OBJECTIVES:

Mixture-cure models (MCM) of overall survival (OS) based upon data from the CheckMate 227 (CM227) trial to a minimum of 3 years of follow-up (Feb. 2020) have previously been presented. This analysis updates these models with 2 years of additional follow-up (Feb. 2022) for nivolumab plus ipilimumab (NIVO + IPI) and platinum doublet chemotherapy (Chemo) in order to improve precision and investigate parameter sensitivity to data maturity under an MCM framework.

METHODS:

A baseline hazard of mortality using matched general population lifetables was derived. A long-term responder (LTR) proportion was modelled at no excess hazard, and the remainder were modelled with a parametrically defined excess hazard. A number of alternative standard hazard functions families (e.g. Weibull, Gompertz) were explored to describe this excess and so investigate structural uncertainty in the model with respect to this choice.

RESULTS:

For NIVO + IPI, Feb. 2020 LTR fractions varied from 29.1% (95% confidence interval (CI): 24.0%, 34.0%) for the Weibull model to 11.1% (95% CI: 0.1%, 21.8%) for the log-normal model. The estimates were reduced to 21.6% (95% CI: 17.4%, 26.0%) for the Gamma model and 8.2% (95% CI: 0.1%, 16.3%) for the log-normal model using Feb. 2022 data. For CHEMO, the previous LTR range was 18.1% (95% CI: 14.3%, 21.8%) [Gompertz] to 5.0% (95% CI: 0.0%, 12.6%) [log-normal] and compared to updated estimates of 11.7% (95% CI: 8.5%, 14.9%) [Weibull] to 5.1% (95% CI: 0.1%, 9.7%) [log-normal].

CONCLUSIONS:

Two years of additional survival follow-up have improved the precision of OS MCM parameter estimates. LTR parameters of models estimating extended survival in the non-LTR population varied less between the DBLs than those estimating lower mean survival in this higher-risk fraction, but confidence intervals were narrowed for all models. Parameter stability may give heuristic insight into MCM model selection from less mature data.