OBJECTIVES: Overall survival (OS) is essential in cost-effectiveness analyses; however, it may not be available from the initial database-locks (DBLs) of a randomized trial. This research compared predicted long-term mean OS gains between the arms of the phase 3 RELATIVITY-047 study (nivolumab [NIVO]+relatlimab [RELA] versus NIVO) obtained from extrapolated OS using reported trial data and from three approaches indirectly modeling OS from progression free survival (PFS).

METHODS: Initial DBL for RELATIVITY-047 study only reported PFS, whereas the successive DBL reported both PFS and OS. Two indirect approaches employed a semi-Markov model to estimate the OS for both arms of the study using the estimated post-progression survival from the NIVO arm of CheckMate 067 study assuming similar subsequent treatment patterns between the trials and the two arms of RELATIVITY-047. The third approach predicted OS through a surrogacy relationship between PFS-OS and extrapolated OS for NIVO from CheckMate 067. All indirect approaches used PFS data from the initial DBL of RELATIVITY-047. OS data from the trial was extrapolated using parametric models and adjusted with background mortality to generate a benchmark for the indirect methods. Projections over a time horizon of 40 years were compared.

RESULTS: Lognormal and gamma distributions were used to extrapolate OS data for NIVO+RELA and NIVO from RELATIVITY-047, respectively. Mean OS gain with NIVO+RELA versus NIVO was estimated as 3.1 years using reported trial data and ranged between 1.1-1.8 years using the indirect methods. As a sensitivity analysis, shortening the time-horizon to 30-years for projections only slightly affected the underestimation margins of mean OS gain by indirect methods (2.9 years with reported trial data versus 1.1-1.7 years with indirect methods).

CONCLUSIONS: This case study showed that in the absence of reported OS data from a trial, indirect methods can be conservative options for subsequent cost-effectiveness analyses by underestimating underlying long-term mean survival benefit.