• Cyrus Mehta • Sam Hsiao • Pranab Ghosh
Cytel/Pfizer Symposium 25 Aug 2016 2
Acknowledgement
• Background IntroducFon • AdapFve Enrichment Design • OperaFonal ConsideraFons • Regulatory Experience • Conclusions
Cytel/Pfizer Symposium 25 Aug 2016 3
Outline
• A rare and aggressive cancer • Reported incidence in EU 1/100000 • 5-‐year survival rate of less than 12% • Current Standard Therapy – Limited treatment opFons – Modest benefit
Cytel/Pfizer Symposium 25 Aug 2016 4
Indication
• Add-‐on to current drug A • Drug A-‐-‐Oral inhibitor of VEGFR • No complete responses observed for such paFents with Drug A
• New drug B targeFng a different pathway • targeFng both pathways concurrently is more effecFve than targeFng either pathway alone
• Non-‐overlapping toxicity profile allowing prolonged dosing in responding paFents
Cytel/Pfizer Symposium 25 Aug 2016 5
New Drug B
• Safety and acFvity being evaluated in a phase 1b/2 study
• Interim efficacy data show radiographic reducFons in tumor volume
• Some paFents with durable complete responses • Two types of paFents (S+ and S-‐) characterized by disease locaFon
• Be^er response observed for paFents in S+ than those in S-‐
Cytel/Pfizer Symposium 25 Aug 2016 6
Clinical development
• Available safety and efficacy data are considered compelling and supporFve of a registraFon-‐enabling clinical study
• Given the extremely small number of paFents in this populaFon
• Single pivotal phase 3 trial to support the claimed orphan indicaFon
Cytel/Pfizer Symposium 25 Aug 2016 7
Single Pivotal Phase 3 Trial
• Considerable uncertainty about the treatment effect of B given with A compared to single agent A
• S-‐ is considered a more aggressive form of cancer and response between S+ and S-‐ may be disparate
• Heterogeneity in response has been observed between S+ and S-‐ in ongoing phase 1b/2 study: S+ has been more responsive
• The uncertainty of the treatment effect puts the study at risk of being underpowered.
Cytel/Pfizer Symposium 25 Aug 2016 8
Design Considerations
Cytel/Pfizer Symposium 25 Aug 2016 9
Design Strategy
124 pa'ents: 70/cohort 1 & 54/cohort 2 95 events: 60/cohort 1 & 30/cohort 2 IA: 40 events from cohort 1 Primary Endpoint: PFS
• Extension of Jenkins, Stone, & Jennison, 2011 • Type I error control is guaranteed by applying the closed tesFng principle together with the combinaFon test
• Allow unblinding other secondary endpoints to facilitate interim decision
Cytel/Pfizer Symposium 25 Aug 2016 10
Type I Error Control
• In case of no enrichment, declare significance on Full populaFon if 𝑤↓1 Φ↑−1 (𝑝↓1↑𝐹𝑆 )+ 𝑤↓2 Φ↑−1 (𝑝↓2↑𝐹𝑆 )> 𝑧↓𝛼 𝑤↓1 Φ↑−1 (𝑝↓1↑𝐹 )+ 𝑤↓2 Φ↑−1 (𝑝↓2↑𝐹 )> 𝑧↓𝛼
• In case of enrichment, declare significance on S+ if 𝑤↓1 Φ↑−1 (𝑝↓1↑𝐹𝑆 )+ 𝑤↓2 Φ↑−1 (𝑝↓2↑𝑆 )> 𝑧↓𝛼 𝑤↓1 Φ↑−1 (𝑝↓1↑𝑆 )+ 𝑤↓2 Φ↑−1 (𝑝↓2↑𝑆 )> 𝑧↓𝛼
Cytel/Pfizer Symposium 25 Aug 2016 11
Final Analysis
Cytel/Pfizer Symposium 25 Aug 2016 12
Simulations
HR for S+, S-‐, Full Popula'on
Zone Prob of Zone
Power
Average Study Dura'on Average Sample Size
Fxd Fxd & SSR
Fxd & SSR & Enrich
Fxd Fxd & SSR
Fxd & SSR & Enrich
Fxd Fxd & SSR
Fxd & SSR & Enrich
0.6 0.6 0.6 Enrich 17% 32% 32% 43% 20 20 25 124 124 168
Prom 44% 67% 86% 86% 20 27 27 124 200 200
Fav 39% 92% 92% 92% 20 20 20 124 124 124
Total 100% 70% 79% 81% 20 23 24 124 157 165
0.5 0.8 0.64 Enrich 23% 43% 43% 75% 20 20 25 124 124 169
Prom 46% 73% 90% 90% 20 27 27 124 200 200
Fav 31% 93% 93% 93% 20 20 20 124 124 124
Total 100% 73% 80% 88% 20 23 24 124 159 169
0.5 1.0 0.72 Enrich 35% 41% 41% 76% 19 19 26 124 124 169
