Abstract 3072
Background
Comparative studies of time-to-event endpoints should be designed to yield a wide range of event times to accurately characterize the hazard function (HF) relationship and ensure valid hazard ratio (HR) estimates. The total sample size (N) is ideally small relative to the required number of events. The primary endpoint of progression-free survival (PFS) in METEOR (NCT01865747) required 259 events; the secondary overall survival (OS) endpoint required 408. As a result, the planned total N (650) was much larger than required to evaluate PFS. Shorter PFS times would be overrepresented if an event-driven analysis was conducted among all 650 pts, potentially undermining the ability to assess the proportional hazards assumption. To address this, METEOR employed a novel “trial within a trial” design: PFS was analyzed in the first 375 randomized pts (PFS Pop); OS was analyzed in all 658 randomized pts (ITT Pop). Both populations follow the intention-to-treat principle.
Methods
To assess the impact of the design, PFS was reanalyzed at the date of the 247th event in the ITT Pop (minimum follow-up [min f/up] 2 days) and compared to both the primary endpoint results for 247 events in the PFS Pop (min f/up 11 mo) and supportive results for 394 events in the ITT Pop (min f/up 6 mo).
Results
The HFs were reasonably proportional between arms in all analyses. The HR and median estimate for the cabozantinib arm in the primary analysis of 247 events in the PFS Pop (0.58, 7.4 mo) are close to the estimates in the larger analysis of 394 events in the ITT Pop (0.52, 7.4 mo) using the same cutoff date. Despite the similar relationship among HFs, an analysis using an earlier cutoff based upon 247 events in the ITT Pop is biased, overestimating the treatment benefit as represented by the HR and underestimating the median PFS in the cabozantinib arm (0.49, 6 mo).
Conclusions
The “trial within a trial” design provided critical data required to characterize the HF relationship in METEOR and demonstrate robust results. This design should be considered when the HF relationship is unknown and the total N is large relative to the number of events needed for a time-to-event endpoint.
Clinical trial identification
NCT01865747
Legal entity responsible for the study
Exelixis, Inc.
Funding
Exelixis, Inc.
Disclosure
C. Hessel: Employee: Exelixis Stock or other Ownership: Exelixis. M. Mangeshkar: Employee and Stock Ownership: Exelixis, Inc. R.J. Motzer: Consulting or Advisory Role: Pfizer, Novartis, Eisai Inc. Research Funding: Exelixis, BMS, Novartis, Pfizer, Genentech/Roche. B. Escudier: Honoraria: Exelixis, Novartis, Pfizer, BMS, Roche. T.B. Powles: Honoraria: Genentech/Roche, Novartis Consulting/Advisory Role: Genentech/Roche, BMS Research Funding: AZ/MedImmune, Genentech/Roche, GlaxoSmithKline. G. Schwab: Employee, Stock Ownership, Officer of the Company: Exelixis, Inc. T.K. Choueiri: Consulting/Advisory Role: Pfizer, GSK, Novartis, Merck, Bayer, Eisai, Roche, Prometheus. Labs Inc, BMS, Foundation Medicine Inc. Research Funding: Pfizer, GSK, Novartis, BMS, Merck, Exelixis, Roche, AstraZeneca, Tracon, Peloton.