1. Norisuke Kawai Clinical Statistics, Pfizer Japan Inc.
Evaluation of a Multi-regional Trial for Global Simultaneous Drug Development
Norisuke Kawai
Clinical Statistics, Pfizer Japan Inc.

2. Agenda Background 3-layer approach
Statistical approaches for exploring regional heterogeneity of the treatment effect
Points to consider for partitioning the overall sample size into each region
Summary

3. Background An Example of Japan-CTD
Apixaban versus Warfarin in Patients with Atrial Fibrillation: Efficacy Results
Overall Results
Japanese Results
Apixaban
(N=9120)
Warfarin
(N=9081)
(N=161)
(N=175)
Primary endpoint:
Stroke or systemic embolism
212 (2.32)
265 (2.92)
3 (1.86)
6 (3.43)
Ischemic or uncertain type of stroke
162 (1.78)
175 (1.93)
Hemorrhagic stroke
40 (0.44)
78 (0.86)
0 (0)
2 (1.14)
Systemic embolism
15 (0.09)
17 (0.10)
Number of events (%)
Sample size: 1.8% (336/18201)
Number of events: 1.9% (9/477)
Reproduced from the PMDA review report for Apixaban

4. Bridging Strategy Phase 3 Data
Typically, a bridging study is a confirmatory study (almost the same sample size as the foreign P2 study).
Phase 3
Data
Extrapolation to
the new region
Phase 2
(Dose-Response)
Data
Phase 2
(Dose-Response)
Data
Comparison
New region
(e.g. Japan)
Foreign region(s)

5. “Basic Principles on Global Clinical Trials” by MHLW in 2007
Question 6
When conducting an exploratory trial like a dose-finding study or a confirmatory trial as a global clinical trial, how is it appropriate to determine a sample size and a proportion of Japanese subjects?
Answers
… A global trial should be designed so that consistency can be obtained between results from the entire population and the Japanese population, and by ensuring consistency of each region, it could be possible to appropriately extrapolate the result of full population to each region….

6. Overall Study Population
A Case of a MRCT
MRCT
Overall Study Population
Total Sample Size:18201
(Number of Events: 477)
Comparison
(too much focused?)
Japan portion
Sample Size: 336
(Number of Events:9 )
Of course, we have no sufficient sample size to conduct subgroup analyses within Japan portion.

7. How should we look at data from a MRCT?
compare
Overall results
Results from rest of the world
Results from “our nation
(ex. Japan)”
“Japan vs.” mentality?
Such a comparison may be reasonable
in the context of the “Bridging Strategy”
Should we focus on this so much in the new era of MRCTs ?

8. Homogeneous (consistent) Heterogeneous (inconsistent)
Objectives of MRCTs
Primary objective
Confirm efficacy and safety of the study drug in the overall study population
A key secondary objective
Evaluate influential ethnic factors on efficacy and safety of the study drug, which includes investigating whether there is regional heterogeneity
Investigation of heterogeneity
Homogeneous (consistent)
Heterogeneous (inconsistent)

9. A Framework to Evaluate Data from a MRCT 3-layer Approach
Overall results (efficacy, safety)
Findings from the other studies
Knowledge of the other drugs in the same drug class, etc.
Layer 2
Layer 2 and Layer 3 have
no prespecified hypothesis, and insufficient power to detect any inconsistency. So, it is important to integrate
any available information.
Benefit:Risk
Assessment
For
Japan
Benefit:Risk
Assessment
For
Region A
Benefit:Risk
Assessment
For
Region B
Layer 3

10. 3-layer Approach
In Layer-1, we look at the overall results of efficacy and safety.
In Layer-2, we conduct comprehensive and rigorous analyses to explore influential factors on efficacy or safety.
Is there inconsistency in efficacy or safety in a particular subgroup?
Is regional heterogeneity observed? etc.
In Layer-3, given the results from Layer-1 and Layer- 2, we consider Benefit:Risk for each region. ×
DO NOT jump

11. In Layer-2
How do we explore regional heterogeneity of the treatment effect?
Graphical presentations
Forest plot, funnel plot, etc.
Modeling approaches
If we find a regional difference by looking at graphical presentations, modeling approaches are useful to investigate how much of the difference can be explained by covariates.

