1. Adaptive Design Workshop ISCTM Autumn Meeting 2012 Marina del Rey, California

2. Working Group Members Radhi Abdulnabi Vlad Dragalin Jean Dries Ginger Haynes Judy Kando Justine Kent Pilar Lim John March Olga Marchenko Tom Parke Jose Pinheiro Joanne Severe Ibo Turkoz Norris Turner Peter Zhang

3. Study Proposal Study Design: – Multicenter – Double-blind, randomized, placebo-controlled – Parallel-group – Dose-response study in male and female subjects with schizophrenia – Multiple fixed doses of Compound X as a monotherapy after 6 weeks of treatment in subjects with schizophrenia Primary objective: – Evaluate the efficacy via change from baseline in the total Positive and Negative Syndrome Scale (PANSS) score – Target: minimal effectiveness = 10 point difference from placebo (SD = 20)

4. Fixed Phase IIa & IIb Design N=75Placebo50mg BID150mg BID N=250Placebo25mg BID50mg BID150mg BID STAGE 1 Proof of Concept STAGE 2 Dose Finding N=250Placebo100mg BID150mg BID 50mg good in 2a 50mg poor in 2a 150mg included in both, as insurance

5. Adaptive Seamless IIa/b Design N=75Placebo50mg BID150mg BIDN=250Placebo25mg BID50mg BID 100mg BID 150mg BID Fixed allocation to control Modify probability of randomisation to favour arm most likely to be Minimum effective Dose (MED) Two doses initiallyOpen up all other arms for ‘IIb’ stage First interim – possible stop for futility Subsequent interims, possibly stop for success or futility, otherwise update randomisation

6. Compare 2 Designs 1.Fixed design of 2 separate trials, a 2a and 2b. a)Test 2 doses in 2a to determine whether to continue and which doses to test in 2b b)Test 2 doses in 2b – either low set with 150 mg (25mg, 50mg & 150mg) or high set (100mg & 150mg) c)Choose minimum effective dose (MED) from 2b to take to phase 3 2.Adaptive design of ‘seamless 2a/2b’, a)Start testing 2 doses b)Then open up other doses c)Adapt allocation to target ‘MED’ d)Allow early stopping for futility or success

7. Adaptive Design Pros Reduce regret – Running trial too long (data conclusive) – Treating wrong patients – Using wrong dose / regimen – Stopping too soon (data inconclusive) Use models Combine data Evaluate more options A good adaptive design is almost always cautious in its adaptations Cons Making many decisions during a trial can increase the chance of making a wrong decision – Abandoning a dose too soon Designs can be complicated Time and cost of running simulations Operational challenges (availability and distribution of doses) Need electronic data capture Unsure of acceptance by key individuals/regulatory bodies Barry DA. Nature Reviews 2006 FDA Draft Guidance for Industry on Adaptive Design Clinical Trials for Drugs and Biologics. Feb 2010

8. ADAPTIVE VS. TRADITIONAL DESIGN

9. Dose Curve Scenarios Placebo 25 mg 50 mg 100 mg 150 mg

10. Illustration Example simulation from ‘middle dose’ scenario. It will stop early (as 67% of them do) and select the correct dose (as 58% of them do). This is the situation at the first interim, 25 subjects have been allocate to each of control and the 50mg and 150mg doses.

11. 2 nd Interim 1/3 rd of new subjects are allocated to control. Most of the rest to the new doses now available and one to 50mg as it looks promising. Notice how large the error estimates are around the doses with no data yet.

12. 3 rd Interim New subjects P(MED) Control:+5 - 25mg+10.10 50mg+20.19 100mg+40.30 150mg+10.25

13. 4 th Interim New subjects P(MED) Control:+8 - 25mg+20.08 50mg+30.08 100mg+60.37 150mg+20.33

14. 5 th Interim New subjects P(MED) Control:+4 - 25mg+10.14 50mg+00.24 100mg+90.53 150mg+00.07

15. 6 th Interim New subjects P(MED) Control:+5 - 25mg+30.14 50mg+20.27 100mg+60.56 150mg+10.02

16. 7 th Interim New subjects P(MED) Control:+5 - 25mg+20.11 50mg+20.25 100mg+3*0.53 150mg+30.10 * Probability of allocation still favours 100mg, but randomisation is probabilistic.

