OBJECTIVES: NICE decision unit (DSU) 21 discusses mixture cure models (MCM) considering one trial arm and hence do not discuss treatment effects (TE). We explored a MCM network meta-analysis (NMA), focussing on TE applied to MCM parameters.

METHODS: We conducted a MCM network meta-analysis (NMA) in melanoma for two trials, which included three treatments (pembrolizumab, ipilimumab and nivolumab). The following MCMs were tested; exponential, gompertz, loglogistic, lognormal and Weibull. The base model applied TE on all MCM parameters (cure, scale and shape) (model1). The Leave-One-Out-Information-Criteria(LOOIC) was used to determine the goodness-of-fit. The MCM with lowest LOOIC was further optimized by only keeping the TEs for parameters for which the value of the SD smaller than the (absolute) value of the TE (model2). We also explored a model with applying TE to the cure parameter only (model3), which is standard setting in flexsurvcure.

RESULTS: The lognormal distribution rendered the best fit when applying TE to all three parameters (model1); only TEs for shape and scale had SD < TE. Therefore, model2 included TE on scale and shape only, and not on cure. The respective LOOICs for model1, model2 and model3 are 8465/8461/8472. The mean overall survival (95%CrI) in years for model1 for nivolumab, ipilimumab and pembrolizumab respectively was 10.73(4.87-20.65), 7.44(5.99-8.98), 8.72(7.24-9.87). For model2 9.27(7.88-10.67), 7.7(6.64-8.82), 8.54(7.57-9.48) and for model3 15.54(9.95-21.84), 6.92(5.41-8.40), and 9.07(7.92-10.17) respectively. The cure fraction for model1 for nivolumab, ipilimumab and pembrolizumab respectively was 0.40(0.01-0.97), 0.29(0.21-0.37), 0.31(0.20-0.39). In model2 all three treatments had cure rate of 0.31(0.25-0.31). For model3 corresponding cure rates are 0.67(0.39-0.99), 0.22(0.16-0.32) and 0.35(0.19-0.41).

CONCLUSIONS: Our results shows that TE parametrization in MCMs results in different (incremental) survival estimates and uncertainty. Further research is needed on parametrization of TE in flexible survival models as these outcomes impact health technology assessment decision making.