OBJECTIVES: Survival heterogeneity poses challenges for assessing long-term clinical and economic value of immune checkpoint inhibitors. This study aimed to visualize unobservable heterogeneity in survival among 1L I/P-risk aRCC patients treated with NIVO+IPI in the Checkmate 214 trial using PMMs.

METHODS: The study population is assumed to consist of two non-overlapping and exhaustive latent subgroups with distinct survival patterns. PMMs were fitted separately to PFS and OS data from the trial with minimum 60-months follow-up to simultaneously elicit the proportion and survival function of each subgroup. For each subgroup’s survival, exponential, Weibull, loglogistic (LL), lognormal (LN), and gamma distributions were considered. Best-fitting PMMs among 15 different combinations were identified based on their statistical goodness-of-fit measures, and visual fits to the observed survival and hazard trends defined by pre-specified violation margins (2% for survival and 5% for hazards). PMMs estimating subgroup(s) with <5% proportion, crossing survival functions, or survival rates > 2% higher than general population mortality were disqualified.

RESULTS: Among all candidate PMMs, 8 and 6 combinations were deemed viable for PFS and OS data, respectively, by satisfying all filtering criteria. Majority of these combinations included an LN model for PFS (5 out of 8) and an exponential model for OS (4 out of 6). Best overall fit to the PFS and OS data were provided by a mixture of two LN distributions (weights: 45%, 55%) and by a mixture of two exponential distributions (weights: 41%, 59%), respectively. Estimated 5-year restricted mean PFS and OS across the selected top 3 PMMs (second and third-best fitting PMMs: LN-gamma & LL-LL for PFS; exponential-Weibull & exponential-LL for OS) showed negligible differences (<0.6 month).

CONCLUSIONS: PMMs may adequately capture the complex survival and hazard trends for 1L I/P-risk aRCC patients treated with NIVO+IPI by offering insights on potential survival heterogeneity without making clinically restrictive assumptions.