OBJECTIVES : Different mechanisms of action and complex underlying risk functions associated with IO treatments have led to exploration of different methods to estimate the long-term survival outcomes . Several ISPOR workshops have addressed the use of various approaches including: parametric survival, spline, landmark, and cure-fraction models. The objective of this study was to examine differences in estimated short and long-term survival using different methods, in previously treated mMCC patients receiving avelumab.

METHODS : Efficacy data from the phase II JAVELIN Merkel 200 trial, with ≥12 months of follow-up was analyzed. Standard parametric survival analyses, landmark analyses and analyses of OS based on time to progression (TTP) plus post progression survival (PPS) duration were used to project OS. In landmark analyses, parametric survival models were fit by response status to OS from the landmark point of 89 days (~13 weeks). In analyses of TTP plus PPS, parametric survival models were fit to PPS and spline models to TTP. Predicted OS was compared with the observed OS from ≥18 months of follow-up, and OS outcome throughout patient lifetime was projected.

RESULTS : Estimated OS derived by any of the 3 approaches fit well with the observed OS curve from ≥18 months of follow-up when based on appropriate statistical distribution assumptions. However, long-term projections differed between methods. Average life expectancy estimated with avelumab ranged between 2.48 years when a lognormal distribution was used to 3.54 years when OS was modeled as TTP plus PPS. The estimated proportion of patients alive varied between 12% and 19% at 5 years and 5% and 11% at 10 years.

CONCLUSIONS : The different methods for projecting OS with IO treatments may have good short-term predictive accuracy, but long-term outcomes vary considerably. Additional data-cuts with longer follow up may provide further insight into the best approach in projecting long-term survival.