Transcript PowerPoint プレゼンテーション - kitasato

Statistical considerations for a
multi-regional trial
Hiroyuki Uesaka, Ph. D
October 28, 2003
Kitasato University-Harvard School of Public Health Symposium
ANA Hotel Tokyo
Acknowledgement
• The multi-regional/national trials were extensively
discussed among
– Mr. Thoru Uwoi (Chairman of JPMA ICH project committee;
Yamanouchi Pharmaceutical Co.,Ltd)
– Dr. Kihito Takahashi (Cordinator of JPMA ICH project committee
efficacy part; Banyu Pharmaceutical Co.,Ltd)
– Mr. Toshinobu Iwasaki (member of JPMA ICH-E5 IWG; Shionogi
Pharmaceutical Co.,Ltd.)
– Dr. Toshimitsu Hamasaki (member of JPMA ICH-E5 IWG; Pfizer
Japan Inc.)
• The speaker would like to thank all of them.
Today’s talk
•
•
•
•
•
General consideration for a multi-regional trial
Primary hypothesis of a multi-regional trial
Testing treatment difference
Sample size allocation and power
Conclusion
Introduction
• There is still considerable gap in the NDA filing time
among regions
• Simultaneous development would be most efficient.
• Multinational study is one possibility in this situation.
• There is an increasing interest in simultaneous
development among regions including Japan as well as the
USA and EU countries.
• However, there is no discussion among regulators,
academia and industries about design and statistical
analysis of a multinational trial.
• This presentation is to give a chance to discuss these topics.
Purpose of a multi regional/national
study
• To establish the efficacy of a drug on a disease where it is
difficult to enroll sufficient number of subjects within a
reasonable time period.
– Rare disease
– A trial whose primary variable is survival or event rate
• To establish the efficacy of a drug among countries where
ethnic differences are assumed negligible
– Multinational trial conducted in EU and USA
– Multinational trial conducted in Asian countries
• To investigate the effect of ethnic differences on response
to a drug
– Bridging study
Multi-regional/national trial to be
discussed here
• Type of a trial
– To establish the efficacy of a drug among countries
where ethnic differences are assumed negligible
• Study design
– Placebo controlled parallel group randomized
study
• Study objective
– To establish efficacy of an investigational
medicinal product against placebo
Prerequisite of a trial
• Assessment of regional differences which may
affect the drug effect
– Factors to be investigated
• Lifestyle, cultural or socioeconomic factors, geographic
environment
• Medical practice, study environment
• Epidemiological characteristics of a disease studied
• Intrinsic factor to produce inter-individual differences
– Actual status of regional difference
– Possible differences in the response and adverse events
• Appropriateness of dose and dose regimen to be
studied
Objective of a trial with a single protocol
• To apply the result of treatment main effect to all
participating regions/countries
But
• It is reasonable to assume some regional
difference in treatment effect
– A design which allows interpretation of the results
– Identify controllable factors
• Influencing baseline variables and patient characteristic
• Subtype of a disease studies
• Severity
– Stratification by controllable factors
Primary hypothesis and its validity
• Primary hypothesis to be confirmed
– The test drug is superior to placebo in an overall mean difference
• Expected result
– Statistically significant difference in the overall mean response
• Applicability and generalizability of the result
– In principle, the primary result is applied to all participating
countries/regions
• Validity of the hypothesis
– Is it possible to assume a priori that the interaction between
treatment-by-region/country is negligible?
• From the information on the existing drugs in the same class or
prior studies
• From pharmacological characteristics of the drug, etiological
or epidemiological nature of the disease
– Confirmation by the study results
Analysis of treatment main effect
-ICH-E9 guideline• Multicenter study
– The main treatment effect may be investigated first using a
statistical model which allows for the center difference but does
not include the term treatment-by-center interaction.
– In the presence of true heterogeneity of treatment effect, the
interpretation of treatment main effect is controversial.
– Alternative estimates of treatment effect may be required, giving
different weights to centers, to substantiate the robustness of
treatment effect.
• Covariate or subgroups
– In most cases, subgroup or interaction analysis are exploratory,…,
they should explore the uniformity of any treatment effect found
overall.
