonstrated in all of the target
20
indications, and the drug was approved based upon
21
this, allowing the drug to be made available to
22
patients who did not have any other options, having
A Matter of Record (301) 890-4188
302
1
failed imatinib or were intolerant to imatinib.
2
However, there was a high percentage of
3
patients who did experience dose reductions or
4
interruptions.
5
that we selected was well below -- there was no MTD
6
that was identified; it was well below the maximum
7
administered dose of 120-milligram BID, there was
8
still, when we treated the patients for a long
9
period of time, dose interruptions.
Remember that even though the dose
In addition,
10
pleural effusions, which is a significant event,
11
adverse event, noted in an earlier speaker, was
12
also seen.
13
Around this time, there was very interesting
14
nonclinical evidence that the continuous inhibition
15
of BCR-ABL may not be necessary for efficacy, and
16
this was a very nice study done by Anita Shah at
17
University of San Francisco.
18
Essentially, what's being shown over here on
19
the left-hand side is the results of a 20-minute
20
incubation of cells.
21
sensitive cells, CML cells, K562 cells I believe,
22
to a 20 minute exposure to dasatinib for washout.
These are cells, the CML
A Matter of Record (301) 890-4188
303
1
And essentially what you see is that soon after the
2
washout, within one hour of the washout, the
3
inhibition of phospho-CrkL essentially goes away.
4
So the inhibition of BCR-ABL and -- phospho-CrkL in
5
this case, a downstream protein from BCR-ABL,
6
represents the BCR-ABL inhibition.
7
shortly after the drug is removed, the inhibition
8
of that particular enzyme goes down.
9
In fact,
However, very interesting, as shown on the
10
right-hand side -- I apologize for the small size
11
of the plot; I took this out from the paper that's
12
cited at the bottom in Cancer Cell -- at the very
13
high doses of 100 and 50 nanomolar, even a short
14
infusion, a short exposure of 20 minutes actually
15
gives you a sustained cell killing ability all the
16
way up to 2 days.
17
concentrations, you do need to have that sustained
18
continuous sort of inhibition.
However, as you go to lower
19
So the idea behind this was that -- what
20
came out of this was that you don't need to have
21
complete inhibition, but it's not quite a threshold
22
effect, because you can actually achieve good cell
A Matter of Record (301) 890-4188
304
1
killing with even low doses if you expose the cells
2
for a long period of time to low concentrations.
3
This sort of leads into the phase 3
4
dose-optimization study, and this was essentially a
5
dream study for pharmacometricians such as me,
6
because we're looking at two different dose levels,
7
100-milligram and 140-milligram daily doses, either
8
given as a once a day dose or a divided dose, BID
9
doses.
10
So this really allows you to tease out
11
alternative measures of exposure that is driving
12
safety and efficacy.
13
slide subsequently, where if you're giving a single
14
dose, you typically have a high peak concentration
15
for a linear PK drug and low trough concentration
16
for a linear PK drug, with the same daily dose.
17
If you're giving a divided dose, you're
And you'll see that in a
18
going to get much lower peak concentrations, the
19
same average exposure over the treatment duration,
20
but higher trough concentration.
21
you to kind of tease out what exposure measure
22
might be driving efficacy and then safety.
A Matter of Record (301) 890-4188
So this allows
305
There was another study that was indicated.
1 2
034 was a study in chronic phase CML, the study in
3
accelerated phase but we're only looking at 2 dose
4
because we knew we needed higher exposure in the
5
sicker patients, so we only looked at 140-milligram
6
QD and 70-milligram BID, which is the approved
7
dose.
8 9
These are some of the key results of the phase 3 dose optimization study.
We did collect PK
10
in this very large study.
11
study of approximately 140 patients per arm.
12
we were able to characterize the PK, and therefore
13
determine a summary measure of exposure for the
14
exposure response.
15
It was a 600 patient And
What was seen was that the 100-milligram QD
16
actually had slightly better or certainly
17
equivalent efficacy to the 70-milligram BID
18
previously approved dose.
19
safety profile was much better with the
20
100-milligram QD dose, and that actually led to the
21
changing of the approved dose to 100-milligram QD.
22
We performed exposure response analysis to
And on this base, the
A Matter of Record (301) 890-4188
306
1
really understand what was actually driving some of
2
the results that were seen.
3
actually able to tease out that for efficacy.
4
Shown here is a property of efficacy on the
5
right-hand panel, showing the property of efficacy
6
versus a weighted C-average exposure over the
7
dosing interval, because remember, we actually had
8
a number of dose interruptions that we needed to
9
take into account.
And again, we were
So we used a weighted C-average
10
exposure that took into account the dose
11
interruptions over the treatment duration, so this
12
is a property of achieving the major cytogenetic
13
response at 6 months. There were several factors that had an
14 15
effect.
16
exposures, you got a more higher property of
17
achieving a response, but also, imatinib resistant
18
subjects were less likely to respond than imatinib
19
intolerant subjects.
20
Certainly, you can see that with higher
Gender was not a factor.
The other very important factor that was
21
incorporated here was the duration of dose
22
interruption.
So intuitively, if you have a long
A Matter of Record (301) 890-4188
307
1
period of dose interruption over and beyond the
2
lower exposure overall, they're less likely you
3
might actually be able to respond.
4
factor as well that was identified.
5
Age was a
So in this particular analysis -- this was
6
done to support the submission -- we did identify
7
dose interruption as a factor, but it was
8
incorporated in this logistic regression model that
9
basically had a yes/no answer.
It didn't account for
10
the time frame and time domain of these dose
11
interruptions.
12
So intuitively, a dose interruption that
13
occurs very early in the treatment might have a
14
different effect than a dose interruption that
15
occurs right before you measure the primary
16
endpoint at six months, and we were not able to, at
17
that point in time, account for that.
18
So what we did, we actually went back, and
19
with the benefit of time did a more sophisticated
20
analysis that actually allowed a more sophisticated
21
time-to-event analysis, actually allowed us to take
22
into account the time of the response together with
A Matter of Record (301) 890-4188
308
1 2
the time of the actual dose interruption. So we took the actual dosing history into
3
account in developing this particular model, and we
4
used a interval censored model to actually account
5
for the fact that you actually only observe the
6
major cytogenetic response at prespecified time
7
points of 3 months, 6 months, and 9 and 12 months
8
in this case, as is shown here.
9
key results that we found were similar to what we
10
had found with the less sophisticated analysis of
11
logistic regression.
12
Fortunately, the
We also went and did exposure response
13
analysis for safety, and the key measure that we
14
were looking at was pleural effusion.
15
adverse event that really was different for
16
dasatinib than the other tyrosine kinase inhibitors
17
targeting CML, and there was a concern for the
18
company.
19
This was one
What we actually were able to find was that
20
for this particular endpoint -- and this was done
21
as a time-to-event analysis, again, to account for
22
the fact that the longer you have a follow-up for a
A Matter of Record (301) 890-4188
309
1
patient, the more likely -- if the risk is
2
constant, the more likely you are actually able to
3
see the event.
4
regression modeling approach, we used a time to
5
event modeling approach.
So rather than use a logistic
6
We were actually were able to identify that
7
it's trough concentration that's driving the safety
8
endpoint.
9
fell in very consistently with the clinical data
And so it made a lot of sense and really
10
that the 100-milligram QD dose was the best dose to
11
pick because that's what gives you a lower trough
12
concentration than with a twice daily BID dosing.
13
So in summary, the phase 1 study was
14
designed with an initial dosing regimen, based on
15
preclinical and toxicology data.
16
amended to include, later on, more advanced CML
17
patients, once activity was seen in chronic phase
18
CML patients, and within subject, dose escalation
19
was allowed.
The study was
20
Again, learning from the data as the study
21
was going on, the study was amended to incorporate
22
continuous BID dosing based on the presumption back
A Matter of Record (301) 890-4188
310
1
then that we needed to have continuous inhibition
2
of the target and based upon the observed PK data.
3
And the 2-day drug holiday was removed because it
4
was seen that the efficacy was not quite as well.
5
Based upon that, a 70-milligram BID dose was
6
chosen for the phase 2 studies.
7
we had noticed that -- and this was done to
8
initiate the phase 2 studies as rapidly as possible
9
to get the drug to patients as rapidly as possible.
10
But importantly,
But at the same time, we had noticed that
11
with a small cohort of subjects who did have QD, we
12
got pretty good efficacy.
13
2 -- phase 3 dose optimizations were actually
14
initiated to actually ensure that the dose
15
selection and dose optimization was actually robust
16
for dasatinib.
17
finally changed to 100-milligrams QD.
18
So two phase
And based upon that, the dose was
I just want to add a few additions.
So that
19
was a story about dasatinib.
And some of the
20
things that actually made this possible I think
21
should be considered in terms of a hematological
22
malignancy, particularly CML, which has a very
A Matter of Record (301) 890-4188
311
1
specific target.
And if you're seeing activity in
2
a phase 1 study, even with the endpoint such as
3
major cytogenetic -- a hematological response,
4
which is not also the registrational endpoint, the
5
correlation with the registrational endpoint of
6
major cytogenetic responses is actually quite well
7
established.
8
So that's one of the things that actually
9
gave the confidence to actually go ahead with the
10
phase 3 study and look at a much complete dose
11
optimization before the approval was obtained.
12
It's not always the case for oncology drugs.
