Evaluation of a Multi-regional Trial for Global Simultaneous Drug Development 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.

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

Background

An Example of Japan-CTD

Overall Results Japanese Results

Apixaban(N=9120

)

Warfarin(N=9081)

Apixaban

(N=161)

Warfarin(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) 3 (1.86) 6 (3.43)

Hemorrhagic stroke 40 (0.44) 78 (0.86) 0 (0) 2 (1.14)

Systemic embolism 15 (0.09) 17 (0.10) 0 (0) 0 (0)Number of events (%)

Reproduced from the PMDA review report for Apixaban

Apixaban versus Warfarin in Patients with Atrial Fibrillation: Efficacy Results

Sample size: 1.8% (336/18201)

Number of events: 1.9% (9/477)

Bridging Strategy

Phase 2(Dose-

Response)Data

New region(e.g. Japan)

Foreign region(s)

Phase 3Data

Phase 2(Dose-

Response)Data

Comparison

Extrapolation tothe new region

Typically, a bridging study is a confirmatory study (almost the same sample size as the foreign P2 study).

“Basic Principles on Global Clinical Trials” by MHLW in 2007Question 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….

A Case of a MRCT

MRCTOverall Study PopulationTotal Sample Size:18201(Number of Events: 477)

Japan portionSample Size: 336

(Number of Events:9 )

Of course, we have no sufficient sample size to conduct subgroup analyses within Japan portion.

Comparison(too much focused?)

How should we look at data from a MRCT?

Overall results

Results fromrest of the world

Results from“our nation (ex. Japan)”

Such a comparison may be reasonablein the context of the “Bridging Strategy”

Should we focus on this so much in the new era of MRCTs ?

compare

“Japan vs.” mentality?

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

Homogeneous (consistent)

Investigation of heterogeneity

Heterogeneous (inconsistent)

A Framework to Evaluate Data from a MRCT

3-layer Approach

Overall results (efficacy, safety)

Benefit:RiskAssessment

ForJapan

Benefit:RiskAssessment

ForRegion A

Benefit:RiskAssessment

ForRegion B

Findings from the other studies

Knowledge of the other drugsin the same drugclass, etc.

Layer 1

Layer 2

Layer 3

Layer 2 and Layer 3 haveno prespecified hypothesis, and insufficient power to detect any inconsistency. So, it is important to integrateany available information.

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 ×

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.

A Case Example of Forest Plot

Change from Baseline in FEV1 for COPD patients PMDA review report for Umeclidinium/vilanterol

We can visually look at inconsistency of the treatment effect across regions. Japan is regarded as one region in Layer 2.

A Case Example of Funnel Plot (1)

Produced from the PMDA review report for Aripiprazole(the treatment effects and SEs were derived from summary statistics)

Indonesia

Japan

China

Philippine

Taiwan

Malaysia

Outlier: -27 in Placebo

Treatment betterTreatment difference of CFB in YMRSin bipolar disorder patients

A Case Example of Funnel Plot (2)

FDA Briefing Information, BRILINTA™ (ticagrelor), for the July 28, 2010 Meeting of the Cardiovascular and Renal Drugs Advisory Committee

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

Treatmentbetter

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.)

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?

0.4 0.6 0.8 1.0 1.2

BMD

Ou

tcom

e V

ari

ab

le

Baseline Variable

Baseline distribution of Region A

Baseline distribution of Region B

Treatment effect in Region B

Treatment effect in Region

A

Idealized Schema

Placebo

Study drug

Imbalance across regions in distributions of factors who impact on the treatment effect

How much of the difference can be explained by covariates?

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)

Systematic residual errorsin a specific region still exist?

An Illustrative Example

Predicted vs. Actual Residual plots by regions

Line of “Predicted” = “Actual”

19

An Illustrative ExampleExamination of residuals (gap between observed data and model)

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

Relationship between change in PANSS total score and body weight in schizophrenia trials

Chen YF, et al. (2010). Trial design issues and treatment effect modeling in multi-regional schizophrenia trials. Pharm Stat. 9(3): 217-29.

A total 12585 patients from 33 clinical trials

Active Drug Placebo

Points to consider for partitioning the overall sample size into each region

Region A

Region B

Region C

Total Sample Size

Weight

Distributions ofknown influential factors

Treatment

Effect

Baseline value of the primary

endpoint

Impact

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

Japanese Data

US DataBootstrapSampling

Weig

ht

Available CT or RWD dataat the design stage

Baseline value

*Chen J et al. (2012). An adaptive strategy for assessing regional consistency in multiregional clinical trials. Clin Trials. 9(3):330-9.

A proposal from Chen et al. (2012)* may be also useful during the study execution period.

An Illustrative Example

24

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)

An Illustrative ExampleMedian Distribution of Weight

25 Japanese Sample Size

Media

n o

f W

eig

ht

60 kg

70 kg

80 kg

90 kg

0 100

Summary

Overall results (efficacy, safety)

Benefit:RiskAssessment

ForJapan

Benefit:RiskAssessment

ForRegion A

Benefit:RiskAssessment

ForRegion B

Layer 1

Layer 2

Layer 3

WorldwideCollaborative

Works!