Estimating Long-Term Survivorship In Patients With Advanced Melanoma Treated With Immune-Checkpoint Inhibitors: Analyses From The Phase Iii Checkmate 067 Trial

Immune-checkpoint inhibitors nivolumab (NIVO) and ipilimumab (IPI), alone and in combination, have demonstrated durable long-term survival patterns in previously untreated patients with advanced melanoma in the CheckMate 067 trial (NCT01844505). Plateaus in overall survival (OS) and progression-free survival (PFS) data can be attributed to survival heterogeneity, which can be better captured by mixture models. We assumed that a subset of patients can be classified as long-term survivors (LTSs) and that their survival trend follows that of the general population, with no excess mortality due to melanoma. A cohort-level background survival distribution was derived using mortality rates from the World Health Organization and demographic information from CheckMate 067. After assessing the suitability of mixture models by comparing hazard functions for the 5-year data in the entire trial population and the general population, we fit mixture models to the OS and PFS data to estimate the proportion of LTSs in each treatment arm. Time-to-event outcomes of non-LTSs were modeled by parametric survival functions, the forms of which were varied to obtain a range for the proportions of LTSs and to test the robustness of the results. Regardless of the data source (OS or PFS), ranges of estimated proportions of LTSs did not overlap across the treatment arms. Based on OS analyses, ranges of estimated proportions of LTSs were 38–46% for NIVO, 49–54% for NIVO+IPI, and 16–26% for IPI. Based on PFS analyses, ranges were 29–33% for NIVO, 38–40% for NIVO+IPI, and 9–13% for IPI. As these ranges represent only a span of point estimates of LTSs across model choices, a formal statistical assessment of the significance of LTSs among the treatment arms would require a comparison of confidence intervals. Mixture models adequately captured the survival plateaus in CheckMate 067 and suggested a higher proportion of LTSs with NIVO and NIVO+IPI than with IPI. These methods may be used in the indirect estimation of auxiliary outcomes, such as time to subsequent treatment and impact of subsequent treatments on LTSs.