13
When we're looking at, let's say, for
14
example -- Dinesh mentioned earlier on when we're
15
looking at a tumor endpoint, and particularly if
16
it's recessed -- we might talk about this more in
17
the panel discussion.
18
of -- the criteria were intended to really ensure
19
that the tumor change that we're seeing were
20
actually real.
21 22
But recessed was sort
The criteria, to my understanding, was never really established as a correlation to overall
A Matter of Record (301) 890-4188
312
1
survival, which is oftentimes the gold standard
2
endpoint in oncology.
3
measure of tumor response, in many cases it would
4
be preferable to doing that.
So looking at a continuous
Moreover, it may also be a durability of the
5 6
tumor response that could be important.
7
point being that the association between a tumor
8
response, particularly by Reese's criteria and
9
survival, for many tumor types is not quite
10
well-established as is the case in many
11
[indiscernible] malignancies.
But the
So the efficacy endpoints can be -- having a
12 13
different efficacy endpoint in a phase 1 study can
14
sometimes be a challenge to extrapolating that to a
15
phase 2 study. The other challenges that others have spoken
16 17
about is the design space in oncology can very
18
varied.
19
given on a continuous basis.
20
flexibility of having drug holidays as the need may
21
arise.
22
play with gives you more freedom, but also it's
For target therapies they're oftentimes But there is the
So having much more wider design space to
A Matter of Record (301) 890-4188
313
1
more challenging to find the most optimal dose.
2
The other aspects that have been spoken
3
about already is the time period of establishing
4
the DLT.
5
typical cycle 1, 21 days or 28 days, for
6
establishing DLT because oftentimes the toxicities
7
for these targeted therapies with continuous dosing
8
occurs much later.
9
increases combinatorially as we introduce
10 11
We need to, I think, move away from the
And of course, the design space
combination therapies into this picture. The other thoughts are, oftentimes we have
12
limited data at low doses.
13
though you can get a very strong signal for the
14
activity of a drug from a phase 2 study, the
15
certainty of the dose response or what the optimal
16
dose is from that phase 2 study, unfortunately, it
17
can be very difficult to really establish.
18
same cases, it may be preferable to actually do a
19
dose-optimization study in phase 3.
20
Jose talked about even
So in
Finally, one needs to account, when we're
21
doing exposure-response analysis for the
22
variability in exposure due to dose reduction, due
A Matter of Record (301) 890-4188
314
1
to adverse events that may occur over the treatment
2
period. With that, I'll end my talk.
3
And I'd like
4
to actually acknowledge Shelly Wang who did all the
5
analyses for exposure response for dasatinib.
6
Shown on the later slides, Neelima Thanneer in our
7
programming group, and Tai who was the lead
8
statistician at the time. (Applause.)
9
DR. RUBIN:
10
So that concludes the speaker
11
presentations for this session.
12
10-minute break I think.
13
bit behind now, so if people could come back at a
14
quarter 'til, we'll have the panel discussion at
15
that time.
We're running a little
Thank you.
(Whereupon, at 3:36 p.m., a recess was
16 17
So we'll have a
taken.)
18
Panel Discussion
19
Lei Nie and Eric Rubin
20
DR. RUBIN:
I think we'll go ahead and get
21
started as people are joining.
22
Rubin from Merck, and I'll be helping to co-lead
A Matter of Record (301) 890-4188
So again, I'm Eric
315
1
this session with Dr. Lei Nie from the FDA. I want to note that we have all the speakers
2 3
from the last session who were up here as part of
4
the panel, so I won't reintroduce them, but we have
5
three folks who will be joining us on the panel
6
that I did want to introduce. So the first is Dr. Donna Dambach, who's
7 8
actually joining us from the earlier session.
9
We've had two people who are not able to make it,
10
Vivek Kadambi from Blueprint Medicines and Sherry
11
Ralston from AbbVie.
12
she's kindly agreed to come help join the panel to
13
help replace those two folks.
So we've asked Donna, and
Then we also have statisticians from the
14 15
FDA.
We have Dr. Vikram Sinha, who's a
16
statistician from the FDA.
17
DR. DAMBACH:
18
DR. RUBIN:
19
(Laughter.)
20
DR. RUBIN:
21 22
He's not a statistician. Oh, I'm sorry.
I love the
statisticians -- pharmacologist from the FDA DR. SINHA:
I think being a statistician is
A Matter of Record (301) 890-4188
316
1
a very close-held group and tough to qualify for
2
it.
3
(Laughter.)
4
DR. SINHA:
5
group.
We'll just be an exceptional
Thank you.
6
DR. RUBIN:
And Dr. Rajeshwari Sridhara who
7
I know is a statistician.
8
(Laughter.)
9
DR. RUBIN:
She's actually head of the
10
oncology statistical department at the FDA.
11
thank them for joining us.
12
So we
Dr. Nie and I did come up with some
13
questions that we thought we'd start off the
14
discussion with.
15
era of breakthrough therapies, how can we resolve
16
the tension between speeding up drug development
17
and taking time in dose finding?
18
improve overall efficiencies in dose finding
19
without slowing down overall development?
20
So the first question is, in the
And can we
I believe we might have had a subheading, if
21
we can advance one more slide.
22
2 and 3, which you're going to take, right?
A Matter of Record (301) 890-4188
These are questions
317
1
So why don't we back up and let's have some
2
discussion around question 1, and maybe I'll start
3
by asking the panel, anyone who feels so inclined
4
to take a crack at this first study.
5
we also welcome questions or participation from the
6
audience.
7
DR. DE ALWIS:
And of course
So in the era of
8
breakthrough, I guess if you have excellent drugs,
9
and there's a significant difference from standard
10
of care, hopefully one of the things in terms of
11
this tension might be addressed by the fact that
12
recruitment won't be a problem.
13
So that's one of the issues that oncology
14
development has had in the past is with these
15
early-phase trials, recruitment can be slow.
16
that can be a limitation when you exploring dose
17
response.
18
So
But if the drug looks very good, then
19
hopefully we can actually explore different doses,
20
and the recruitment is good.
21
fill this up and do this quickly.
22
that addresses the question, Eric.
So you'll be able to
A Matter of Record (301) 890-4188
I don't know if
318
1
DR. BAILEY:
Maybe I can add something.
I
2
think one of the challenges of -- Laura presented
3
it on the case study for ceritinib, LDK and the ALK
4
inhibitor.
5
2013 was designated as a breakthrough therapy while
6
we were still trying to really understand the
7
compound in terms of the level of activity, the
8
exact dose that we needed.
9
This was a compound that in March of
I think one of the big challenges is that
10
upon of designation of breakthrough from the
11
company perspective, is that you are now in the
12
confirmatory stage.
13
you are full speed trying to get this drug into the
14
market to allow as many patients as possible to
15
benefit from this therapy.
16
sense, challenge the ability, or the concept that
17
we are trying to push across, that we still need to
18
take that time to be able to understand.
19
You are now at a point where
And that can, in some
It's very easy to say, with the limited
20
information we've got, we're going to pick a dose
21
and going to run.
22
and submit the information, but then starts some
Yes, we're going to look to try
A Matter of Record (301) 890-4188
319
1
confirmatory trials at the dose that we've
2
submitted on this very limited information.
3
the reality is that we should not be afraid to
4
continue to explore.
5
study for dasatinib with the changing doses for
6
different populations, but this is still within the
7
same population.
And I presented the case
So I think from a regulatory perspective as
8 9
And
well, I think there's a need for guidance or for a
10
push from the regulators as well to not just, let's
11
say, sit back on your haunches.
12
that's not something that has been done.
13
definitely postmarketing commitments that have been
14
made.
15
tomorrow about dose optimization of ceritinib.
16
I think it is something that companies also have to
17
bear a stronger role in pushing that.
18
And I can tell you There's
And I mentioned Dan Howard will talk
DR. PINHEIRO:
Maybe I can comment.
But
Part of
19
the issue is that people in industry, they tend to
20
plan development programs.
21
quickest path to market.
22
operations, blah, blah, blah, what is quickest way
They think about the Okay, these are the
A Matter of Record (301) 890-4188
320
1
to get this drug approved?
But they don't
2
typically assign probabilities of success to
3
different paths.
4
maybe the chance of succeeding is 1 in a 1000, so
5
maybe that's not the best.
So well, this is the fastest, but
But that was an issue that was discussed at
6 7
the last December workshop.
8
well.
9
talk about that, but the reality is it's very, very
That issue came up.
Dr. Sinha was there as So okay, it's easy to
10
hard to break that paradigm because the temptation
11
is just too much.
12
in development, the potential is huge.
13
If you can cut two, three years
So one of the ideas that came up I think is
14
a request from industry to be considered -- but I
15
don't think regulators were completely opposed to
16
it at the beginning -- was the idea that would it
17
be possible to actually make a proper dose-finding
18
study, one that's done as it should be, let's say,
19
with a request to proper characterize dose
20
response, efficacy, safety and all that, be
21
considered as one of two pivotal studies.
22
So not just have as a whole, but say, if you
A Matter of Record (301) 890-4188
321
1
do it this way, but not intend it should look like
2
a phase 3, but do what you have to do, explore and
3
present your case.
4
show efficacy, to be a demonstration of efficacy,
5
and then companies would have to have a second
6
study in which they would confirm one or two doses
7
that came out of that study.
8 9
Then that would be enough to
There was no agreement and that was an idea, a lot of discussion, a lot of interest seen back
10
and forth.
11
the lines of what we have here.