Definition of the treatment main effect
• Difference between the treatment’s overall
means
– A simple average of the mean of each region
– A weighted average of the mean of each region
• The precision of mean difference of each region:
reciprocal of variance of the mean difference at each
region
• Other region specific weight
Mean response of each region
Mean response and difference
Multi-regional/national trial
10
10
9
9
8
8
7
7
6
6
5
5
test drug
4
4
3
3
2
2
Placebo
1
1
0
0
0
1
2
3
Nations
Population means
4
5
Definition of the treatment main effect
Region
Overall
treatment mean
difference
J
A
E
U
Mean (test)
6
5
4
3.5
Mean (control)
0
1
1
0.5
Difference
6
4
3
3
Equal weight
1x6
1x4
1x3
1x3
16/4=4
Sample size
5x6
5x4
45x3
45x3
320/100=3.2
Sample size
45x6
45x4
5x3
5x3
480/100=4.8
Definition of the treatment main effect
M ean response and difference
Multi-regional/national trial
10
10
9
9
8
8
7
7
6
6
(45,45,5,5)
5
5
(equal weight)
4
4
3
3
(5,5,45,45)
2
2
1
1
0
0
0
1
J
2
3
EU
A
Nations
Population mean difference and overall mean difference
4
USA
5
Treatment main effect and power
•
•
•
•
•
Weighted analysis(model without Interaction: A, with interaction B),
Unweighted mean: C, Simple two sample: D
Treatment difference: 4.0, error SD=10 (Effect size =40%)
Significance level of test treatment effect：One-sides 2.5%; Interaction test: 5%
Sample size =100: 80% of the power of the detecting 40% effect size
Region
A
B
C
D
Interaction
4.0
80.5
80.7
80.7
80.4
9.9
45
3.2
61.9
61.9
39.1
61.5
5.8
5
5
4.8
92.1
92.2
39.0
92.1
8.9
10
40
40
3.4
66.9
67.0
61.4
66.5
7.4
40
40
10
10
4.6
90.0
90.1
61.6
89.9
9.6
15
5
40
40
3.5
69.5
69.7
51.8
69.3
9.1
J
A
E
U
Test
6
5
4
6
control
0
1
1
3
Case1
25
25
25
25
Case2
5
5
45
Case3
45
45
Case4
10
Case6
Case6
Treat
ment
difference
Test of treatment main effect
Effect of sample sizes imbalance on the
power of test for treatment main effect
•
•
•
•
•
Weighted analysis(model without Interaction: A, with interaction B),
Unweighted mean: C, Simple two sample: D
Treatment difference: 4.0, error SD=10 (Effect size =40%)
Significance level of test treatment effect：One-sides 2.5%; Interaction test: 5%
Sample size =100: 80% of the power of the detecting 40% effect size
Region
A
B
C
D
Interaction
4.0
80.5
80.7
80.7
80.4
9.9
45
3.2
61.9
61.9
39.1
61.5
5.8
5*9
5*9
3.2
61.3
61.4
61.4
61.2
5.4
45
5
5
4.8
92.1
92.2
39.0
92.1
8.9
5*9
5
5
4.8
92.3
92.7
92.3
92.2
6.2
J
A
E
U
Test
6
5
4
6
control
0
1
1
3
Case1
25
25
25
25
Case2
5
5
45
Case2’
5
5
Case3
45
Case3’
5*9
Treat
ment
difference
Test of treatment main effect
Treatment main effect and power
(A case of no interaction)
•
•
•
•
•
Weighted analysis(model without Interaction: A, with interaction B),
Unweighted mean: C, Simple two sample: D
Treatment difference: 4.0, error SD=10 (Effect size =40%)
Significance level of test treatment effect：One-sides 2.5%; Interaction test: 5%
Sample size =100: 80% of the power of the detecting 40% effect size
Region
J
A
E
U
Test
6
5
4
4.5
control
2
1
0
0.5
Case1
25
25
25
25
Case2
5
5
45
Case3
45
45
Case4
10
Case6
Case6
Treat
ment
difference
Test of treatment main effect
Interaction
A
B
C
D
4.0
80.1
80.1
80.1
79.9
4.6
45
4.0
79.9
79.9
39.5
79.8
5.1
5
5
4.0
80.2
80.2
39.0
80.0
5.0
10
40
40
4.0
80.5
80.4
61.9
80.3
4.9
40
40
10
10
4.0
80.5
80.4
61.2
80.3
4.3
15
5
40
40
4.0
80.4
80.3
51.2
80.3
5.0
Test of treatment by country/region (Null case)
•
•
•
•
•
•
Weighted analysis(model without Interaction: A, with interaction B),
Unweighted mean: C, Simple two sample: D
Treatment difference: 4.0, error SD=10 (Effect size =40%)
Significance level of test treatment effect：One-sides 2.5%; Interaction test: 5%
Sample size =100: 80% of the power of the detecting 40% effect size
No treatment effect
region
Treat
ment
differ
ence
Treatment main effect
A
B
C
D
Intera
ction
effect
J
A
E
U
Test
2
0
-1
-1
control
0
0
0
0
Case1
25
25
25
25
0
2.44
2.52
2.52
2.43
10.04
Case2
5
5
45
45
-0.8
0.56
0.58
2.54
0.56
6.49
Case3
45
45
5
5
0.8
8.28
8.34
2.61
8.33
9.56
Case4
10
10
40
40
-0.6
0.78
0.80
2.44
0.77
8.00
Case5
40
40
10
10
0.6
6.07
6.17
2.59
6.08
9.48
Case6
15
5
40
40
-0.5
0.91
0.91
2.75
0.88
8.34
Summary of testing treatment main
effect
• When there is no interaction effect
– Weighted analysis is more powerful than unweighted
analysis
• Not affected by imbalance in sample sizes among regions
• Statistically more powerful
• When there is interaction effects
– Sample size imbalance among regions may severely
inflate the type I error rate
– To apply the significant result of the treatment main
effect, unweighted mean should be used
A trial to observe the treatment
difference greater than MCSD
• If a region shows treatment difference close to
zero?
