OBJECTIVES: Mixture cure models (MCMs) explore survival heterogeneity when there is potential for cure by estimating cure rates and the survival function of the uncured. We propose a novel unified network meta-analysis (NMA) framework to simultaneously estimate relative treatment effects on both outcomes of MCMs.

METHODS: MCMs are fit independently to each trial arm of each trial to obtain the cure rates and survival function parameters for the uncured, which are subsequently analyzed with a multivariate NMA. The results of the NMA are the relative treatment effects versus a pre-specified reference treatment regarding cure rates expressed as odds ratios (ORs) and differences in survival function parameters for the uncured to describe time-varying hazard ratios (HRs). Post-NMA, cure rates and survival curves for the uncured subpopulation of each treatment can be obtained by applying the estimated relative treatment effects to corresponding parameter estimates for the pre-specified reference treatment and use of external background mortality rates. The methodology was employed to analyze recurrence-free survival (RFS) data in resected stage III/IV melanoma comparing nivolumab (NIVO), pembrolizumab (PEM), dabrafenib+trametinib (DAB+TRAM), ipilimumab (IPI), and placebo.

RESULTS: Based on a log-normal MCM, the resulting ORs for cure rates for active treatments relative to placebo were 2.61 (95% credible interval [CI] 1.57, 4.36) for NIVO, 2.10 (95%CI 1.18, 3.68) for PEM, 1.86 (95%CI 1.30, 2.68) for DAB+TRAM, and 1.45 (95%CI 1.00, 2.09) for IPI. Three- and five-year RFS HRs for the uncured population were not statistically significant between treatments.

CONCLUSIONS: Given the increased use of MCMs in estimating cure rates in the immunotherapy era, the presented multivariate NMA approach addresses limitations due to violation of proportional hazards and enables synthesis and indirect comparisons in the presence of long-term survivors.