12
But I wonder if that would be go along
So there's a conundrum in there that it
13
would be nice if people did what's right, and do
14
the scientific approach of dose finding,
15
exploration, and all that, but the commercial
16
appeal of not doing that may be too much to break
17
just by itself.
18
So I wonder if given some regulatory
19
incentive to that would work, and if that can be
20
done remains to be seen.
21
want put for discussion and see what people think;
22
not my idea, by the way.
But I don't know.
A Matter of Record (301) 890-4188
I just
322
1
DR. RUBIN:
Could I also ask the panel
2
members please get up very close to the microphone
3
because it's hard to hear, even I can tell you from
4
down here, what you guys are saying down there.
5
DR. SRIDHARA:
So I just want to say you
6
know breakthrough therapy is a designation rather
7
than that it's a given approval.
8
there is a lot of uncertainty around it, but it's
9
certainly something that's extraordinary that we
And we know that
10
are seeing as a response or activity more than
11
anything else, that it deserves a breakthrough
12
therapy designation.
13
So certainly at the agency, we don't see
14
this as this is the exact dose that should be given
15
or what have you, but we are just saying that here
16
is -- the activity of the drug is what's calling it
17
a breakthrough therapy for now.
18
to come up with more data to say, okay, this is
19
ready for approval.
20
But then you have
But certainly it does bring in issues with
21
how are you going to conduct future studies?
22
the designation allows you to come and talk to us
A Matter of Record (301) 890-4188
And
323
1
also in advance or multiple times in designing your
2
trials.
3
that's available and see how we can make the design
4
better.
5
the designs of such trials.
And I think we should look at all the data
Certainly, we are open to discussing about
Audience Q&A
6
DR. RUBIN:
7 8
to this?
9
panel.
10
Mark, is yours going to relate
Mark, go, and we can continue on the
DR. RATAIN:
Mark Ratain, University of
11
Chicago.
12
tension here, which is the commercial tension.
13
tension of time to market, but the tension of
14
pricing per milligram.
15
I think there's one other piece to this Not
So the conundrum is hypothetically let's say
16
you do a randomized, dose ranging, phase 2 trial.
17
You show clear dose response.
18
accelerated approval and says you need -- but go
19
study two of these doses in a confirmatory trial
20
with some other control group as a definitive
21
trial, and there's the whole conundrum around
22
pricing.
A Matter of Record (301) 890-4188
FDA grants you
324
1
So I think that that's another piece to
2
this, that in our current pricing models, where
3
prices are set per milligram or per tablet or per
4
capsule, as opposed to per value of -- per life
5
year saved or whatever, that we're a little bit in
6
this conundrum. So I think that we almost need to think
7 8
really out of the box and blue sky this, and then
9
maybe come back down to reality at some point.
But
10
I think that that's a commercial tension that I
11
personally don't the answer to under current price
12
models.
13
DR. RUBIN:
I do think it's a good point,
14
although it also argues that on the R&D side, you
15
might want to be more concerned about studying
16
earlier doses lower so that you're not finding out
17
a lower dose is just as good as a high dose when
18
you've already set price based on the older
19
paradigms.
20
DR. RATAIN:
Well, that's right.
I mean,
21
the dasatinib example where you raise the dose
22
after approval doesn't have any commercial harm.
A Matter of Record (301) 890-4188
325
1
In fact, commercial people will say, oh good, we'll
2
sell more drug.
Well maybe yes, maybe no.
But I think we have to keep all this in
3 4
mind.
5
implications for some of the drugs for which
6
there's postmarketing requirements to study lower
7
doses are very profound.
8 9
And I think that potential commercial
I guess we've been through this before, not in oncology, but certainly with AZT, where there
10
was a marked decrease in the labeling after
11
approval, although prices and what's at stake is
12
nowhere near the revenue stream that we have today
13
with any of these drugs.
14
But I do think we have to consider this
15
whole issue.
And it's all the more reason to get
16
the dose right before approval, rather than after
17
approval.
18
approved at a dose in order of magnitude greater
19
than the non-inferior lower dose.
I mean, the last thing you want is to be
20
DR. RUBIN:
Thanks.
21
DR. SINHA:
Just to add and maybe respond to
22
what Dr. Pinheiro mentioned.
Remarks from the panel?
So indeed, there was
A Matter of Record (301) 890-4188
326
1
this meeting last December that the EMA hosted
2
around dose response.
3
focus was not oncology, to be specific.
4
was a more general discussion around dose response
5
and what we are actually experiencing.
6
Now, at that meeting the I think it
Actually, from the FDA, we had actually done
7
an analysis, and Yaning Wang would have actually
8
presented this when he had summarized, by
9
therapeutic area, what were we typically seeing in
10
the number of doses that were studied in phase 3,
11
or leading up to phase 3, how many doses were being
12
evaluated.
13
listened to it, and you probably know the breakup
14
of that.
And many of you were in the audience or
15
So what I think is interesting here is that
16
oncology has, over the years, been somewhat special
17
in the way it's assessed dose response.
18
therapeutic areas have taken the more classical
19
pharmacology approach to assessing dose response,
20
which oncology, largely from the cytotoxic, as
21
Dinesh mentioned, rightly so, drove much of its
22
decision-making around dosing.
A Matter of Record (301) 890-4188
Other
327
1
So even though we say breakthrough
2
designations, the information that is actually
3
required to tell us about the pharmacology is still
4
the same.
5
decisions.
6
I mean, you still need to make good
All through the day, I think the common
7
themes have been that if we are investing a little
8
more in the preclinical to clinical translation, if
9
we are doing a little bit more as we transition
10
from phase 1 to phase 2, and a little more as we do
11
from phase 2 to phase 3, even by addition of one
12
dose arm, we are substantially improving our
13
ability to understand dose response.
14
So I actually think the tension is a good
15
tension, in the sense I think it is compelling
16
sponsors to integrate their information much
17
earlier, more efficiently.
18
breakthrough designation is doing.
19
faster, but do it more skillfully, and you don't
20
necessarily have all the time to spend in phase 2
21
and in large phase 3 studies.
22
That's what I think the It's saying
So I actually don't necessarily think that
A Matter of Record (301) 890-4188
328
1
these things are hugely at odds.
2
compelling us to move in a slightly different
3
direction more quickly.
4
DR. RUBIN:
I think it's
I noticed some of the speakers
5
actually did some literature reviews.
I think
6
Dinesh, he did this.
7
movement towards these types of approaches where
8
we're doing a little more, spending a little more
9
time either in the transition or in the actual
Are we seeing any trends of
10
clinical space towards, I think again, randomized
11
dose-finding studies in oncology, which are
12
exceedingly rare?
13
I'm just wondering if that's being done more
14
now or we do have to really start from now trying
15
to make this happen.
16
DR. DE ALWIS:
So reviewing the literature,
17
there are some examples that stood out, and I think
18
they started to occur actually from around 2005,
19
2006 onwards.
20
more of the examples from 2005 onwards, rather than
21
sort of the earlier part.
22
So yes, to your point, we're seeing
Actually, I just want to touch on Vikram's
A Matter of Record (301) 890-4188
329
1
point around using all information, and I think
2
probably goes towards the question 2 in fact.
3
I think this is something that we need to more of.
But
When we look at experiments that we get from
4 5
the preclinical setting, there has been -- and I
6
think we've started to see some change, but there
7
has been, from biology, to really show efficacious
8
experiments. They love to show wonderful survival data or
9 10
impact tumor growth reduction in animals.
11
do this with doses which are very high.
12
actually when you translate these doses, they
13
aren't doses that are going to be able to be
14
tested, in fact, in many cases, because they're too
15
high.
16
And they And
So one of the things that we need to do more
17
concertedly is really to do proper dose-ranging
18
studies, investigating especially the minimal
19
effective dose.
20
on don't like to do this because, unfortunately, in
21
company settings, it doesn't get your candidate
22
selected in many cases.
Pharmacology and biologists early
A Matter of Record (301) 890-4188
330
But this is really important information
1 2
because at these sort of doses is where you really
3
understand what is driving the efficacy.
4
time on target?
5
And based on these kinds of information, you can
6
then start to feed in,
7
dependencies and other factors.
Is it AUC-driven?
to understanding schedule
So I think, Eric, to your question, I think
8 9
Is it Cmax?
Is it a
we're beginning to see this and I think we're
10
beginning to understand that back and forth is
11
happening more.
12
setting, and then it's sort of feeding back into
13
the preclinical setting to do more refined
14
experiments based on what we're seeing.
So you're going to the phase 1
DR. SRIDHARA:
15
So, yes.
Another thing is I think with
16
breakthrough therapy, actually you may have more
17
data.
18
very quickly once the designation is given.
19
probably you learn about safety particularly faster
20
than the other products, so you may have more
21
information then.
22
It seems like the more diseases are studied
DR. ROY:
Yes.
I just want to make a
A Matter of Record (301) 890-4188
So
331
1
comment about -- certainly, I think we'd all agree
2
that it's advisable to explore multiple doses,
3
certainly more than one dose in early-phase studies
4
to inform the dose taken to pivotal studies.
5
But I would argue that every case has to be
6
looked at for its own attributes.
So certainly,
7
for something that has breakthrough therapy,
8
meaning you have clinical data that's showing high
9
levels of activity, enrollment may not be an issue
10
as Dinesh said.
11
look at patients in more than one dose level
12
because the enrollment may go fairly quickly, and
13
you may have the answer fairly soon.
14
So it may be more acceptable to
If it's an orphan drug indication, the
15
enrollment may be very slow.
