Assessing the Impact of Modeling Non-Disease-Related Mortality on Long-Term Survivorship Rates in Previously Untreated Advanced Melanoma - A Case Study from CheckMate 067

Mixture cure models (MCM) provide a useful framework to explore survival heterogeneity and to estimate the proportion of long-term survivors (LTS) in a trial. We examined the sensitivity of LTS rates and resulting survival projections to modeling background mortality in an MCM for previously untreated advanced melanoma patients in the CheckMate 067 trial. For each arm, MCMs were fitted to 5-year progression-free survival (PFS) and overall survival (OS) data from the trial under two scenarios. In the first scenario, LTS were subject to no disease-related mortality and their background hazard was modeled using publicly available age, gender and country specific lifetables for the general population. In the second scenario, LTS were assumed to follow a survival trend similar to that of complete responders (CR) in the trial. In both scenarios, disease-related time-to-event trends for non-LTS were modeled using parametric distributions. After estimation of overall LTS rates, lifetime mean PFS and mean OS were calculated and Bayes’ rule was employed to indirectly derive LTS rates among 5-year survivors. The difference in the estimated LTS rates between the two scenarios was marginal in each arm, 1.6-2.0% for overall population and 0.4-1.6% for 5-year survivors. Under both scenarios considered for LTS background mortality, the difference between LTS rates estimated from PFS versus OS data was lower for 5-year survivors (1.0-11.7% difference across arms) than for the overall population (13.9-16.1% difference across arms). Modeling background mortality with CR survival trend rather than general population mortality generated more conservative mean PFS (14-36 months shorter) and mean OS (5-26 months shorter) estimates across arms. Relaxing the assumption that LTS experience no disease-related mortality can render MCMs more acceptable to health technology assessment reviewers. In this study, it had substantial impact on mean survival projections despite marginal changes in estimated LTS rates.

In Value in Health, 24(S55)