An adaptive phase II basket trial design

Transcription

1 Novartis Translation Clinical Oncology An adaptive phase II basket trial design Matt Whiley, Group Head, Early Development Biostatistics BBS / PSI One-day Event on Cancer Immunotherapy Basel 15 June 2017

2 An adaptive phase II basket trial design 1. Trial design overview 2. Bayesian hierarchical model: EXNEX 3. Some design issues 4. Last words 2

3 Bayesian adaptive signal finding BAYESIAN ADAPTIVE SIGNAL FINDING INDICATION A INDICATION B INDICATION C INDICATION D INDICATION E INDICATION F INDICATION G INDICATION H INTERIM ANALYSIS OF ORR USING BAYESIAN ADAPTIVE MODEL N=n 1 / INDICATION UP TON=n 2 / INDICATION 3

4 Bayesian adaptive signal finding Multiple indications Small (initial) sample sizes Initial futility decision after response outcome observed for N=n 1 patients How to make optimal use of available data? 4

5 Bayesian adaptive signal finding Making optimal use of available data Independent strata Each stratum has independent P(response) Pooled Common P(response) 5

6 Bayesian adaptive signal finding Making optimal use of available data Independent strata Each stratum has independent P(response) Exchangeable strata Similar P(response) Hierarchical model allows sharing of information Pooled Common P(response) 6

7 Hierarchical models Increased precision Shrinkage towards common mean 7

8 Hierarchical models Increased precision Shrinkage towards common mean 8

9 EXNEX Mixture of hierarchical and stratified approaches Exchangeable vs Independent is not a dichotomy Each strata is assigned a probability of belonging to the exchangeable group That probability updates as data accumulates Strata initially expected to have similar outcome may prove otherwise Equally, strata initially thought dissimilar may prove to have similar outcome Dynamic borrowing of information between indications More borrowing between indications with similar outcome 9

10 EXNEX Mixture model Exchangeable Non-exchangeable ~ HN 0, ~ N, ~ N, ~ N, 1 log ~, ) 10

11 EXNEX model 11

12 EXNEX model 12

13 Design considerations Exchangeability groups Multiple exchangeability groups are possible 13

14 Design considerations Exchangeability Exchangeability can be defined on: Probability of response Appropriate if we expect to see similar response rates in some or all indications Odds ratio of response More suitable for examples where indications have different historical Hypothesis is to see similar improvement over historical rates 14

15 Design considerations Futility decision making Intention is to make an interim stop/go decision for each indication after n 1 patients have been enrolled At each interim data from all patients is used Futility decision rule applied to each stratum P(π i c i ) > ε 15

16 Design considerations Futility decision making INDICATION A INDICATION B INDICATION C INDICATION D INDICATION E INDICATION F INDICATION G INDICATION H Enrolment rates can vary substantially Only strata reaching a minimum enrolment limit are eligible for closure Strategy for interim futility analyses allowing for: Decisions being taken at the appropriate time Operational feasibility! 16

17 Design considerations Interim decision strategies INDICATION A INDICATION B INDICATION C INDICATION D INDICATION E INDICATION F INDICATION G INDICATION H 1. Trigger an interim as each strata reaches N=n 1 With flexibility to shift decision time points based on projected enrolment 17

18 Design considerations Interim decision strategies INDICATION A INDICATION B INDICATION C INDICATION D INDICATION E INDICATION F INDICATION G INDICATION H 2. Trigger initial interim when first strata reaches N=n 1 Subsequent interims held periodically (e.g. quarterly) 18

19 Last words Flexible, adaptive, signal finding framework Simulation studies demonstrate: Appropriate decision making at interim Good control of overall false positive/negative rates High level of complexity Statistically Operationally 19

20 Acknowledgements Shiling Ruan Simon Wandel Satrajit Roychoudhury Beat Neuenschwander Suman Sen Niladri Roy Chowdhury Stuart Bailey 20

21 Thank you