So in that kind of
16
situation, maybe it's advisable to get a drug out
17
there and then do some optimization later on.
18
that's one aspect of it.
So
19
The other aspect is that I think we
20
oftentimes overlook -- that Jose actually referred
21
to it as really the difficulty -- even though you
22
might have a signal that the drug is working, the
A Matter of Record (301) 890-4188
332
1
difficulty of really precisely defining that
2
dose-response curve.
3
did a dose expansion cohort with more than one
4
dose, let's say, typically these are maybe 20
5
patient cohorts.
6
So in oncology, even if you
Take into account the fact that in oncology,
7
oftentimes you have a heterogeneous patient
8
population -- I gave the example about CML, which
9
is more homogeneous than other typically patient
10
populations.
11
mutations that might occur, and that causes a
12
heterogeneity in the response.
13
But even then, you have multiple
Particularly you saw there are patients who
14
are resistant to imatinib were less likely to
15
respond to dasatinib, and patients were intolerant
16
to imatinib.
17
So all these factors I think make it very
18
difficult to really precisely characterize the dose
19
response from a limited number of patients.
20
think one needs to have realistic expectations
21
about what that could actually -- how precise that
22
you could actually get that estimate.
A Matter of Record (301) 890-4188
So I
333
1
DR. RUBIN:
So can I ask just again, my
2
industry colleagues, maybe we should touch a bit
3
about what are the impediments for having larger
4
dose-finding cohorts in oncology.
5
been noted, typically there's an algorithmic
6
3 plus 3, or 3 plus 6, or something like that with
7
a small confirmatory cohort.
8
done, and you're off to multiple expansion cohorts
9
and various cancer types to look at efficacy.
10
I think, as has
Then, it's sort of
At least, again, what I've seen, there's not
11
commonly a lot of dose finding among those.
12
emphasis is across various cancer types or maybe in
13
combinations.
14
points have been in the talks -- it seems that
15
spending a little more time upfront with some of
16
the designs that have been talked about would be
17
overall better than having to go back months or
18
years later to revisit dose and schedule, sometimes
19
after approval.
20
The
But in retrospect -- and I think
So again, the question is, what are the
21
impediments to that?
Why not have a larger
22
dose-finding cohort in the beginning, and then move
A Matter of Record (301) 890-4188
334
1
to the indication specific groups. DR. BAILEY:
2
Maybe I can comment to that
3
one.
So I think there's a couple of different
4
factors.
5
go to your organization and say we need to make a
6
change from a small phase 1 paradigm into -- from
7
15 patients into a case where for every single
8
drug, we're now potentially going to be treating
9
maybe up to a hundred patients before we make a
10
decision either to move it forwards or kill it.
11
You're asking for significant investment on
12
compounds with no more preclinical data than we had
13
before.
14
it better, but in reality it's no more than we had
15
before.
Clearly, when working in industry and you
The key is we're actually hopefully using
16
Messaging from different areas will be -- I
17
know I've heard this message many times -- "I could
18
do exactly the same thing with 15 patients in a 3
19
plus 3."
20
You cannot do the same kind of dose finding with
21
15 patients in a 3 plus 3 design as you can with an
22
adaptive approach with significantly more patients.
And the reality is, you really can't.
A Matter of Record (301) 890-4188
335
1
Now, what will the significantly more
2
patients do?
3
the dose that you finally select, and hopefully
4
will then lead to the reduction of postmarketing
5
changes, which, to Mark's comment earlier on, is a
6
way to mitigate some of those risks of pricing
7
issues from a company perspective.
8 9
It will give you more confidence in
So you have to look in communicating, not just the impact on a phase 1 portfolio, but the
10
overall portfolio.
11
environment whereby you may well keep drugs alive
12
longer that really shouldn't necessarily be alive.
13
And a company has to be willing to take that step
14
in order to benefit from the real positives that
15
come, and that be very challenging.
16
buy-in from far more than just statisticians in
17
order to be able to get that.
18
DR. RUBIN:
19
DR. ROY:
But you're also in an
That requires
Thanks, Stuart. If I could also add to that.
20
Also, in the era of combination therapies, where
21
now if you have several drugs in a company's
22
pipeline or actually you're partnering with the
A Matter of Record (301) 890-4188
336
1
other companies, there are the possibilities of the
2
combination that you may want to evaluate in the
3
clinic, expand exponentially.
4
terms of enrollment, that may be an issue with
5
large studies. DR. RUBIN:
6
And then again, in
But I think again, if you don't
7
understand the monotherapy dose, it again seems
8
risky to start off with multiple combinations. DR. TANG:
9
Hi.
Shenghui Tang, FDA.
Eric,
10
your questions are about the extension cohort touch
11
a little bit in my questions.
12
haven't seen lots of patients in the extension
13
cohort, particularly when you're in the phase 1
14
study.
15
Right now, if you see a signal, then you extend to
16
hundreds of patients within each cohort.
17
Right now, we
Usually we have a small phase 1 study.
In this case, I know, Novartis, you talked
18
about this that sometimes it is a business decision
19
to go to each cohort.
20
start criteria?
21 22
Have you -- think about any
So enough is enough.
So maybe we can look at -- because anyway, it's a single-arm study.
You can look at the
A Matter of Record (301) 890-4188
337
1
response rate, but you cannot keep going and
2
include all the patients.
3
some starting criteria for these kind of extension
4
cohorts?
5
DR. BAILEY:
Yes.
So have you looked at
I think one of the
6
questions earlier on was about these very large
7
extension cohorts.
8
give was that really it's not about defining a very
9
quick escalation, then throwing hundreds of
10 11
And the presentation I tried to
patients on to one dose level. I think one of the keys is to be able to
12
actually get more patients during the escalation,
13
understand a lot more about the drug to be able to
14
make a recommendation on taking not just one dose
15
into expansion; maybe taking two doses into
16
expansion.
17
Between escalation expansion, really, the
18
primary difference is that during escalation you're
19
limiting risk to very small number of patients.
20
You're escalating maybe 3 to 6 patients at a time.
21
At some artificial cut point within the trial, you
22
say I have just about enough confidence in the
A Matter of Record (301) 890-4188
338
1
safety of this dose level, but I'm willing to
2
recruit a bulk of patients to it in one go. Now, I may be now asking a different
3 4
question.
I may be recruiting a slightly different
5
subpopulation.
6
between the expansion and the escalation is the
7
amount of risk I'm willing to expose, or the number
8
of patients I'm willing to expose to that given
9
risk.
But in essence, the difference
I think that has been one of the challenges,
10 11
is that going through the escalation, most
12
companies will still favor to get to this point
13
very quickly, and then get this bulk of data.
14
the point I was trying to make about Novartis
15
is -- or the business case about investing earlier
16
is not just investing in big expansions.
17
about investing in learning about the dose right
18
from the very outset and investing in more than
19
just 3 patients at any dose level across the dose
20
range.
21
determine with 15 patients, 1 dose.
22
not an effective way.
And
It's
I don't think you can wait until you just
A Matter of Record (301) 890-4188
That's just
339
1
DR. RUBIN:
Mark?
2
DR. RATAIN:
Mark Ratain.
I think if we
3
were here discussing something other than cancer,
4
there'd be no discussion.
5
this.
6
know how to identify a set of doses for phase 2.
7
They know how to dose-ranging phase 2's.
8
how to select 1 or more doses for phase 3, and they
9
know how to register an active drug.
Companies know how to do
And companies know how to do phase 1.
They
They know
So the question is, in the year 2015, why
10 11
should oncology be different?
12
how many -- how many oncology drugs do we have now?
13
There's medical need in Alzheimer's disease.
14
Right? DR. HOWARD:
15 16
So it's like this.
If I'm
drowning, if I -DR. RUBIN:
17 18
Because we only have
Can you just state your
affiliation, please? DR. HOWARD:
19
Oh, yes.
I'm from Novartis.
I'm sorry.
I'm Dan
20
Howard.
If I'm drowning, Mark,
21
I don't want you to look for the best rope to throw
22
me.
I want you to throw me any freakin' rope you
A Matter of Record (301) 890-4188
340
1
have available.
2
(Laughter.)
3
DR. HOWARD:
And that's really what's going
4
on.
5
medical need who are about to die.
6
they're going to die in less time than it takes us
7
to recruit fully for an individual study.
8 9
We have patients who are under a high unmet In some cases,
I would argue that it doesn't necessarily take oncology and separate it from any other
10
therapeutic area; it's any therapeutic area where
11
this kind of medical need exists, you're going to
12
have the same issue.
13
Now, what's interesting for me is that when
14
you say to a company, I'm going to grant you
15
breakthrough therapy designation, and I'm going to
16
give you a preferential status for accelerating
17
your development, you're actually taking away the
18
motivation to study multiple doses.
19
them to find the first rope they can find that has
20
a reasonable benefit-risk.
21
suggesting is that we'll optimize the dose at a
22
later time.
You're asking
And what you're
A Matter of Record (301) 890-4188
341
1
I think that's maybe something that we're
2
going to have to deal with.
3
struggle with the fact that maybe the selection of
4
the dose under certain circumstances is going to
5
not have the finality of saying we know this is the
6
best benefit-risk ratio we can achieve.
7
acceptable at this time, and we'll optimize at a
8
later time.
9
DR. BAILEY:
We're going to
It is
I also think that it's not
10
necessarily accurate to say that in all other
11
indications, everything is perfect.
