OBJECTIVES: This study explores the potential for mixture-cure models (MCM) to capture heterogeneity in life expectancy with nivolumab plus ipilimumab (N+I), given the novel mechanism of action and durable, deep survival response, using the CheckMate-227 (CM227) Part 1 trial of advanced non-small cell lung cancer (aNSCLC).

METHODS: MCM relative survival models were fitted by maximum likelihood to OS data, with a minimum of three years’ follow-up, from all patients randomised to N+I or chemotherapy in CM227 (database lock: February 2020). Baseline hazard was informed by nationality, age and gender-specific lifetables matched to each patient. In the base case, long-term survival in the “cured” fraction applied the matched general population’s hazard of death; the complementary “uncured” fraction with shorter survival assumed a parametric hazard function above this baseline. The sensitivity of these models to the distribution family for the “uncured” and the baseline hazard was also investigated.

RESULTS: In the base case scenario, assuming general population baseline hazard, MCM generated “cure” fractions of 0.291 (95% CI: 0.240, 0.340) for N+I with excess hazard in the “uncured” described by a Weibull distribution, greater than for chemotherapy - 0.173 (95% CI: 0.137, 0.214). Applying a lognormal to the “uncured” gave a “cure” fraction of 0.111 (95% CI: 0.001, 0.218) for N+I. Assuming no baseline hazard “cure” fractions were approximately 10% lower.

CONCLUSIONS: Models of OS in CM227 can be represented by MCMs. The “cure” fraction is sensitive to the OS parametric distribution for the “uncured”, with short-tailed distributions (e.g. exponential, Weibull) generating “cure” fractions around 29% vs 17% (N+I vs chemotherapy). This fraction is also sensitive to the baseline hazard: parameters are specific to the hazard assumptions made during fitting. Validation of long-term OS using external sources and longer follow-up can help to select appropriate parametric distributions and assess the plausibility of assumptions.