Abstract
Bayesian parametric survival models provide a flexible framework for time-to-event analysis, particularly when structural assumptions may enhance efficiency. However, performance in low-event settings remains an important practical consideration. We evaluated three Bayesian parametric survival models: Exponential, Weibull, and Lognormal, using data from 155 hypertensive patients, among whom 14 events (9.1%) were observed. Five clinically relevant covariates were included a priori in all models. Posterior inference was obtained via Markov Chain Monte Carlo methods. Model comparison was performed using WAIC (Watanabe-Akaike Information Criterion) and LOOIC (Leave-one-out cross-validation). Sensitivity analyses were conducted under alternative prior specifications to assess robustness. The Bayesian Lognormal model (SE = 50.5, LOOIC = 220.4, WAIC = 220.0) outperformed other models. The study showed that occupation is significantly associated with survival time of hypertension patients across all the models. These findings underscore the need for careful consideration of the choice of the parametric model to employ in survival data analysis. The results also suggest the need for the provision of job-related intervention in people's health regarding hypertension.