12
on non-oncology indications, implementing exactly
13
the same methodology for children that have
14
terminal disease, that's not oncology, but as Dan
15
says, it's about medical need.
16
So I've worked
We've had organizations come to us to
17
request how to implement these kinds of designs,
18
purely because standard approaches are not going to
19
be able to facilitate the patients to get effective
20
doses fast enough.
21 22
I take the point that we can certainly learn from other areas, and the whole point of this
A Matter of Record (301) 890-4188
342
1
session is how do we learn to do things better.
2
But I don't think that it's also the case that in
3
other indications everything is hunky-dory. DR. BELLO:
4
Akintunde Bello, Bristol-Myers
5
Squibb.
So the last two questions really stole the
6
questions that I wanted to ask, but I was going to
7
ask the question around, okay, breakthrough therapy
8
designation.
9
oncology predominantly.
It's not just oncology, probably
But what has been the agency's experience in
10 11
terms of these other indications, other therapeutic
12
areas, with regards to their development paradigms?
13
Have they kind of followed a more traditional path
14
with regard to drug development in terms of
15
establishing and investigating exposure response,
16
dose response, et cetera?
17
DR. SINHA:
I'll let Dr. Sridhara respond to
18
this.
I think it's quite a mix, Akintunde.
I
19
think we see a breakthrough designation across many
20
different therapeutic areas.
21
from the area of pharmacology, I naturally lean
22
towards using that as a strong basis for making
And I think being
A Matter of Record (301) 890-4188
343
1
some sort of drug development decisions.
2
where I actually come from.
3
That's
So I actually think the elements of -- I
4
think all we heard today, the PK/PD elements, our
5
ability to link exposures to intermediate
6
biomarkers that are now linking in some way to
7
outcomes, et cetera, are huge.
8
innovation that has happened over the last decade,
9
and we should be maximizing this.
10
This has been the
So we see this in all forms in a sense.
11
There are areas in certain rare diseases,
12
et cetera, where we don't get a lot of data, and we
13
are having to make very difficult decisions on very
14
little information.
15
oncology's probably a little ahead.
16
You know in that contrast,
So I think the agency makes these decisions
17
based on unmet need and really tries to do the
18
right thing for the patient at the end of the day.
19
It's very conscious about adding more needs and
20
adding more demands on drug development.
21
decision is very carefully weighed.
22
translates into postmarketing optimization,
A Matter of Record (301) 890-4188
So every
Some of this
344
1
et cetera. But fundamentally, where I actually come
2 3
from -- and I think this is unique.
If you start
4
from a position of pharmacology and you build your
5
story, whether it's clinical, nonclinical data, and
6
you're getting a sense of what is happening on
7
dosing as best as you can, I think at the end, you
8
will win, better than if you do nothing. That's always the position I've actually
9 10
had.
11
anything that makes me feel otherwise.
12
been cases where, yes, miracles have happened where
13
people have done nothing and drugs have been
14
successful.
15
sustainable long-term thing.
16
balance.
17
advice on each drug case-by-case.
18
And actually, I don't think I've seen There have
But that's highly -- that's not also a So I think it's a
I think the agency tries to give the best
DR. BELLO:
Also, a follow-up question.
In
19
terms of breakthrough therapy designation, I think
20
that kind of denotes a certain level of
21
collaboration between the agency and the sponsor
22
company.
And I'd think this type of meeting, in
A Matter of Record (301) 890-4188
345
1
which we are kind of upfront addressing the issue,
2
is clearly the first step in terms of kind of
3
highlighting the gaps and the needs and our
4
thoughts about how we can address this.
5
So there's got to be two areas of attack of
6
the issue in terms of those discussions with the
7
regulators.
8
company with our development plans, if you really
9
desire us to investigate this, there's a contact
When we as sponsors come to the
10
point at which you can make those suggestions.
11
that kind of in my mind has a very strong impact on
12
the organizations that we work in, to actually
13
indicate that this is a requirement, an expectation
14
of our companies as we develop our drugs.
15
DR. SINHA:
And
I totally understand that, and I
16
think we are very conscious about this, that
17
sometimes this room is probably not the group that
18
we are trying to convert here.
19
(Laughter.)
20
DR. SINHA:
I think we understand that quite
21
well, and I think that was, to the large extent, a
22
very similar case at the meeting in December on
A Matter of Record (301) 890-4188
346
1
dose finding.
The converts were in the room.
The
2
people who were writing the checks were somewhere
3
else.
I think we totally understand that. So I think what we should be also -- I think
4 5
what sponsors should really be thinking about is,
6
you need to articulate your questions when you come
7
in for these meetings with these things in mind,
8
versus asking, this is my study design, do you
9
agree?
10
You I think as a sponsor, have the means to
11
change the dialogue to something that starts making
12
you focus on these things. That is not to say that we at the agency are
13 14
not responsible for bringing these things up, but
15
sometimes it is difficult to know all that you
16
know.
17
to be very honest, especially in the IND phase.
18
And if you articulate these questions that you
19
really think you want to be pressing on and to be
20
making these difficult decisions of including
21
another dose; how am I going to use a
22
exposure-response analysis, what are the challenges
You know more about these drugs than we do,
A Matter of Record (301) 890-4188
347
1
you're going to face if you did an exposure
2
response analysis in some kind of decision-making?
3
You need to ask those questions.
4
compels us to be more clear, more definitive, so
5
there are no surprises at the end of the day.
6
DR. SRIDHARA:
It also
So your first question about
7
comparison to other disease areas with respect to
8
breakthrough therapies, I don't think I can comment
9
fully on other disease areas, but certainly it's
10
very clear that in oncology, the tolerance for
11
toxicity is very different.
12
So how we weigh a breakthrough therapy in
13
oncology may be very different from what they see
14
in other areas.
15
higher efficacy standards, or activity standard
16
more than efficacy, because if the efficacy's
17
already established, then there is no breakthrough
18
therapy; you have an approval there.
19
So they may be looking for even
So I think we have to understand that
20
breakthrough therapy is only a designation.
21
just that we are acknowledging that we are seeing
22
something more than what we would normally see as
A Matter of Record (301) 890-4188
It's
348
1
an activity of the product, so we are willing to
2
say yes.
3
as possible, and we will help you in whatever way
4
we can.
5
that.
6
Maybe you need to develop this as quickly
And that's what we're trying to say with
DR. RUBIN:
So I think we want to -- Mark,
7
you can maybe add a follow-up.
But I do want to
8
put up questions 2 and 3, to make sure we have time
9
to address those as well.
10
Is yours with regard to question 1?
11
MR. RATAIN:
I just want to follow up on the
12
discussion as to why oncology should be different.
13
I think we can agree that there are certain drugs
14
that are so important that they need to be moved to
15
patients as quickly as possible, and we designate
16
those now with breakthrough status.
17
So logically, it then follows, if a drug is
18
not breakthrough status, it should not be treated
19
differently, and therefore, there should be more
20
time spent optimizing dose.
21
at least a reasonable place to start for the agency
22
and industry to consider.
I think that would be
A Matter of Record (301) 890-4188
349
1
DR. RUBIN:
2
DR. NIE:
Thanks, Mark. Thank you.
We move to question
3
number 3.
4
related.
5
good strategy.
6
extensively to questions, will we allow it to
7
continue to dominant oncology dose finding?
8
also, why don't we use the integrated approach
9
more?
10
Actually, these two questions are very We all know MTD with 3 plus 3, it's not a But we are still using it
So these two questions:
And
What has hindered
11
the oncology drug development from using a more
12
tailored approach, integrate all existing data
13
together, as we all discussed, the toxicity data,
14
preclinical, clinical, PK/PD data, all together?
15
The second question relates to what is a
16
pertinent resolution from moving away from MTD
17
3 plus 3 to integrated adaptive approach.
18
discussed obstacles, but more importantly, we need
19
to find a solution to promote more effective dose
20
finding.
21 22
DR. DE ALWIS:
We
I'll have a go at question 2.
I think, if I remember, when I started getting into
A Matter of Record (301) 890-4188
350
1
this field and started working in oncology, an
2
oncologist told me once, he said, "We learn two
3
things in training.
4
dose and body surface area, neither of which are
5
very useful."
6
One is the maximum tolerated
So I think this is partly what has kind of
7
hindered us in many ways is that paradigm, of the
8
cytotoxic paradigm, and we're trying to kind of go
9
back and really revisit this whole thing and
10
address.
11
worked for a certain class of compounds is
12
difficult.
13
And changing a paradigm that may have
So that's what we're faced with.
The other part I think we're faced with is
14
actually, as I mentioned earlier, in the
15
preclinical domain, and really getting to
16
understanding, as I mentioned, the minimal
17
efficacious dose, where we had the cusps of where
18
you're not seeing efficacy and you're just seeing
19
efficacy.
20
preclinical study that can really inform us, and
21
then doing studies with IV infusions, where you
22
have control, precise control over the Cmax, the
That is the most important kind of
A Matter of Record (301) 890-4188
351
1 2
AUC, and the time on target. So well designed studies that help you
3
really define what is driving the efficacy, and
4
this will help determine the schedule dependencies,
5
et cetera, and it's the body of data.
6
So I think we're very good -- and companies
7
actually are at fault is well, is having people
8
move around from one particular compound to
9
another, one particular target to another, and you
10 11
lose institutional knowledge. So it's the ability for people to stay on
12
things, and that body of evidence and body of data
13
really lends itself to much better well-designed
14
studies, and as Stuart pointed out, in terms of how
15
you do a phase 1 study to really maximize that
16
information so you don't have this highly
17
uninformed priors, but actually very tight priors.
