A Bayesian Model Leveraging Multiple External Data Sources to Improve the Reliability of Lifetime Survival Extrapolations in Metastatic Non-Small-Cell Lung Cancer.

This article focuses on extending the Bayesian multiparameter evidence synthesis (B-MPES) framework to incorporate historical trial data alongside registry and general population data to improve long-term survival extrapolations from immature clinical trial data in metastatic non-small-cell lung cancer (mNSCLC). Applied to early data cuts from the phase III CheckMate 9LA trial comparing nivolumab plus ipilimumab plus chemotherapy (NIVO+IPI+CHEMO) versus chemotherapy alone, B-MPES models better anticipated the emergent survival plateau characteristic of immunotherapy than standard parametric models (SPMs), although incorporating certain historical trial data led to slight overestimation due to confounding effects. The study found that survival extrapolations were relatively robust to the choice of external data sources when adjustments and rescaling were applied to mitigate confounding, and that B-MPES offers a transparent and clinically plausible approach for health technology assessments requiring lifetime survival predictions. The authors emphasize the importance of carefully selecting and justifying external data and model assumptions to avoid overfitting and ensure reliable extrapolations.