Nathalie Frétault Cytel European East User Group Meeting / Friday, October 14, 2011

Use of Bayesian approach in phase II single arm study with possible subgroup selection during an interim analysis

Context

 Drug targeted therapy (T)

  Indication high unmet medical need

 Population 40% of the overall population is expected to have the molecular alterations M+ (that may preclude responsiveness to targeted therapy)

 Primary endpoint: objective response rate

To design a phase II trial in order to establish Proof of Concept

| Presentation Title | Presenter Name | Date | Subject | Business Use Only 2

Phase II Study – Key features

  Primary objective

to establish the activity of T as measured by ORR in « all patients » or in « M+ patients »

 Success criteria •  to observe ORR ≥ 20% in « all patients » •  to observe ORR ≥ 25% in « M+ patients » 20% and 25% are clinically meaningful threshold (go criteria)

 Futility criterion •  to observe ORR < 10 %

Phase II Study – Design

All

patients (n=60)

M+patients (n=24)

Continue « all patients enrollment »

Stop the study

Enroll «M+ patients »

Interim analysis -> futility, population

selection

Single arm study (n=140 patients) Primary endpoint = ORR

Bayesian interim look- Predictive probability

 Decision making process Bayesian framework helps to make a decision at interim analysis based

on - probability to observe the target effect at the end of the trial (probability of success - PPS)

- risk to observe an effect below a defined effect at the end of the trial (futility)

Phase II Study – Bayesian approach

  The decision making process at interim analysis is based on: Predictive Probability that the observed ORR at the end of the study in the “all patients” reach the target 20 %

PPS all pat = Prob [Final Observed ORR All patients ≥ 20% | x , n]

Predictive Probability that the observed ORR at the end of the study in “M + patients” and in “M- patients” is below 10 % -defined weak activity-

Fut m+ = Prob [Final Observed ORR M+ patients < 10%| x , n]

Fut m-= Prob [Final Observed ORR M- patients < 10%| x , n]

Predictive Probability that the observed ORR in the second part of the study in the “M + patients” reach the target 25 %

PPS m+= Prob [Final Observed ORR M+ patients ≥ 25 %| x , n]

Phase II Study – Bayesian approach

  Decision making process at interim analysis :   continue enrolment in all patients

 PPS all pat = Prob [Final Observed ORR All patients ≥ 20 | x , n] ≥ 10 %

 Fut m+ = Prob [Final Observed ORR M+ patients < 10%| x , n] < 20%

 Fut m- = Prob [Final Observed ORR M- patients < 10%| x , n] < 20%

  restart study - M+ patients

 at least one of the above criteria is not met

 PPS m+ = Prob [Final Observed ORR M+ patients ≥ 25 %| x , n] ≥ 10 %

  Stop the study

 Fut m+ = Prob [Final Observed ORR M+ patients < 10%| x , n] > 20%

Phase II Study – Bayesian approach / Beta prior distribution

 θ ~ Beta(a,b) is a convenient choice

 Phase II setting so uncertainty about efficacy profile of T

 Vague prior beliefs about the ORR distribution

 Minimally informative beta distribution priors p ~ Beta(a,b) with prior median 0.10 (10%=clinical threshold for futility)

with a= ln(0.5) / ln(0.10), b=1

Phase II Study – Bayesian approach / Posterior distribution

 At the time of interim analysis, the posterior parameters of the beta distribution of ORR are computing using available data

 The number of responses in the potential future patients follows a Beta distribution with the same parameters

θ/y ~ Beta(a+y*,b+n1-y*) with y*, observed objective responses at interim analysis, n1 the number of patients at interim analysis

 The (posterior) predictive distribution is Beta-Binomial with parameters a+y*,b+n1-y*, n2

with n2 the number of patients to be enrolled in second stage

Phase II Study – Operating characteristics of the design

 Simulations were performed to evaluate the operating characteristics of the design based on several scenarios

•  the probability to stop the study after interim analysis is 97% if actual ORR is 5% (5%,5%)

•  the probability to enroll 140 patients (all patients) is 85.5% if actual ORR is 25% (25%, 25%)

•  the probability to re-start in “M+ patients” is 69% if actual ORR is 25% for “M+ patients” and actual ORR is 9% for “M- patients”

Reference

 Neuenschwander B, Branson M, Gsponer T (2008) Critical aspects of the Bayesian approach to phase I cancer trials. Statist. Med., 27;2420-2439