Integration of Knowledge from Late Phase Trials to Support

Download Report

Transcript Integration of Knowledge from Late Phase Trials to Support

Integration of Knowledge from
Late Phase Trials to Support
Regulatory Decisions
Yaning Wang, Ph.D.
Deputy Director
Division of Pharmacometrics
Office of Clinical Pharmacology
OTS/CDER/FDA
ASA NJ Chapter, June 6, 2014
Disclaimer: My remarks today do not necessarily reflect the official views of the FDA
1
Outline
• Pharmacometrics scope at FDA
• Motivation for learning from late
phase trials
• Case studies
• Summary
2
Pharmacometrics Scope at FDA
Tasks
• NDA/BLA reviews
• IND reviews
– Dose-Finding trials
– Registration trials
• QT Reviews
– Central QT team
• Model-based drug development tool
evaluation
• Research
Decisions Influenced
• Evidence of Effectiveness
• Labeling
• Quantify benefit/risk
– Dose optimization
– Dose adjustments
• Trial design
– Disease Models
– Pediatrics
– PBPK
• Knowledge Management
3
Importance of Learning
• The understandable focus of commercial drug development on
confirmation, as this immediately precedes and justifies regulatory
approval, has led, in my view, to a parallel intellectual focus that slights
learning. The predictable result … is that clinical drug development is
often inefficient and inadequate.
• Phase 3/4. Although learning is not the primary goal of this phase, it is a
most important subsidiary one. Because confirming, the primary
purpose of this phase, must not be compromised, one must learn while
confirming, not instead of confirming.
• The most obvious reason is that at no earlier stage is it possible to study
as large a number and as wide a variety of patients as in phase 3/4.
…Thus, important “learning” about the response surface for purposes of
appropriate labeling will either take place at this stage or will not occur
at all.
• The most important reason to learn while confirming … is because
confirmation sometimes fails. …What is needed at this point to plan
future action is a diagnosis: why did the confirmation fail? Only if there
are learning-oriented features to the study can this question be
answered.
Learning versus confirming in clinical drug development, Lewis B Sheiner, CPT, 1997
4
Comments from FDA Officials
•
“Pharmacometrics brings much-needed quantitative, mechanistic reasoning to the
clinical review process. Use of modeling and simulation can fill in the huge gaps left
by even the most comprehensive development program, and can save less
informative programs. As we continue to synthesize clinical data using
pharmacometric techniques, we will advance the science of drug development in
ways we have not anticipated. I will continue to support pharmacometrics in CDER”-Janet Woodcock, Director, CDER
•
“One of the most important benefits of pharmacometric analyses is providing insight
into dose-response/concentration-response by using blood levels collected
(population PK), together with patient effects to relate blood level to response,
providing potentially richer data than the D/R data from the fixed dose D/R studies
(where patients, despite the dose assignment, experience a range of drug
concentrations)”--Robert Temple, Deputy Director for Science, CDER
•
“At NDA review, exploration of exposure response relationships can complement
planned analyses, providing supportive evidence of effectiveness, and helping with
decisions relating to choice of dosing regimens to approve”--Norman Stockbridge,
Director, Division of Cardiovascular and Renal Products
5
Case Studies
• Everolimus
– Non-inferiority margin justification
– Approved based on “totality of evidence”
• Paliperidone
– Approved regimen not studied in phase 3
– Strategy to handle real life scenarios derived from
population PK simulation
6
Case 1: Everolimus (EVR)
• A kinase inhibitor (mammalian Target Of Rapamycin,
mTOR inhibitor)
• Approved as
– Zortress for kidney transplantation
