1. Adaptive Clinical Trials: A New Paradigm in Drug Development Donald A. Berry dberry@mdanderson.org Donald A. Berry dberry@mdanderson.org

2. 2 2 Financial Disclosure Part owner Berry Consultants, LLC. Designs adaptive trials for medical device and pharmaceutical companies. Part owner Berry Consultants, LLC. Designs adaptive trials for medical device and pharmaceutical companies.

3. 3 3 Outline l Background: Bayesian adaptive approach in drug development l Seamless phase II/III trial from Critical Path Initiative l Biomarker-driven trials & I-SPY2 l Background: Bayesian adaptive approach in drug development l Seamless phase II/III trial from Critical Path Initiative l Biomarker-driven trials & I-SPY2

4. 4 4 Janet Woodcock, Director CDER FDA “Improved utilization of adaptive and Bayesian methods” could help resolve low success rate of and expense of phase 3 clinical trials

5. 5 5 Current use of Bayesian adaptive designs l MDACC (> 300 trials) l Device companies (> 25 PMAs)* l Drug companies (Most of top 40; many biotechs)** l MDACC (> 300 trials) l Device companies (> 25 PMAs)* l Drug companies (Most of top 40; many biotechs)** *http://www.fda.gov/MedicalDevicesDeviceRegulationandGuidance/G uidanceDocuments/ucm071072.htm **http://www.fda.gov/downloads/DrugsGuidanceCompliance RegulatoryInformation/Guidances/UCM201790.pdf

6. 6 6 Some areas of application of Bayesian adaptive drug trials l Oncology l Migraine l Rh Arthritis l Lupus l Sepsis l Diabetes l Obesity l Stroke l Gastroparesis l Oncology l Migraine l Rh Arthritis l Lupus l Sepsis l Diabetes l Obesity l Stroke l Gastroparesis l Spinal Cord Injury l HIV l Hepatitis C l Pre-term labor l Constipation l Overactive bladder l Libido l Alzheimer’s l Parkinson’s l Spinal Cord Injury l HIV l Hepatitis C l Pre-term labor l Constipation l Overactive bladder l Libido l Alzheimer’s l Parkinson’s

7. 7 7 "Our current approach [to trials] is horribly inefficient, and we need to do something better," says Roger Lewis, an emergency medicine physician at Harbor-University of California, Los Angeles, Medical Center. Lewis helps advise a company called Berry Consultants …

8. 8 8 Lewis and Berry, along with emergency medicine physician William Barsan at the University of Michigan, will be studying whether "adaptive" trial designs that incorporate new information in midcourse can answer medical questions. They also want to learn what concerns researchers might have about this approach.

9. 9 9 Bayesian adaptive trials l Stopping early (or late) n Efficacy n Futility l Dose finding (& dose dropping) l Seamless phases l Population finding l Adaptive randomization l Ramping up accrual l Stopping early (or late) n Efficacy n Futility l Dose finding (& dose dropping) l Seamless phases l Population finding l Adaptive randomization l Ramping up accrual

10. 10 Why? l Smaller trials (usually!) l More accurate conclusions l Can focus on better treatment of patients in trials l Smaller trials (usually!) l More accurate conclusions l Can focus on better treatment of patients in trials

11. 11 References l Berry DA (1996). Statistics: A Bayesian Perspective. Duxbury. (ISBN: 978-0534234720) l Berry DA (2006). Bayesian clinical trials. Nature Reviews Drug Discovery. (Free download: http://archlab.gmu.edu/people/jthompsz/Berry 2006.pdf) l Berry DA (2009). Statistical Innovations in Cancer Research. In Cancer Medicine e.8. Ch 35, pp 446-463. London: BC Decker. (Ed: Holland J, Frei T et al.) l Berry DA (1996). Statistics: A Bayesian Perspective. Duxbury. (ISBN: 978-0534234720) l Berry DA (2006). Bayesian clinical trials. Nature Reviews Drug Discovery. (Free download: http://archlab.gmu.edu/people/jthompsz/Berry 2006.pdf) l Berry DA (2009). Statistical Innovations in Cancer Research. In Cancer Medicine e.8. Ch 35, pp 446-463. London: BC Decker. (Ed: Holland J, Frei T et al.)