17. 8 th Interim New subjects P(MED) Control:+4 - 25mg+10.12 50mg+20.22 100mg+20.60 150mg+10.05

18. 9 th Interim – trial stops New subjects P(MED) Control:+3 - 25mg+50.07 50mg+00.23 100mg+40.65 150mg+10.05

19. Final analysis 6 weeks later Total subjects P(MED) Control:64 - 25mg200.02 50mg370.12 100mg390.79 150mg340.02

20. Final estimate of MED

21. ScenarioFail PoCFail before Cap Fail at CapSuccessMean Sample Size Null - fixed0.8140.00.1680.018118 / 325 adaptive0.3790.5630.0280.007125 / 273 Weak0.5860.00.2140.200142 / 325 adaptive0.1430.4480.0690.197169 / 288 High Dose0.1830.00.1940.623204 / 325 adaptive0.0170.0460.0020.898155 / 248 Middle Dose0.1480.00.1510.701201 / 325 adaptive0.0060.0120.0010.968111 / 241 Low Dose0.1170.00.1960.687232 / 325 adaptive0.0060.00900.974124 / 247 Peak 50 mg0.2590.00.1870.554180 / 325 adaptive0.0570.130.020.698152 / 252 Peak 100mg0.2450.00.2150.540192 / 325 adaptive0.0250.0300.911114 / 253

22. Probability of early failure

23. Probability of late failure

24. Probability of success

25. Mean sample size

26. ScenarioControl25mg50mg100mg150mg High Dose-16-17-18-20-28 fix’d P(MED)0.000 0.005 0.0710.542 adpt P(MED)0.0000.0170.0240.1510.701 Middle Dose-16-17-22-27-28 fix’d P(MED)0.000 0.0200.1140.3680.119 adpt P(MED)0.0000.0380.2930.5770.058 Low Dose-16-21-25-27-28 fix’d P(MED)0.000 0.2040.1960.2070.080 adpt P(MED)0.0000.2060.5250.2010.039 Peak 50 mg-16-21-26-22-18 fix’d P(MED)0.000 0.0370.5020.0040.012 adpt P(MED)0.0000.2030.4780.0070.000 Peak 100mg-16-20-24-28-24 fix’d P(MED)0.000 0.0540.1430.1540.190 adpt P(MED)0.0000.1360.4710.2960.003 Dose Selection

27. The adaptive design Has lower type-1 error: 0.7% compared to 1.8% Has higher power: 89% on average compared to 69% Is more likely to be successful and select the correct MED: 51% on average compared to 35% – In “Peak at 50mg” scenario the fixed does better at selecting the MED: fixed: 50.2% adapt: 47.8% In success the adaptive has a lower expected sample size In futility in the Null and Weak scenarios when the fixed has lower expected sample size – – Null: 118, adapt: 125 – Weak: 142, adapt:169. But is that enough to make it worth it?

28. ASSESSING VALUE OF EACH DESIGN

29. Thus in each success scenario We need to calculate the probability, cost and revenue of 10 outcomes: P2a P2b P3 w 25 mg P3 w 50 mg P3 w 100 mg P3 w 150 mg Fail

30. Success depends on dose Probability of success in P3 depends on the dose, and its effect size in that scenario To simplify we assume P3 is the same and fixed for both programs 2 phase 3 trials, each one: – 2 arms – 2-sided alpha 0.05 – Power 0.9 for assumed mean difference of 8 points – 132 per arm – Actual power depends on true effect size

31. NPV Total Revenue = o 0.5 * Peak revenue * remaining market premium * discount Remaining market premium: o 12yrs – development time Discount = (1 – 0.09) development time

32. Peak revenue Peak revenue modelled as dependent on effect size: sigmoid with $500M at mid point of 10pt treatment difference

33. Penalty for overdosing Peak revenue reduced for doses above “MED” Revenue multiplier: Scenario100mg150mg Top Dose1.00 Middle Dose1.000.85 Low Dose0.850.70 50mg peak0.850.70 100mg peak1.000.85

34. Cost assumptions Development overheads: $1M pa Fixed P2a and P2b $1M overhead cost each Adaptive P2a/b $2.5M overhead cost P3 $4M overhead cost Trial cost per subject – US: $45,000 + $1100 per hospital day – Ex US: $34,000 + $280 per hospital day P2 all in US, 14 hospital days P3 half in US, 14 hospital days

35. Time assumptions (fixed) 2 years elapsed already 6 months to prepare P2a 30 subjects per month (3 months to reach mean rec rate) 6 weeks follow up (P2a – 5.5 months elapsed) 6 months to prepare P2b … (P2b - 10.5 months elapsed) 6 months to prepare P3 … (2xP3 – 17.5 months elapsed) 12 months to register 5.4 yrs peak market life remain

36. Time assumptions (adaptive) 2 years elapsed already 9 months to prepare P2a/b 30 subjects per month (3 months to reach mean rec rate) Take mean sample size given scenario and outcome, rounded up 6 weeks follow up 6 months to prepare P3 … (2xP3 – 17.5 months elapsed) 12 months to register 6.1 yrs peak market life remain

37. eNPV

38. NPV per scenario ($M) ScenarioFixed DesignAdaptive seamless P2a/b Null-10.32-11.68 Weak-20.17-22.06 High Dose650.051024.96 Middle Dose477.05756.06 Low Dose282.23431.13 Peak at 50mg314.47363.04 Peak at 100mg201.90484.58

39. Summary Best depends on relative expected likelihood of different scenarios. Say we think – 40% Null – 40% Weak – 20% Positive (4% each of 5 scenarios) Probability weighted eNPV – Fixed:$65M – Adaptive:$109M