– Is it due to too small sample from that region?
– Is it due to too low power to detect regional/country
difference
– Does it suggest regional/country difference
• Points to consider for study design
– Assume that regional/country difference is negligible
– Enroll enough subjects to give a point estimate greater
than MCSD
To get observed mean difference greater
than MCSD assuming no interaction effect
Mean response and difference
Multi-regional/national trial
10
10
9
9
8
8
7
7
6
6
5
5
(equal weight)
4
4
3
3
MCSD=delta/2
2
2
1
1
MCSD=delta/3
0
0
0
1
J
2
A
3
EU
4
USA
Nations
Overall population mean and MCSD (half and one thirds of overall mean)
5
A trial to observe the treatment
difference greater than MCSD
• Assumption
– There is no regional difference in treatment means
• Sample sizes
– 4 regions have common treatment difference: 
– Power of test for treatment main effect 90%at one-sided
2.5% significance level
– Equal sample size at all regions: n
– Probability of getting observed mean difference >MCSD
is 80, 90 and 95%, respectively.
• MCSD＝/280％=> 1.06n, 90％=>2.262n, 95％=>3.62n
• MCSD＝/380％=> 0.67n, 90％=>1.34n, 95％=>2.1n
Sample size
• Determine the target number of subjects to be
enrolled in each region/country
– The method of testing treatment main effect should be
determined prior to sample size estimation
– Equal numbers among regions/countries is most
desirable from statistical perspective
– The number enough to give point estimate which is
greater than minimum clinically significant difference
between treatments in every or a specific
region/country
Conclusion
• Design and statistical method should be discussed
• Method of analysis of treatment main effect
should be pre-defined
• Result of treatment main effect may vary
depending on the definition of treatment main
effect and regional sample sizes
• Equal sample size is important for controlling both
type I and Type II errors
• To give sample size for assuring point estimate
which is greater than MCSD
Backup
Ethnic factors to be considered for study
design
• Definition of a disease and diagnosis
• Epidemiological characteristics of patients and enrolled
subjects
– Distributions of disease subtypes and severity
• Dose and dose regimen of the test drug and control
• Treatment objective, primary variables, timing of
measurement and criteria of efficacy
• Evaluation and reporting safety information
• Medical practices
– Hospitalization/outpatient, patient care, practitioners/specialized
hospital, etc.
• Available concomitant treatments and actual uses
Interpretation of the result
• Is the result applicable to all regions/countries
• What is the significance of the result in the
regional culture, socioeconomic and geographical
conditions, and medical practices and environment
Statistical analysis plan
• Definition of analysis set
• Comparability among regions/countries
–
–
–
–
Attrition of subjects and reasons for attrition
Protocol violations: reasons and frequency
Concomitant medication/treatments, dose and dose regimen
Demographic factors, disease type and severity
• Confirmation of efficacy
– Definition of treatment main effect and statistical model for the
analysis of treatment main effect
– Analysis of treatment by region/country interaction
– Adjustment for covariates
– Important interaction effect between covariate
• Analysis of safety
Evaluation of interaction effect
• Clinically significant size of the interaction effect
– Relative to the size of the mean difference between
treatments
– If there exists an cross-over interaction, evaluate
treatment difference by region/country
• Is non-cross over interaction of no importance?
– The region where there is no significant difference
between treatments.
– Is it necessary for the point estimate of the treatment
difference to be greater than minimum clinically
significant difference
Assessment of the interaction effect
• In case that is no evidence of treatment by
regional interaction effect
– Evidence that there is no interaction effect
• If the test of treatment main effect is significant, testing
treatment by region interaction is performed
• In case some data suggest appreciable interaction
effect
– Non-cross over interaction
• Sample size to show at least the point estimate is greater than
minimum clinically significant difference