Prom 45% 70% 87% 87% 19 26 26 124 200 200
Fav 20% 90% 90% 91% 19 19 19 124 124 124
Total 100% 64% 71% 84% 19 23 25 124 158 174
• Monitoring of enrolment and event arrival – Track paFents belonging to each cohort and when the enrolment switches from cohort 1 to cohort 2
– Track PFS events belonging to each cohort • Establish appropriate firewall to protect the study integrity – Document who saw what and when – Only DMC and ISC have access to unblinded data
• Web based system and document management system (ACES) to manage the storage of data and documents – Role based access – Full audit tracking capabiliFes
Cytel/Pfizer Symposium 25 Aug 2016 13
Operation Considerations
• Handling Fme difference between arrival of 70 paFents for cohort 1 and arrival of 40 PFS events 1) The 40 PFS events arrive before 70 paFents are enrolled
! Arrival of 40 PFS events will trigger the interim analysis for adapFve decision ! Data cleaning and data base lock ! Transfer data to independent staFsFcian for unblinded analysis ! DMC meeFng ! AdapFve decision ! For final analysis, cohort 1 will comprise the first 70 paFents who will be followed unFl arrival
of 60 events
2) The 40 PFS events arrive aher 70 paFents are enrolled ! Cohort 1 is closed as soon as the 70 patents have been enrolled who will be followed unFl 60
events ! All future arrivals belong to cohort 2 ! Limit the Fme between start of cohort 2 recruitment and interim adapFve decision making ! Perform interim analysis either 30 days aher the arrival of 70th subject or 14 days aher the
arrival of the 40th event whichever is earlier
Cytel/Pfizer Symposium 25 Aug 2016 14
Operation Considerations
• To maintain the independence of cohort 1 and cohort 2 – Cohort 1 is closed when 70 paFents and 60 events arrive
– In case of SSR or enrichment, only cohort 2 will be expanded or modified
– AddiFonal events from cohort 1 other than the 60 events will not be included in the final analysis
Cytel/Pfizer Symposium 25 Aug 2016 15
Independence between Cohort 1 and Cohort 2
• JusFficaFon of the adapFve design proposal • Large convenFonal design should be considered and discussed
• Impact of adapFve enrichment on hypothesis test and parameter esFmaFon – Type I error is strongly controlled – Parameter esFmaFon is sFll an acFve research topic
– No rigorous soluFon to esFmaFon problem in case of adapFve enrichment
Cytel/Pfizer Symposium 25 Aug 2016 16
Regulatory Feedback
• If the interim result from cohort 1 enters the favorable zone, use the convenFonal maximum likelihood esFmate of hazard raFo from the Cox model.
• If the interim result from cohort 1 enters the promising zone, use the "backward image" method of Gao, Liu and Mehta (2013) to esFmate the log of the hazard raFo.
• If the interim result from cohort 1 enters the enrichment zone, use only the enriched data from cohort 2 to esFmate the hazard raFo. This ensures that the parameter esFmate will be unbiased though the resulFng confidence interval may not be consistent with the hypothesis test.
Cytel/Pfizer Symposium 25 Aug 2016 17
Proposal for estimation
• There is ohen significant uncertainty and heterogeneity with the treatment effect among subpopulaFons for rare oncology area
• The adapFve enrichment design provides an alternaFve opFon to miFgate the risk of being under powered
• Pros and cons of such adapFve design need to be evaluated thoroughly on a case-‐by-‐case basis
• ProspecFve planning is criFcal for successful implementaFon of such design
Cytel/Pfizer Symposium 25 Aug 2016 18
Conclusions