– Afinitor for HER2-negative breast cancer, advanced renal
cell carcinoma, adults with progressive neuroendocrine
tumors of pancreatic origin, adults with renal
angiomyolipoma and tuberous sclerosis complex (TSC), and
subependymal giant cell astrocytom
• Studied extensively for heart transplantation, not
approved due to safety
• New indication: prevention of rejection in liver
transplantation
• Single phase 3 trial with non-inferiority (NI) design
7
Other Products Approved for Prevention of Rejection
in Liver Transplantation
• Tacrolimus (Prograf®) in combination with steroids
– TDM to achieve recommended whole blood
trough concentrations of 5-20 ng/mL
• Mycophenolate Mofetil (MMF, CellCept®) used
concomitantly with cyclosporine and corticosteroids
– Off-label: Tacrolimus used with MMF, most
frequent regimen in the US
• Cyclosporine-A (Sandimmune®, Neoral®) used
concomitantly with azathioprine and steroids
Mechanism of Action and Sites of Action in 3
Signal Model of T-cell Activation
• Signal 1: Antigen-specific stimulation – calcineurin
pathway inhibited by calcineurin inhibitor (CNI,
tacrolimus or cyclosporine)
• Signal 2: Co-stimulation signal not affected by CNI,
everolimus or mycophenolic acid (MPA, active
moiety of MMF)
• Signal 3: Stimulation of activated T-cells by
interleukin-2 (IL-2) binding to IL-2 receptor –
everolimus and MPA act downstream and inhibit
cell proliferation
9
Individual Immunosuppressive Drugs and Sites of Action in
the Three-Signal Model of T-cell Activation (Halloran, NEJM 2004)
10
Trial Design
Target level (ng/ml)
TAC
<3M >3M
tacrolimus
(TAC)
Control
N=243
EVR
<3M >3M
8-12 6-10
One month later*
1378
screened
1147
transplanted
TAC/Steroids
+-MMF
719
randomized
reduced
tacrolimus
(rTAC)+EVR
N=245
3-5
3-5
3-8 3-8
0
3-8 6-10
By randomization,
TAC8ng/ml
eliminated
tacrolimus
(eTAC)+EVR
N=231
3-5
*To minimize wound healing complications, edema, and thrombosis
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3533764/pdf/ajt0012-3008.pdf
11
Efficacy Results
Endpoint
tBPAR/Death/GL
EVR/reduced TAC
N=245
16 (6.7%)
TAC control
N = 243
23 (9.7%)
Difference
-3.0 (-8.7%, 2.6%)
Sponsor’s non-inferiority (NI) margin: 12%
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3533764/pdf/ajt0012-3008.pdf
12
Putative Placebo and M1
What if…
Placebo
Reduced TAC
(Putative placebo)
M1
Reduced TAC+EVR
M2
Standard TAC
13
Planned TAC Levels
(Defined targets without overlap, 50%)
Tacrolimus (TAC) trough level (ng/mL)
14
Control arm: standard TAC
Putative placebo arm: reduced TAC
12
10
10
8
6
6
5
4
3
2
0
0
50
100
150
200
Days after randomization
250
300
350
14
M2 @ Planning Stage
• Assuming target TAC levels achieved and M1
realized as planned
• Sponsor selected M2 (non-inferiority margin)
to be 12% (tBPAR/Graft loss/Death)
• Extensive analyses to justify NI margin due to
the lack of historical data under the special
trial design
• Study powered based on 12% NI margin
15
Actual TAC Levels
(Wide ranges with significant overlap)
10
6
5
3
16
Observed Median TAC Exposure
(Clinically meaningful separation, but not as large as planned)
8
4
17
M1 Re-assessment
• What would be M1 under the actual TAC
levels (wider than TDM targets and smaller
reduction than planned)?
• Given the wide range of TAC exposure in the
TAC control arm, we can derive M1 from TAC
control arm alone
– Exposure-response analysis
– Estimate the difference in efficacy between two
TAC exposure levels (profiles)
• Standard TAC
• Reduced TAC (without everolimus)
18
Atypical Scenario
1. Data source
2. Analysis
3. M1
Typical NI study
Historical data
Meta-analysis
Point estimate (lower, upper)
Everolimus Phase 3 Study
Current data from control arm
Exposure-response
Point estimate (lower, upper)
FDA NI Guidance:
The point estimate…may be closer to the true effect of the active control
It (lower bound of the 95% CI ) is, on average, a low estimate of the effect of the
drug, and is “conservative” in that sense.
