Background to Adaptive Design Nigel Stallard Professor of Medical Statistics Director of Health Sciences Research Institute Warwick Medical School [e-mail]

1
Background to Adaptive Design Nigel
StallardProfessor of Medical StatisticsDirector
of Health Sciences Research InstituteWarwick
Medical Schooln.stallard_at_warwick.ac.uk
2
Outline

• What are adaptive designs?
• Types of adaptive designs
• Advantages and challenges
• Advantages
• Statistical challenges
• Logistical challenges
• Example adaptive seamless design in MS
• Adaptive seamless phase II/III clinical trial
• Evaluation of design options
• Implications for research funders

3
1. What are adaptive designs?
Adaptive design Use interim analyses to assess
accumulating data Adapt design for remainder of
trial
4

• Types of adaptive designs
• Possible adaptations can include
• - Up-and-down type dose-finding
• - Adaptive randomisation (rand. play-the-winner
  etc.)
• - Sample size re-estimation based on nuisance
  parameter estimates
• - Sample size re-estimation based on efficacy
  estimates (including self-designing trials)
• - Early stopping for futility
• - Early stopping for positive results
• - Selection or modification of subgroups or
  treatments
• - Stopping for safety or logistical reasons

5

• Focus on methods for confirmatory trials
• - Sample size re-estimation based on nuisance
  parameter estimates
• - Sample size re-estimation based on efficacy
  estimates (including self-designing trials)
• - Early stopping for futility
• - Early stopping for positive results
• - Selection or modification of subgroups or
  treatments

6
2. Advantages and challenges

• Advantages
• Efficiency
• - reach conclusion with (on average) smaller
  sample size
• - avoid wasting further resources on trials
  unlikely to yield useful results
• - ensure trials are appropriately powered
• - focus resources on evaluation of most
  promising treatments
• Ethics
• - use right number of right patients on right
  treatments

7

• Statistical challenges
• Type I error rate
• E.g. Interim analysis in phase III trial to
  compare two arms
• Significant at 5 level stop trial
• Not significant continue with trial
• Probability of false positive at interim analysis
  5
• Overall probability of false positive gt 5
• Other adaptations may also increase type I error
  rate
• e.g. sample size increased after less promising
  interim data

8

• Treatment effect estimation
• Trial may stop because of extreme positive data
• Conventional estimates will overestimate true
  treatment effect
• Specialist statistical methodology is required

9

• Logistical challenges
• Up-front planning
• Designs may be more custom-made
• Design properties may need to be assessed prior
  to trial
• e.g. by simulation studies
• Management of unblinded data
• Breaking of blind may lead to bias, limit
  recruitment or lead to lack of equipoise
• Release of information and decision-making
  process needs to be carefully considered

10

• Conduct of interim analyses
• Timely and accurate data management required
• Trial modification
• May require ethical approval
• May require revision of patient information
  sheets
• Randomisation and drug supply needs careful
  consideration

11
3. Example Adaptive seamless design in MS

• Setting
• Primary/secondary progressive Multiple Sclerosis
• Challenges
• No current effective disease modifying therapy
• Several potential novel drug therapies to
  evaluate
• Outcomes
• Phase II Short-term MRI data (6-12 months)
• Phase III Long-term disability scales (2-3
  years)
• Clinical trials are very long and costly

12
Adaptive seamless phase II/III clinical trial
Experimental treatments T1, ..., Tk Control
treatment T0 Select treatment(s) at
interim analysis using MRI data Final analysis
uses combination test to control overall type I
error rate allowing for selection/multiple
testing
13

• Evaluation of design options
• Choice of design options
• sample size, timing of interim analysis,
• decision rule for selecting arms
• Simulation study
• estimate power to reject at least one false null
  hypothesis
• estimate selection probabilities
• based on wide range of assumptions
• treatment effect on primary outcome
• treatment effect on short-term outcome
• correlation between outcomes
• from extensive literature review
• 10,000 simulations for each of gt 25,000 scenarios

14

• Example simulation results
• 3 experimental treatments
• Interim analysis
• midway early
• one effective treatment
• one effective treatment
• one partly effective

15
4. Implications for research funders

• Advantages
• Adaptive designs could lead to efficiency gains
• Resources are targeted most effectively
• Challenges
• Need to ensure appropriate methodology is used
• Additional methodological development may be
  needed
• May need to allow extra time/funding for design
  work and evaluation
• More flexible trials may require more flexible
  funding model