ExL Pharma’s Multiple Comparisons in Clinical Trials Highlights January 25-27, 2010 Rockville, MD
ExL Pharma’s Multiple Comparisons in Clinical Trials HighlightsJanuary 25-27, 2010Rockville, MD
Subgroup analyses in Pharmaceutical Development- Must we always adjust
for multiplicity?
What Does the FDA Say About Subgroup Analyses?•Not Much!•The need for conducting subgroups
analyses is acknowledged •No methodological guidance is provided•Subgroup analyses are lumped together
with other multiplicity issues
FDA Position on Subgroup Analyses
• Subgroups of interest must be pre-specified in the protocol• Inferences about subgroups following the ITT analysis is
subject to multiplicity Type I error adjustment• Generally, subgroup analyses are exploratory only
▫ Hypotheses generation▫ Identify heterogeneity w.r.t baseline, demographic, geographic
variables• Generally, NDA approval requires significance of the
primary endpoint in ITT• Significance in pre-specified subgroup is not sufficient
Fundamental Question:
Do the problems associated with subgroup analyses raise multiplicity
issues?
MultiplicityMultiplicity issue arises when a single inference is based on
multiple repeated testing▫ Interim analyses (multiple looks)▫ Multiple comparisons (e.g. multiple doses of a drug)▫ Multiple endpoints
Error to be controlled = Family-wise Error Rate
Multiple Comparisons Paradigm
Regulatory claim:Drug is efficacious Patient population A
Stat Decision Rule:Drug is efficacious if
Sig. on V1, OR Sig. on V2, ORSig. on V3, etc.
Testing
V1 Sig?
V2 Sig?
V3 Sig?
Yes
Yes
Yes
EFFICACIOUS
Control “family-wise” Error Rate
Subgroup AnalysisExample:
• Placebo-controlled global trial of a new ACE inhibitor
• Sponsor is interested in investigating the drug’s efficacy in African patients
• Randomization stratified by country• Primary efficacy variable – DiPB• Target population – Patients with moderate
hypertension
Analysis Strategy
•Test for efficacy in the ITT•Proceed to test in the subgroup of
African Patients
Possible Outcomes
P 0.05P > 0.05
P 0.05P > 0.05P 0.05P > 0.05
Test in SubgroupTest in Subgroup
Test in ITT
A B C D
Inferences
P 0.05P > 0.05
P 0.05P > 0.05P 0.05P > 0.05
Test in SubgroupTest in Subgroup
Test in ITT
Treatment Selection in Multi-Armed Trials Using Confirmatory Adaptive Designs
The term adaptive
• Adaptive randomization
• Adaptive test selection
• Adaptive dose selection
• Bayesian adaptive designs
• Confirmatory adaptive designs
13
Multi-armed designs• Consider many-to-one comparisons, e.g., G treatment
arms and one control, normal case.
• In an interim stage a treatment arm is selected based on data observed so far.
• Not only selection procedures, but also other adaptive strategies (e.g., sample size reassessment) can be performed.
• Application, e.g., within an “Adaptive seamless designs” using the combination testing principle, but investigation of more than one dose in phase III is also encouraged.
14
Standard2 phases
AdaptiveSeamlessDesign
Plan & DesignPhase III
Dose Selection
Learning
A
B
C
DControl
Confirming
Learning, Selecting and Confirming
Plan & DesignPhase IIb
Plan & DesignPhase IIb and III
Adaptive seamless designs
15
A
B
C
DControl
ExampleComparison of three test procedures
•Inverse normal Dunnett
•Pure conditional Dunnett
•Separate stage conditional Dunnett
16
Comparison of the three procedures
17
Design: two-stage, = 0.025 one-sided, u1 = , u2 = 1.96 linear dose-reponse relationship with drift
120 i.e., ,20;20 220
13
12
11
10 Nnnnnnn S
140 i.e., ,20;20 2220
13
12
11
10 21
Nnnnnnnn SS
- always select the two best:
- select all:
160 i.e., ,20;20 23
22
21
20
13
12
11
10 Nnnnnnnnn
- always select the best:
Consider three selection procedures:
18
19
20
The comparison shows that • the conditional second-stage Dunnett test performs
best
• it is identical with the conventional Dunnett test if no adaptations were performed
• becomes complicated if, e.g., ▫ allocation is not constant
▫ variance is unknown
• the inverse normal technique is not optimum but enables early stopping and more general adaptations
• is straightforward if, e.g., ▫ allocation is not constant
▫ variance is unknown
21
Still have any questions? For additional information on ExL Pharma’s Multiple Comparisons in
Clinical Trials Conferences, please visit www.exlpharma.com