12. 12 Adaptive Phase II/III Trial

13. 13 IoM Report on Cancer Cooperative Groups 2010

14. 14 Similar Example from Critical Path Initiative l Type II diabetes l Seamless Phase II/III: Dose finding plus confirmation l Active comparator & placebo l Primary endpoint: Clinical Utility Index (12 months) l Type II diabetes l Seamless Phase II/III: Dose finding plus confirmation l Active comparator & placebo l Primary endpoint: Clinical Utility Index (12 months)

15. 15 Some Details l Longitudinal modeling l Phase II: 7 doses experimental drug l Phase III n 1 or 2 doses experimental drug n Sample size via predictive power considering available Phase II data n Adaptive transition: Bayesian predictive probs l Both phases driven by CUI l Longitudinal modeling l Phase II: 7 doses experimental drug l Phase III n 1 or 2 doses experimental drug n Sample size via predictive power considering available Phase II data n Adaptive transition: Bayesian predictive probs l Both phases driven by CUI

16. 16 Clinical Utility Index l Dose-response modeling l Longitudinal modeling l Dose-response modeling l Longitudinal modeling Exp – AC 12 mos Exp – AC 12 mos Exp – P 12 mos Exp – P 12 mos 0 0 0 0

17. 17 Analogy with Composite Endpoint Exp – AC 12 mos Exp – AC 12 mos Exp – P 12 mos 0 0 0 0 Exp – P 12 mos

18. 18 Adaptive Biomarker Trial

19. 19 EXAMPLETRIALEXAMPLETRIAL

20. 20 Simulations Usually Required l To find operating characteristics: n Type I error rate n Power n Sample size distribution l Prospective design essential l Longitudinal modeling l Many scenarios l Accrual rate matters l To find operating characteristics: n Type I error rate n Power n Sample size distribution l Prospective design essential l Longitudinal modeling l Many scenarios l Accrual rate matters

21. 21 Savings possible in sample size when using biomarkers …

22. 22 Berry et al. SABCS 2009

23. 23 Each subset shows a statistically significant benefit from paclitaxel with small sample size—could have been small, focused phase III No paclitaxel benefit

24. 24

25. I-SPY2: The Cartoon (Press conference* slides) *

26. Standard Phase 2 Cancer Drug Trials Experimental arm Population of patients Outcome: Tumor shrinkage? Experimental arm Population of patients Outcome: Longer time disease free Standard therapy RANDOMIZERANDOMIZE

27. Standard Phase 2 Cancer Drug Trials Experimental drug Outcome: Longer time disease free Standard therapy RANDOMIZERANDOMIZE Outcome: Tumor shrinkage? Population of patients Population of patients Consequence: 60-70% Failure of Phase 3 Trials

28. I-SPY2 TRIAL Experimental arm 1 Outcome: Complete response at surgery Experimental arm 2 Experimental arm 3 Experimental arm 4 Experimental arm 5 Population of patients Standard therapy ADAPTIVELYADAPTIVELY RANDOMIZERANDOMIZE

29. I-SPY2 TRIAL Outcome: Complete response at surgery Experimental arm 2 Arm 2 graduates to small focused Phase 3 trial Experimental arm 1 Experimental arm 3 Experimental arm 4 Experimental arm 5 Standard therapy Population of patients ADAPTIVELYADAPTIVELY RANDOMIZERANDOMIZE

30. I-SPY2 TRIAL Outcome: Complete response at surgery Experimental arm 1 Experimental arm 3 Experimental arm 4 Experimental arm 5 Standard therapy Population of patients Arm 3 drops for futility ADAPTIVELYADAPTIVELY RANDOMIZERANDOMIZE

31. I-SPY2 TRIAL Outcome: Complete response at surgery Experimental arm 5 Arm 5 graduates to small focused Phase 3 trial Experimental arm 1 Experimental arm 4 Standard therapy Population of patients RANDOMIZERANDOMIZE ADAPTIVELYADAPTIVELY

32. I-SPY2 TRIAL Outcome: Complete response at surgery Arm 6 is added to the mix Experimental arm 6 Experimental arm 1 Experimental arm 4 Standard therapy Population of patients RANDOMIZERANDOMIZE ADAPTIVELYADAPTIVELY Goal: Greater than 85% success rate in Phase 3, with focus on patients who benefit

