A Randomized Adaptive Platform Trial Comparing Multiple Treatments

A Randomized Adaptive Platform Trial Comparing Multiple Treatments for Ebola Virus Disease: A Discussion of Ethical Ramifications Scott Berry 28 Nov 2017 1

2

Treatment Control Classical Trial Disease 3

Tx M Tx 3 Tx 2 Tx 1 Control Platform Trial (perpetual) Disease • Does the tx 1 work in the disease? • Does tx 1 work better than tx 2? • What is the best way to treat the disease? 4

Adaptive Platform Trial Design: Master Protocol – There is one Protocol – The master protocol assigns patients – Protocol is disease focused – No treatment names in protocol – Treatment arm appendices – Potentially evolving arms Master Protocol Tx 1 Tx 2 Tx 3 Tx 4 5

EV-003 Adaptive Platform Design • Reviewed and approved by: – Duke University IRB – University of Sierra Leone ethics committee • Master Protocol dictates trial behavior, each treatment included as an appendix • Multiple Agents – SOC? – Primary & Secondary agents – Combination + Single agents Anti-virals, Sunitinib, Erlotinib, statin, Irbesartan, Azithromycin… • Response Adaptive Randomization (RAR) – Run by a single algorithm – Assigns treatment regimens that are performing better using collection of primary endpoint data • Protocol is built so trial arms evolve (part of the protocol!), trial is perpetual • Endpoint is 14 -day mortality 6

EV-003 Adaptive Platform Design Burn-In Enrollment Analyze Available Data Remove Agent? Accrue More Yes No Revise Allocation Rules Analyze (report) Results Add Agents 7

EV-003 Adaptive Platform Design Burn-In Enrollment Analyze Available Data Remove Agent? Accrue More Yes No Revise Allocation Rules Analyze (report) Results Add Agents 8

Example Trial Regimens Agents 1 2 3 4 1 Agents 2 3 4 9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

Scenario 3 Mean N Truth 1 A 2 B Prob Wins (fixed) 3 C Mean N & Fails 4 D 1 A 2 B 3 C Design Mean Deaths Adapt 49. 4 Fixed 69. 9 4 D 37

FDA Interaction: 3 -Arm Example • Not for evaluation of a single drug for FDA/EMA approval as originally designed • Modifying for regulatory approval with FDA input • 3 arms (SOC plus treatment) + SOC alone – SOC vs. A, B, C [+SOC] • Same trial design! – Modifying “success over SOC” probability – Was 95%. . . Use 99. 8% for the following… 38

Null Arm True N SOC 0. 40 42 Arm True Mean N > SOC > ALL SOC 0. 40 62. 5 A 0. 40 61. 7 0. 025 0. 013 B 0. 40 61. 7 0. 025 0. 013 C 0. 40 61. 7 0. 025 0. 013 Total 0. 40 247. 6 0. 063 0. 037 >SOC 0. 025 Arm True N SOC 0. 40 42 B 0. 40 42 >SOC Arm True N SOC 0. 40 42 0. 025 Arm Deaths N >SOC > ALL Total 0. 40 252 0. 060 ? >SOC 0. 025 39

One Nugget Arm True N SOC 0. 40 42 Arm True Mean N > SOC > ALL SOC 0. 40 29. 4 A 0. 40 21. 6 0. 004 0. 003 B 0. 40 21. 6 0. 003 C 0. 15 81. 1 0. 970 0. 948 Total 0. 268 153. 7 0. 970 0. 948 >SOC 0. 025 Arm True N SOC 0. 40 42 B 0. 40 42 >SOC Arm True N SOC 0. 40 42 C 0. 15 42 0. 025 Arm Deaths N >SOC > ALL Total 0. 358 252 0. 724 ? >SOC 0. 71 40

Varying Good Arm True N SOC 0. 40 42 A 0. 30 42 Arm True Mean N > SOC > ALL SOC 0. 40 26. 9 A 0. 30 25. 0 0. 017 0. 000 B 0. 20 46. 9 0. 202 0. 005 C 0. 10 89. 6 0. 951 0. 664 Total 0. 196 188. 4 0. 970 0. 969 >SOC 0. 147 Arm True N SOC 0. 40 42 B 0. 20 42 >SOC Arm True N SOC 0. 40 42 C 0. 10 42 0. 490 Arm Deaths N >SOC > ALL Total 0. 300 252 0. 953 ? >SOC 0. 892 41

Ethical Ramifications • In a pandemic (LMIC) the struggle over SOC/PBO will be recurring – Trial design functions with or without, and removes as soon as answer is known • In multi-arm setting appropriate RAR improves inference and improves care • Improves global ability to find effective therapies among the ‘universe’ of therapies – Ethical considerations rarely address type III errors – Are 1: 1 trials in a pandemic situation ethical? • Complexity trade-off with improved inference + treatment – Speed of FPI? Speed of 2 nd, 3 rd arm in… 42