Discussion on "Bayesian meta-analysis of penetrance for cancer risk" by Thanthirige Lakshika M. Ruberu, Danielle Braun, Giovanni Parmigiani, and Swati Biswas.

This article discusses a Bayesian hierarchical modeling approach for meta-analysis of cancer risk penetrance, as introduced by Ruberu et al. (2023). The authors highlight the method's use of parametric assumptions—specifically, Weibull distributions for study-specific penetrance functions and Gamma priors for population parameters—and propose extensions involving nonparametric Bayesian models to enhance robustness and accommodate heterogeneity across studies. They also explore likelihood-free inference techniques, such as approximate Bayesian computation, to handle diverse reporting modalities without relying on normal approximations. An application to meta-analysis of cancer immunotherapy survival data illustrates the flexibility of nonparametric Bayesian priors, probabilistic clustering of studies, and regression on study-level covariates to improve inference and prediction for future cohorts. The discussion emphasizes the potential of these Bayesian methods to provide coherent, efficient, and flexible inference in complex meta-analytic settings.