34. 34 I-SPY2 Adaptive Design Process  PI: Laura Esserman, UCSF  Multi-disciplinary team  Funded by FNIH: NCI, FDA, industry, academia  Coordinated with FDA’s CDER & CDRH from inception  PI: Laura Esserman, UCSF  Multi-disciplinary team  Funded by FNIH: NCI, FDA, industry, academia  Coordinated with FDA’s CDER & CDRH from inception

35. 35 I-SPY2: Adaptive Phase II Neoadjuvant Breast Cancer l Moderate to high-risk primary breast cancer l Baseline biopsy: assess biomarkers l Primary endpoint: pathCR l Longitudinal modeling: MRI volume On Study MRI Blood Surgery Biopsy Blood MRI Biopsy Tissue Taxane +/–New Drug (12 weekly cycles) AC (4 cycles)

36. 36 Overview of Design  Randomized control: T  AC  Randomization to new drugs, balanced initially then adaptively  Find predictive probability of success in focused 300-patient phase III trial  Evaluate many drugs & combinations  Successes graduate to Phase III  Duds dropped for futility  Randomized control: T  AC  Randomization to new drugs, balanced initially then adaptively  Find predictive probability of success in focused 300-patient phase III trial  Evaluate many drugs & combinations  Successes graduate to Phase III  Duds dropped for futility

37. 37 Screening Process: Matching Drugs to Populations l Many drugs; 5 to start: s PARP inhibitor s IGFR inhibitor s Angiogenesis inhibitor l Pair drugs/biomarker signatures l Many drugs; 5 to start: s PARP inhibitor s IGFR inhibitor s Angiogenesis inhibitor l Pair drugs/biomarker signatures s TRAIL agonist s Pan HER inhibitor s TRAIL agonist s Pan HER inhibitor

38. 38 Patient strata Estimated prevalences based on I-SPY1: MP: MammaPrint, High+ vs High- HR+: Hormone Receptor+: Either ER+ or PgR+ MP: MammaPrint, High+ vs High- HR+: Hormone Receptor+: Either ER+ or PgR+

39. 39 pathCR rates in I-SPY 1

40. 40 l Graduate drugs/signatures from trial: nBased on effectiveness nBased on prevalence l Biomarker signatures (2^8 combinations of subtypes): B 1, B 2, …, B 256 l But restrict to (10) marketable signatures: l Graduate drugs/signatures from trial: nBased on effectiveness nBased on prevalence l Biomarker signatures (2^8 combinations of subtypes): B 1, B 2, …, B 256 l But restrict to (10) marketable signatures: Biomarker signatures

41. 41 Some details  Sample size for each drug, 20 to 120 (minimum 60 if “success”)  Maximum of 5 exp drugs at a time  Patient enters trial, identify subtype  Find (Bayesian) prob each drug >> control; based on all current results  Covariate modeling (across subtypes)  Sample size for each drug, 20 to 120 (minimum 60 if “success”)  Maximum of 5 exp drugs at a time  Patient enters trial, identify subtype  Find (Bayesian) prob each drug >> control; based on all current results  Covariate modeling (across subtypes)  Assign in proportion to current prob drug >> control, by subtype

42. 42 More details  For each possible biomarker signature B, calculate prob drug >> control in B  If Bayesian pred prob 300-pt Phase III success < 10% for all B, drop drug  If > 85% for some B then drug graduates  At graduation we provide predictive probability Phase III success for each B, including B on drug’s diploma  For each possible biomarker signature B, calculate prob drug >> control in B  If Bayesian pred prob 300-pt Phase III success < 10% for all B, drop drug  If > 85% for some B then drug graduates  At graduation we provide predictive probability Phase III success for each B, including B on drug’s diploma

43. 43 Computer Simulations: Probability graduate (power) … l P1: True signature l P2: All subtypes that benefit l P3: Only subtypes that benefit l P4: Some subtype that benefits l P5: No subtypes that benefit l P1: True signature l P2: All subtypes that benefit l P3: Only subtypes that benefit l P4: Some subtype that benefits l P5: No subtypes that benefit

44. 44 I-SPY 2 Effects l Match drugs (& combos) with biomarker signatures l Graduate drug/biomarker pairs to smaller (n = 300), more focused, more successful Phase III

45. 45 OutlineOutline l Background: Bayesian adaptive approach in drug development l Seamless phase II/III trial from Critical Path Initiative l Biomarker-driven trials & I-SPY2 l Background: Bayesian adaptive approach in drug development l Seamless phase II/III trial from Critical Path Initiative l Biomarker-driven trials & I-SPY2