Justifications for conservative estimate of M1:
Similarity of the current NI trial to the historical studies: “constancy assumption”
Historical evidence of sensitivity to drug effects (HESDE)
19
Response
dnorm(TAC, mean = 9, sd = sd(dsn1$TAC))
Concept
0.25
0.5
0.15
2
0.05
-1
0
12
0.4
0.05
-2
1
2
xrange
-1
0
10
TAC
12
14
TAC exposure-efficacy
data from TAC control
arm only
-1-2
0.1
0.2
M1
8
xrangexrange
01
0.3
6
0.05
yrange
TAC exposure data from TAC reduced+everolimus arm
0.25
0.0
-2
yrange
0.05
6
8
10
Exposure
12
14
20
0.25
yrange
Observed Exposure-Response
(TAC Control Only)
10
5
Below 25th percentile: 10/19
0
TAC level (ng/mL)
15
Above 75th percentile: 3/19
3
4
6
12
14
21
28
29
31
39
52
Days after randomization
58
76
136
153
174
263
21
Exposure-Response Model
• Data from TAC control arm only
ITT data
Exposure-response data
Patients with
event
23
22 (95.6%)
Patients without
event
220
209 (95.0%)
Total
243
231 (95.1%)
• TAC predicted (prior day) daily average
exposure as time-dependent covariate
• Efficacy endpoint: time-to-event
(tBPAR/GL/D)
• Cox proportional hazard model
Estimate (95%CI)
Hazard ratio* 0.68 (0.55, 0.84)
*: One unit increase in TAC exposure
p-value
0.0004
22
TAC daily average level (ng/mL)
(mean+-SD)
TAC daily average level (ng/mL)
Two Exposure Profiles
23
Steps to Derive M1
0.5
Replicate1
2
1 0
xrange xrange
2 1
0.3
0.2
97.5th percentile
Upper 95% CI
-1 -2
0.1
0 -1
Δ1
0.25
0.0
0.05
Response
0.4
16.5% (2.4%, 26.6%)
-2
yrange
0.05
0.25
yrange
6
8
10
12
14
Exposure
1000 Δi
…
Exposure
240 patients/arm
Median
Point estimate
2
0.5
Replicate1000
1
20
xrange
0.3
1 -1
0 -2
0.25
-2
-1
yrange
0.05
0.0
xrange
0.2
0.1
Lower 95% CI
Δ1000
0.05
0.25
yrange
Response
0.4
2.5th percentile
6
8
10
Exposure
12
Exposure
14
24
Sensitivity Analyses
• TAC concentration not randomized
– Impact of confounding factors
•
•
•
•
•
•
Only on-treatment data used
One patient with tBPAR when TAC=0
PK outliers
PK correlation over time
Bootstrap runs with underestimated M1
Linear relationship between TAC
concentration and log-hazard
25
TAC Concentration Not Randomized
0.5
Over-estimate exposure-response slope and
M1
0.0
y3
y Da
35
0.4
0.6
ent b
v
e
5
.
Pevent
Probability
Day 335
r.4ob 0by
bility0.0
a
a
b
0
bilit
Pro
y
0.3 0.1
0.0
0.1
0.2
0.3
0.4 eve0.5
0.2
nt b 0.6
0
.