12. A Case Example of Forest Plot
We can visually look at inconsistency of the treatment effect across regions. Japan is regarded as one region in Layer 2.
Change from Baseline in FEV1 for COPD patients
PMDA review report for Umeclidinium/vilanterol

13. A Case Example of Funnel Plot (1)
Outlier: -27 in Placebo
Malaysia
Taiwan
China
Japan
Philippine
Indonesia
Treatment better
Treatment difference
of CFB in YMRS
in bipolar disorder patients
Produced from the PMDA review report for Aripiprazole
(the treatment effects and SEs were derived from summary statistics)

14. A Case Example of Funnel Plot (2)
The PLATO trial: a MRCT that compared ticagrelor and clopidogrel for the prevention of cardiovascular events in 18,624 patients admitted to the hospital with an acute coronary syndrome, with or without ST-segment elevation
FDA Briefing Information, BRILINTA™ (ticagrelor), for the July 28, 2010 Meeting of the Cardiovascular and Renal Drugs Advisory Committee
Treatment
better

15. If we find regional heterogeneity
A next question is “Are the observed regional differences Real or Not?”
Possible answers (they are mixed in practice)
Imbalance across regions in distributions of intrinsic ethnic factors who impact on the treatment effect
We could explain it by available data in the MRCT
Extrinsic ethnic factors impact on the treatment effect
Play of chance
Others (outliers, treatment compliance, dropout rates, etc.)

16. Modeling Approaches
If we find certain regions look “different” from others, as a next step, we need to examine what may have caused the difference.
We can use statistical models to examine the difference.
How much of the difference can be explained by covariates?
Are there observations that cannot be explained, such as outliers?
Idealized Schema
Imbalance across regions in distributions of factors who impact on the treatment effect
Treatment effect in Region A
Treatment effect in Region B
Outcome Variable
Placebo
Baseline distribution of Region A
Baseline distribution of Region B
Study drug
Baseline Variable

17. How much of the difference can be explained by covariates?
Systematic residual errors
in a specific region still exist?
Predicted values by the model
Actual values
Line of “Predicted” = “Actual”
e.g., Predicted value = f (treatment group, baseline value, body weight)
Residuals
= unexplained by the model
Covariates
(influential factors)

18. An Illustrative Example
Line of “Predicted” = “Actual”
Residual plots by regions
Predicted vs. Actual

19. An Illustrative Example Examination of residuals (gap between observed data and model)

20. Points to consider for partitioning the overall sample size into each region
Consistency perspective: Minimize the chance for observing apparent differences across regions when the treatment effect is truly uniform across the regions.
e.g. Method 1 or Method 2 in “Basic Principles on Global Clinical Trials” by MHLW
Another perspective: Be able to evaluate influential factor(s) on important efficacy/safety endpoints, considering distributions of known influential factor(s) in each region

21. Relationship between change in PANSS total score and body weight in schizophrenia trials
A total patients from 33 clinical trials
Active Drug
Placebo
Chen YF, et al. (2010). Trial design issues and treatment effect modeling in multi-regional schizophrenia trials. Pharm Stat. 9(3):

22. Baseline value of the primary endpoint
Points to consider for partitioning the overall sample size into each region
Distributions of
known influential factors
Total Sample Size
Region A
Weight
Impact
Region B
Treatment Effect
Region C
Baseline value of the primary endpoint

23. Available CT or RWD data
Points to consider for partitioning the overall sample size into each region
Need to check distributions of known (or potentially) influential factors by simulating various scenarios for partitioning the total sample size into individual regions at the design stage
Available CT or RWD data
at the design stage
Japanese Data
Weight
US Data
Bootstrap
Sampling
Baseline value
A proposal from Chen et al. (2012)* may be also useful during the study execution period.
*Chen J et al. (2012). An adaptive strategy for assessing regional consistency in multiregional clinical trials. Clin Trials. 9(3):330-9.

24. An Illustrative Example
Target Patients: hypercholesterolemic patients
Primary endpoint: Change from baseline (CFB) of LDL-C in 12W
Known influential factors on the primary endpoint: Weight (and baseline LDL-C value)
Regions: US and Japan
Sample size: Total 100 (50/group)

25. An Illustrative Example Median Distribution of Weight
90 kg
80 kg
Median of Weight
70 kg
60 kg
100
Japanese Sample Size

26. Summary Worldwide Collaborative Works! Layer 1 Layer 2 Layer 3 For
Overall results (efficacy, safety)
Worldwide
Collaborative Works!
Layer 2
Benefit:Risk
Assessment
For
Japan
Benefit:Risk
Assessment
For
Region A
Benefit:Risk
Assessment
For
Region B
Layer 3