Non-parametric Frailty Model for the Natural History of Prostate Cancer; Using Data from a Screening Trial

Mixed-effects models for survival, known as frailty models, can be used to capture individual or cluster-specific unobserved heterogeneity. A common choice is assuming the random effects follow a parametric distribution, e.g. a normal distribution. However, computing the marginal likelihood can be computationally expensive and infeasible in a high-dimension setting. Alternatively, a non-parametric approach avoids the normality assumption for random effects and can be less computationally demanding. The data used are from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. The aim is to model prostate cancer progression to evaluate screening strategies, considering the PSA longitudinal biomarker, accounting for unobserved heterogeneity, interval-censored, left-truncation, and right-censored data.