18
Going to I think the talk by Jose, he
19
indicated that, unfortunately, with the
20
dose-response study, there was a much larger
21
confidence around that and not as tight as just
22
looking in pairwise.
But part of that can be
A Matter of Record (301) 890-4188
352
1
reduced if you actually know the shape of your
2
exposure response relationship, for example, and
3
you can reduce that uncertainty.
4
So that's all there.
In fact, if you
5
actually were to delve into all the information
6
that's available for a particular compound, it
7
isn't just that compound, it's that whole class of
8
compounds.
9
going to be the very first in class.
10
It's very unlikely that you're totally
There may be situations like that in which
11
case you will have less information.
12
most part, you're going into either a chemistry
13
space or understanding of biology where others have
14
gone before.
15
really reduce those priors and use a much more
16
Bayesian approach to drug development.
17
But for the
So it's using all that information to
We do that in most things we do day-to-day,
18
so why not use this and, rather, not look at every
19
study as a one-off.
20
DR. PARIVAR:
Kourosh Parivar, Pfizer.
I
21
would wish that we were in an environment that we
22
had preclinical models which we could rely on to
A Matter of Record (301) 890-4188
353
1
translate into a human setting.
2
see being generated is TGI models, PDX models,
3
xenograft models.
4
when you see those models and the assets coming for
5
assessment, what you see is 100 percent more or
6
less elimination of the tumor.
7
cancer would have been gone 10, 20 years ago.
8 9
Currently, what
And the majority of the time
If that was true,
So we know per definition that's not true. Those models grossly overestimate the efficacy of
10
the products.
11
know our models, the preclinical models, are
12
grossly overestimating the outcome in the lack of
13
having any reliable model, I guess that limits some
14
of the proposals they're putting in place, Dinesh;
15
that I would wish that we were in a situation that
16
we had a reliable model for which we could do more
17
PK/PD modeling preclinically and translate that to
18
human setting.
19
So if we are in a situation that we
DR. DE ALWIS:
So I will actually challenge
20
that, I think, concept.
I think I agree that -- as
21
someone put up with a quotation from Box, "All
22
models are wrong and some are useful."
A Matter of Record (301) 890-4188
And it's
354
1
getting to the "some are useful" part. I remember a particular example I was
2 3
involved with some years ago, and it actually was
4
in fact in -- this is now probably about 10 years
5
ago -- no, 15 years ago.
6
losing my memory.
Sorry.
I think I must be
So it was about 15 years ago, and it was
7 8
actually around bone marrow suppression.
And I
9
remember the particular example we talked about
10
where the toxicologist said, "I don't know how
11
you're going to predict what's going to happen in
12
the clinic," and this is just -- what you'll do is
13
actually the doubling of the black arts [ph], in
14
fact.
15
But what we did was actually concertedly
16
look at that preclinical experiment, and we
17
incorporated things like the colony-forming unit
18
granular macrophage assay, the Ralph Parchment
19
assay, incorporated a very well-defined study;
20
understood incorporated unbound concentrations.
21
That was one of the things they were doing, would
22
just bound total drug.
And there was a key issue
A Matter of Record (301) 890-4188
355
1
because the compound was highly bound. It's really going to understanding the
2 3
mechanisms, understanding the pharmacokinetics,
4
understanding what's driving it, incorporating a
5
well designed study, seeing if it was receptive to
6
a quantitative approach, looking at the differences
7
of sensitivity between the preclinical species in
8
man.
9
As you know with the colony forming unit
10
granular macrophage, you can look at the in vitro
11
bone marrow in animals and man.
12
translation, we were accurately able to predict
13
where we had grade 3/4 toxicities in man in terms
14
of bone marrow.
15
And based on this
Obviously, now we look back, and I think the
16
preclinical session show that with respect to bone
17
marrow, we do a good bone marrow NGI.
18
do a good job.
19
I think we
There are others that we can't.
Yes, I agree.
But I think we need to look
20
at those things very carefully, work with these
21
other functions, statisticians, pharmacometricians,
22
pharmacologists, biologists, and get to that right
A Matter of Record (301) 890-4188
356
1
spot.
I think it's easy for us to walk away and
2
say you know --
3
DR. PARIVAR:
4
DR. DE ALWIS:
5
DR. PARIVAR:
No, no, don't get me wrong -Yes? I hitting more function for
6
PK/PD.
I guess they're doing a lot better work
7
when it comes to prediction on the safety side.
8
The neutropenia model and others are much, much
9
better developed, much more predictive.
What I was
10
challenging was going on the efficacy side.
11
want to predict the efficacy of a product, that's
12
much more challenging that predicting, say,
13
thrombocytopenia or neutropenia and such, based on
14
your preclinical data. DR. DE ALWIS:
15
If you
I think again -- yes, I
16
agree, xenograft tumor models have significant
17
issues.
18
reflecting something more of a tumor that is in man
19
rather than some of the xenograft cell lines that
20
we have.
21 22
But we've got better with things
The other is looking at pharmacodynamic endpoints, something with biomarker endpoints, and
A Matter of Record (301) 890-4188
357
1
seeing how translatable that is.
2
variations on -- I mean, I agree, it isn't easy.
3
The other is I think getting into the clinic, and
4
as a mentioned in my talk, looking at endpoints
5
like change in tumor change and not going to
6
something like a resist response where you're
7
losing information.
8
information that we're getting.
9
DR. PARIVAR:
There are
So we could do more with the
No, that I agree.
10
[Indiscernible] and us, and a few others have been
11
publishing on that.
12
DR. BAILEY:
13
challenges as well.
14
statistician working in industry, I can tell you
15
that Bayesian methodology is the best to use.
16
for statisticians, you put two statisticians in the
17
room, you get three different opinions.
18
(Laughter.)
19
DR. BAILEY:
Just one comment back to the From the perspective of a
And
The number or the amount of
20
papers, the literature that is published around
21
debunking the Bayesian methodologies that come
22
forward or the Bayesians who debunk the algorithmic
A Matter of Record (301) 890-4188
358
1
approaches, depending on how you implement the
2
designs, we can't agree ourselves. It's very easy for someone to say, well,
3 4
you're introducing prior information, but how could
5
you have any confidence in the prior?
6
the potential to have priors are very weak
7
information.
8
information, I really want to use it, and I
9
incorporate it.
We showed
And yet, when I have a lot of
I need to be able to allow for
10
dissimilarity between the actual patient group and
11
what I thought would happen, and that often doesn't
12
happen.
13
Even within Bayesians, we argue over did you set
14
the right prior.
15
And then you have interpretation issues.
So in that framework, you're then asking
16
clinicians to trust you to implement designs that
17
within your own peers don't have agreement.
18
think that's been a very challenging thing.
19
And I
When we submit Bayesian phase 1 designs to
20
the authority, we submit them with simulation
21
showing not just the long-term operating
22
characteristics in term of determination of MTD,
A Matter of Record (301) 890-4188
359
1
but the frequency that you treat different patients
2
within trials at different dose levels.
3
show what you would kind of look at as the power of
4
the study to detect the correct dose.
5
give operating characteristics from a frequentist
6
mindset.
7
And we
So we try to
But I would ask the authority, when you
8
receive a 3 plus 3 study, do you request the
9
comparison of the characteristics of that design to
10
an adaptive Bayesian approach, and request the team
11
who is submitting that to assess whether there's a
12
better approach for their study?
13
DR. SRIDHARA:
I think that's the reason
14
that we have this meeting, that we want to come out
15
and tell you that we don't ask the sponsors to do 3
16
plus 3 trials at all.
17
what have you, for adaptive designs, we ask more in
18
the confirmatory phase 3 trials if somebody is
19
suggesting a Bayesian design.
20
And yes, the simulations or
Certainly in drugs, we don't have that much
21
of experience with Bayesian designs in the
22
confirmatory phase 3 randomized studies.
A Matter of Record (301) 890-4188
But for
360
1
these dose-finding studies, if anyone comes to ask,
2
we are saying, do the Bayesian.
3
3 plus 3; 3 plus 3 has basically no statistical
4
properties and it's just that it sounded good.
5
get a standard deviation if you do three.
6
(Laughter.)
7
DR. SRIDHARA:
Don't go with the
You
If you do 2 plus 2, you don't
8
get that, and that's how it comes out.
I mean,
9
there's no rhyme or reason why it should be cohorts
10
of three versus cohorts of 2 or cohorts of 5.
So I
11
don't think we are saying that you should do that
12
at all, and still we see a lot of 3 plus 3 designs. If you ask, well, are you telling them go
13 14
back and do Bayesian designs in those cases?
No,
15
we don't sort of say, okay, this is the comfort
16
level.
17
how we are doing.
18
done.
19
design for dose finding, we are certainly not
20
saying no to it.
21
you're characterizing your toxicity and what's your
22
model, but certainly we are not saying no to that.
This is the expertise you have, and this is Fine.
And this has it has been
But if somebody comes up with a Bayesian
We may have questions on how
A Matter of Record (301) 890-4188
361
1
DR. NIE:
Following this question, following
2
your discussion, I just wonder what other
3
impediments do we see moving from MTD 3 plus 3.
4
Could you please comment on that?
5
DR. RUBIN:
Being a clinician who -- in some
6
cases, I guess I'm a wanna be statistician, so I
7
tend to be inclined more towards these things.