2
yD
0.1
ay 3
0.3
3
5
0.6
(Two ways of confounding with opposite effects)
Under-estimate exposure-response slope
and M1
X
M1
14
X
12
10
6
8
6
8
8
6
T AC
le
T AC
g/mL
n
(
l
ve
le 10
0
TAC1level
(ng/mL)
)
12
14
26
Confounding Factors
• HCV positive at screening (Yes/No)
Slope
Estimate
0.045
p-value
0.006
Higher TAC, HCV positive
• eGFR at randomization (continuous)
Slope
Estimate
-0.006
p-value
0.099
Higher TAC, lower eGFR
• MMF use prior to randomization (Yes/No)
Slope
Estimate
0.017
p-value
0.4
Higher TAC, MMF use
All favor a flatter ER relationship and conservative M1
27
Results
Scenario
Original (not adjusting for covariates)
M1
16.5%
Including only MMF patients
19.7%
Excluding 1 patient with tBPAR when
TAC=0
16.4%
Excluding 13 patients with outlier TAC
observations
17.1%
Including off-treatment data
13.7%
28
Rationale for M1=13.7%
• Current data without constancy concern from
historical data
• Not for designing another study
• By definition, the point estimate is the most likely
estimate of M1 in this study
• Multiple strategies to ensure a conservative
estimate
–
–
–
–
Not adjusting for confounding factors
Median of all bootstrap runs
Highly correlated PK to minimize M1
Taking the minimum of all sensitivity analyses
29
Summary of Case 1
• Innovative method can be applied to derive NI
margin under special conditions
• Post hoc learning analysis served as the
foundation for M1 and the approval decision
• Valuable knowledge can be learned from
confirmatory trials
• Rigorous learning analyses and application of
integrated knowledge led to
– Unnecessary trials avoided
– Faster access of medication for patients
Case 2: Paliperidone
• Indication: schizophrenia
• Extended-Release tablet is already approved
(QD regimen)
• New monthly long acting injection
formulation
• Regimens studied in phase 3 trials:
– 25 mg, 50 mg, 100 mg, 150 mg (day 1, 8, 36, 64)
25 mg (days 8, 36, and 64)
– 150 mg (day1)+
100 mg (days 8, 36, and 64)
150 mg (days 8, 36, and 64)
• Proposed regimen:
– 150 mg (day1), 100 mg (day 8) and 75 mg (monthly)
31
Change from Baseline in
PANSS Total Score
0
Benefit/Risk Assessment
-2
• N160/arm
• All active arms better than placebo
-4
-6
P<0.05
-8
-10
P<0.001
-12
P<0.001
-14
0
25
50
75
100
125
150
Maintenance Dose (mg)
Safety: one death at 150 mg and dose-dependent increase
in body weight and serum prolactin levels
32
15
20
Median Cmax (6 mg QD ER at steady state)
5
10
Median Cmin (6 mg QD ER at steady state)
Dosing Interval at Steady State
(75 mg im. Q 4week)
0
Concentration [ng/mL]
25
30
Additional Support for 75 mg
19
20
21
22
Time [Week]
23
24
25
33
Real Life Challenges
• Dosing window for the 2nd dose
• Dosing window for the monthly maintenance
dose
• What to do after missing a dose
• Switching from ER tablet or other
antipsychotics
• Dosing regimen for patients with renal
impairment
34
PK Simulation for Dosing Window
35
Simulation Based Solutions
• Dosing window for the 2nd dose:  4 days
• Dosing window for the monthly maintenance dose: 
7 days
• What to do after missing a dose
– 2nd dose: 3 re-initiation regimens
– maintenance dose: 3 re-initiation regimens
• Switching from ER tablet or other antipsychotics
– Different maintenance doses and starting times
• Dosing regimen for patients with renal impairment
– 100 mg (day 1), 75 mg (day 8) and 50 mg (monthly)
36
Summary for Case 2
• Unstudied regimen was approved
• FDA’s alternative strategy and individualized
maintenance dose were accepted by the sponsor
• PK simulation led to recommendations for dosing
window, strategy for handling missing dose,
switching from prior treatments, dosing regimen for
special patients
• All these model based recommendations are
included in the product label
37
Conclusion
• Valuable knowledge can be learned from
confirmatory trials
• Rigorous learning analyses and application
of integrated knowledge led to
– Unnecessary trials avoided
– Faster access of medication for patients
– Rational dosing regimen
– Strategy to handle real life issues
• Routine practice in drug development and
regulatory review
38
Acknowledgment
•
•
•
•
•
•
•
•
•
•
•
•
Yoriko Harigaya, Pharm.D.
Philip Colangelo, Pharm.D., Ph.D.
Marc Cavaillé-Coll, M.D., Ph.D.
Joette M. Meyer, Pharm.D.
Hongling Zhou, Ph.D.
Karen Higgins, Sc.D.
Renata Albrecht, M.D.
Hao Zhu, Ph.D.
Kofi Kumi, Ph.D.
Raman Baweja, Ph.D.
Jing Zhang, M.D.
Scientists from the sponsors
39