8
I do think, Stuart, I think that's the first time
9
I've actually seen someone put up simulations of
10 11
But
iterative types of dose-finding type things. Typically I think -- I know when we're
12
developing studies, early studies, and we ask the
13
statistician to provide the approach, if it is a
14
Bayesian approach, what you tend to see as a
15
clinical person is the equation --
16
(Laughter.)
17
DR. RUBIN:
-- a little bit of the -- some
18
of the predictive probability tables.
But I think
19
the picture part that you showed tends to be
20
missing.
21
you're coming from where you've always done dose
22
finding with 3 plus 3, it can be hard to be
So I think that, again, Dinesh said, if
A Matter of Record (301) 890-4188
362
1
convinced why you should take the time to do
2
something different without that kind of
3
explanation. DR. BAILEY:
4
Well, I think the message
5
earlier on really is communication is the absolute
6
key.
7
very complex -- an impediment is if you think 3
8
plus 3 is something that doesn't require a
9
statistician.
And to be able to take, again, what can be
So an investigator can run their own
10
IRT study with a 3 plus 3 without the need for a
11
statistician to analyze the safety data at all.
12
Moving into an adaptive environment where
13
you need statisticians, you start having another
14
language introduced into the discussion.
15
absolutely critical that you find a common
16
understanding for what you're really trying to
17
achieve.
18
In terms of moving away from 3 plus 3,
19
again, a simple algorithm.
20
therefore I stay where I am.
21
one, therefore I go down.
22
It's
I've seen one DLT, I've seen another
Now we're in a point where we're saying
A Matter of Record (301) 890-4188
363
1
we're not just wanting to rely on this DLT, but I
2
want to see all my other data to be able to help me
3
choose from this potential set of doses.
4
that data's not available, all I have to rely on is
5
that DLT.
6
an operational standpoint, we can design the most
7
amazing trials in the world, but if we don't have
8
data to be able to implement them, it's still a
9
major issue.
10
And if
And that's something, which again, from
DR. PARIVAR:
Stuart, at Novartis, you were
11
able to successfully move away from the 3 plus 3
12
paradigm, right?
13
DR. BAILEY:
14
DR. PARIVAR:
Yes. I think it's the only company
15
that I know that has been able to do that, based on
16
communication.
17
DR. BAILEY:
18
DR. PARIVAR:
I'll let you say that. I didn't participate in that,
19
but I was there witnessing it at the time.
20
it's about communication, and that's probably the
21
answer to that.
22
DR. SRIDHARA:
I think
So Stuart, I have a question
A Matter of Record (301) 890-4188
364
1
for you.
You are saying that among statisticians,
2
if they're arguing between 3 plus 3 and Bayesian
3
designs for these dose-finding type trials, not the
4
confirmatory trials, who are the ones who are
5
against?
6
institutions, or who are these people?
Are they typically from academic
7
(Laughter.)
8
DR. BAILEY:
9
I can tell you that within our
company, we've had a very strong push to implement
10
these designs, not just from statisticians, from
11
clinicians, from pharmacologists, who all believe
12
that we need to do better.
13
implement the designs, there are even
14
statisticians -- within our team I had a
15
statistician come to me and say, "You're either
16
born a Bayesian or you're not."
17
(Laughter.)
18
DR. BAILEY:
But when we start to
And I'm not.
There were presentations about
19
the concept to escalate until you see a DLT, then
20
implement a Bayesian paradigm.
21
you've a DLT, you have no information.
22
is you do have information.
Because until The reality
You have information
A Matter of Record (301) 890-4188
365
1
the drug is safe, or at least safe -- I use a term
2
"safe" that I know regulatory and a phase 1 view of
3
safe would be quite different.
4
But the reality is there are academic
5
members -- I mentioned my PhD was in this
6
environment, and I designed what I thought was a
7
perfect phase 1 trial that incorporated safety,
8
activity, PK data, all in one.
9
came into industry was that I can't get that
10
The reality when I
activity data at the same time.
11
So whilst I try to implement something from
12
a pure methodological standpoint, the realities are
13
that a phase 1 design is not just about the
14
methodology.
15
outweigh some of the ability to implement certain
16
designs, and that's the challenge to get people
17
from an academic standpoint to understand.
18
The operational challenges can
DR. NIE:
So another related question, are
19
there any concerns that the interpretation of the
20
result will be dictated by statistical models?
21
there any concern about that?
22
DR. BAILEY:
Is
Sorry, I didn't quite get --
A Matter of Record (301) 890-4188
366
DR. SRIDHARA:
1
So what he is asking is you
2
have a model telling you what to do next.
3
plus 3, they are looking at, okay, do you have
4
toxicity?
5
a problem?
No.
DR. NIE:
6
Then move on to the next.
In the 3
Is that
So the investigators feel very
7
comfortable with 3 plus 3.
8
understand.
9
feel lost and have to rely on statisticians to
But right now, with the model, they
10
interpret that.
11
that?
12
They feel they
Do you have any concern about
DR. BAILEY:
I think that is very true.
13
With most investigators, there's an initial
14
apprehension to using what they consider to be a
15
black box.
16
data in and it told you what dose to give.
17
some people's mindset of what the CRM really is.
18
And it tells me what dose to give, and if I give
19
that dose, I optimize the MTD.
The CRM was a design where you threw That's
20
However, you're taking away the ability of
21
an investigator to then sit there and say, well, I
22
don't agree with this.
This is not a dose I would
A Matter of Record (301) 890-4188
367
1
want to give.
The reality is you're not taking
2
that away.
3
patient onto the study if they don't agree with
4
that dose level.
No investigator is required to put a
In the case that we had, we made the point,
5 6
the model is not recommending the dose.
7
is identifying a set of doses that based on certain
8
prespecified safety criteria, this is a range that
9
we should look at, and then we consider the other
10
The model
information. Now, you may have other models that are then
11 12
working to pick that dose, but I think an
13
investigator is now, through more and more
14
experience -- I mean, their understanding of PK/PD
15
relationships that are presented to them
16
graphically are to be able to then look at that and
17
say, well, okay now I understand. I can understand the justification of this
18 19
dose.
The longer term toxicities, which will make
20
me pull back on my dose, I understand this.
21
once they've started to use it, then they really
22
find the power of it because they're actually the
A Matter of Record (301) 890-4188
And
368
1
ones deciding the dose.
2
DR. NIE:
3
question from the floor.
4 5
Thanks.
DR. BULLOCK:
We are going to take a
Yes, thank you for letting me
stand for so long because it's cold out there.
6
(Laughter.)
7
DR. BULLOCK:
Julie Bullock, D3 Medicine.
I
8
have a question for you, Stuart.
I was surprised
9
by the number of patients that are required for the
10
CRM method to find a dose.
11
the patient cohorts were rather large compared to
12
some of the 3 plus 3 designs that I'd seen and
13
reviewed at the agency.
14
I thought the N's for
I was wondering, how predictive are these
15
studies?
16
methods, so now you've got your CRM dose and you're
17
going into phase 3 or your pivotal trial.
18
just say that there are no phases in oncology
19
anymore.
20
I know that you've used these CRM
Let's
So you're in your pivotal study.
How predictive was your work and the dose
21
finding to the pivotal study?
How many patients
22
still needed dose reductions?
That's what I kind
A Matter of Record (301) 890-4188
369
1
of want to know, because this is a complicated
2
design.
3
Three plus three, I think the reason why we
4
haven't moved away from it is because it's simple.
5
I mean it doesn't need a statistician, and it
6
doesn't need models.
7
pitfalls that we can learn from failed trials or
8
failed drugs that haven't incorporated PK/PD well
9
enough, but I'm wondering if we need to get as
I think there are definite
10
complicated as CRM in order to get really doses for
11
a pivotal study.
12
I also have a comment about breakthrough
13
therapies.
14
understand the industry's hesitation for dose
15
finding and taking the time for a breakthrough
16
therapy.
17
be a little more open and maybe using a multi-dose
18
phase 2 study for accelerated approval for a
19
breakthrough therapy drug.
20
Since I'm up here, I'll just say it.
I
But I think that the agency then needs to
I think that showing that you have a dose
21
response for efficacy and safety is way more
22
important than just a single arm trial at one dose
A Matter of Record (301) 890-4188
370
1 2
for a breakthrough therapy drug. DR. BAILEY:
I'll try to be brief.
The
3
sample size question, we have examples of phase 1
4
trials that have used 21 patients.
5
goes through no toxicities, then we see toxicity,
6
and the next, now our dose is completely safe, you
7
can already make a decision.
8 9
When escalation
So at the minimum point, you're approximately the same.
We have some bound on the
10
sample size of about 21 to 24, just based on
11
simulations around how confident we are about MTD.
12
The larger sample size comes when you really
13
need to answer different questions.
14
showed had 103 patients, it had 3 doses levels that
15
each had 16, 18, and 24 patients.
16
questions that we were asking were not about were
17
we identifying a maximum tolerated dose.
18
questions there were, can we identify between these
19
doses, changes between the onset of ocular
20
toxicities, the diarrhea, nausea, frequencies, and
21
the longer duration of treatment.
22
size to be able to answer those questions is
A Matter of Record (301) 890-4188
The study I
And the
The
And the sample
371
1
completely different to sample size needed to
2
estimate an MTD. So when necessary, you are going to have a
3 4
larger trial, because you're not answering the same
5
questions.
So that's the quickest way.
DR. KARRISON:
6
Ted Karrison from the
7
University of Chicago.
My question is whether it's
8
3 plus 3 or CRM or whatever the design, how we
9
settled on the -- 20th to the 30th percentile is
10
the target percentile for toxicity that seems about
11
right.
12
patients that could answer this best.
13
And maybe it's clinicians who deal with
I was wondering how do we think about what's
14
an acceptable level of toxicity?
15
that doubled my median survival time, I might be
16
willing to put up with more toxicity that only
17
increased it by 20 percent.
18
If I had a drug
So it seems like the whole toxicity efficacy
19
needs to be tied together a bit, and yet we're
20
always in this 20 to 30th percentile.
21 22
DR. RUBIN:
I think it's a good point.
when you get in combinations, it also can make
A Matter of Record (301) 890-4188
And
372
1
things tricky because if you're combining with an
2
older standard of care drug where the dose is the
3
MTD, you actually have very little room then to add
4
on the second drug and not hit a 30 percent DLT
5
rate, for example, that's fully driven off of the
6
prior drug.
7
Others may want to comment.
I think
8
sometimes we've actually gone higher than the
9
30 percent rate in settings like that, and I
10
believe the agency's accepted those arguments along
11
the lines as what you've said.
12
So I think it has to be a dialogue, as you
13
know, in terms of the particular patient population
14
that you're studying, and there may be cases where
15
you might want to consider higher or lower rates,
16
again, depending on the patient population.
17
don't know if others want to add to that.
18
DR. DE ALWIS:
I
I think, Eric, it goes back
19
to some of the review data that I showed earlier,
20
in that it depends on the kind of MTD.
21
sort of subjective assessment as well, there might
22
be more room to go higher.
If it's a
And if it's a
A Matter of Record (301) 890-4188
373
1
reversible -- I guess it goes down to the nature of
2
the toxicity as well.
3
DR. BAILEY:
I think one comment from
4
experience of trying to have higher boundaries
5
accepted -- that's not been very
6
successful -- we've generally had push back that it
7
is 33 percent, and that's it.
8
where did this magic number come from?
9
it's not one that always gets answered.
10
UNIDENTIFIED SPEAKER:
11
DR. BAILEY:
12
(Laughter.)
13
DR. BAILEY:
And my question is, I mean,
Three plus three?
I know, it's 3 plus 3.
I think the one thing I would
14
question, I mean we're still talking about
15
identifying MTD.
16
toxicity.
17
into more and more targeted treatments and target
18
indications, and if you take it out of the kinase
19
into the immuno therapy space, it's not necessary
20
to identify toxicity.
21 22
We're still talking about seeing
And the reality is that as we're going
It's definitely a requirement to mitigate against exceeding toxicity, but are we really
A Matter of Record (301) 890-4188
374
1
trying to drive, in the chemo mindset, to toxicity,
2
which in previous drugs was an indicator we were
3
killing cells and that was our on-target effect?
4
mean, do we want to stay in this paradigm that MTD
5
is the primary goal?
6
believe that's the case.
7
DR. NIE:
8
DR. MCKEE:
9
from FDA.
Because for me, I don't
Questions from the floor? Yes, my name is Amy McKee.
I'm
So my question is directed to Dr. Janne
10
because we've heard a lot from industry and from
11
the FDA side.
12
these first in human trial designs, what are the
13
hindrances at your institution, both from the
14
investigators and from the IRB?
15
percentages do you see for these different kind of
16
studies in oncology, and the investigators at your
17
institution, what do they think of them?
18
I
But as an investigator, when you see
DR. JANNE:
And what sort of
You're talking about the
19
different types of designs in terms of three?
So I
20
think there's certainly some growing pain.
21
talked about, there's a certain degree of comfort
22
with an older design.
As we
And as we go into different
A Matter of Record (301) 890-4188
375
1
model-based designs, I think that's taken some
2
learning, both from the investigator side and from
3
the IRB side, quite frankly.
4
But I think it's a slowly evolving process,
5
and it's a continual educational process to be able
6
to show that this a better way to do it and can
7
lead to a better way to find an efficacious dose.
8
But I think it's in evolution, I would say.
9
DR. NIE:
That is a very good comment
10
about -- so actually we are talking about the
11
integrated approach, but how do we actually do it?
12
I remember, Stuart, in your last presentation, you
13
showed a lot of data.
14
sample them together to make a decision?
15
get more comments on that.
16
DR. BAILEY:
But how do we actually Maybe we
The advances in electronic
17
databases have allowed us to be able to see data
18
only a day or a few days after it's been entered
19
into a database at the site.
20
significant more operational challenges.
21
want to have pharmacokinetic data, pharmacodynamic
22
data available, in time to be able to make
There's clearly
A Matter of Record (301) 890-4188
If you
376
1
decisions, you need to have a very strong
2
operational component, to the running of the trial
3
to have samples shipped to be analyzed, to have
4
that data brought in.
5
mean that you're delaying your meetings, and that
6
can be challenging because everybody wants to
7
escalate as quickly as possible.
Just a couple days delay may
8
So very, very clear understanding.
And as a
9
company, there's definitely an onus on you to make
10
sure that you've helped the investigator understand
11
what is critical information to have for the next
12
meeting.
13
then it's in some sense your own fault if you don't
14
necessarily receive something.
15
If you haven't told them what's critical,
There is the need to make sure that the
16
sites are entering the data and that teams are
17
reviewing that data, and that's an investment.
18
I mentioned in the past, it wasn't about
19
statistics.
20
2005, was four people.
21
people running the phase 1 studies.
22
has grown, but not grown by such an enormous amount
As
I mean our group, when I joined in Now we're a team of 30
A Matter of Record (301) 890-4188
The portfolio
377
1 2
since then. So there is a significant amount more
3
investment, not just into the drugs, but into the
4
teams that are needed to analyze them.
5
DR. RUBIN:
I'll just add that, as you know,
6
making sure if you have a pharmacodynamic assay
7
that you need to have ready, that can be a
8
challenge; to make sure that you have a reliable
9
assay that's ready to go as you're starting and the
10
escalation cohorts.
11
DR. NIE:
Questions?
12
DR. RATAIN:
Yes.
I just want to respond a
13
bit more to some of these issues that you've got up
14
on the screen.
15
have -- I've been doing this for a long time.
16
wrote a paper in 1993 on what we would now call an
17
adaptive phase 1 trial design, where we've modeled
18
and simulated.
19
want to define the MTD, there's no question that we
20
want to use all of the data.
21 22
I think the real concern some of us I
And I completely agree that if we
But I think we now know that, number one, if you really have a really spectacular drug, you
A Matter of Record (301) 890-4188
378
1
don't need to dose it at a dose that causes what we
2
accept today as dose-limiting toxicity in
3
25 percent of patients.
To me, it doesn't compute.
Then, the purpose of understanding the MTD
4 5
is more to measure, to estimate our therapeutic
6
window, to estimate what would happen in an
7
overdose setting; to estimate what happens in
8
pharmacokinetic outliers. I guess, why are we so consumed with
9 10
precisely trying to estimate the MTD?
I noticed in
11
your ceritinib trial, you were doing doses of 600,
12
650, 700, 750.
13
got inter-individual variability in adherence,
14
inter-individual variability in bioavailability, a
15
food effect, which you didn't know at the time or
16
didn't incorporate at the time.
We've got an oral drug here.
We've
So where is this drive for precision coming
17 18
from?
19
We've got to get the dose precisely right for each
20
individual patient?
21
DR. BAILEY:
22
Is it because we're back in the BSA mindset?
That's what doesn't compute. Yes.
In some sense, I agree
with you.
A Matter of Record (301) 890-4188
379
1
DR. RATAIN:
Okay.
2
DR. BAILEY:
I think that when you're in the
3
settings -- before your start the trial, you're
4
never quite sure whether you're going to need to
5
end up at that MTD or not.
6
significant preclinical information that helps us
7
understand that we're probably never going to get
8
there.
9
You may have
But possibly, you may need to end up there. The ceritinib case was a very interesting
10
one because the dose we finally ended up at is
11
still lower in exposure than the dose that we think
12
you need in order to see the activity that, from a
13
preclinical side, we expected.
14
patients have brain metastases.
15
that you still need to get as much in as possible
16
to be able to treat these is still there.
17
Many of the So the concept
The precision in some trials is absolutely
18
necessary.
In other trials where you start to see
19
significant activity and you're just trying to
20
quantify maybe a supra-dose -- to some sense,
21
though -- and I know for any DDI for a compound I'm
22
going to combine this with, I'm still within a very
A Matter of Record (301) 890-4188
380
1
safe window, you may not need that level of
2
confidence.
3
25, 30 patients at those dose levels.
4
agree.
And you're never going to require the I completely
Adjournment
5
DR. NIE:
6
Sorry.
The time is up.
I would
7
like to ask a voting question from the panel.
8
you like to see moving away from MTD with 3 plus 3?
9
Just, we'll take this vote.
10
Don't need to answer,
just a vote, yes or no.
11
(Laughter.)
12
DR. NIE:
13
Will
Yes or raise your hand if you say
yes.
14
(Show of hands.)
15
DR. NIE:
16
Thank you.
17
(Applause.)
18
(Whereupon, at 5:02 p.m., the meeting was
19
Okay.
Great.
I would like to conclude.
adjourned.)
20 21 22
A Matter of Record (301) 890-4188
View more...
Comments
