Spatial Survival Analysis of Stroke Hospitalizations: A Bayesian Approach

Abstract

Survival analysis encompasses a range of statistical techniques used to evaluate data where the outcome variable represents the time until a specific event occurs. When such data is collected across different spatial regions, integrating spatial information into survival models can enhance their interpretive power. A widely adopted method involves applying an intrinsic conditional autoregressive (CAR) prior to an area-level frailty term, accounting for spatial correlations between regions. In this study, we extend the Bayesian Cox semiparametric model by incorporating a spatial frailty term using the Leroux CAR prior. This approach aims to enhance the model's capacity to analyze stroke hospitalizations at Labuang Baji Hospital in Makassar, with a particular focus on exploring the geographic distribution of hospitalizations, length of stay (LOS), and factors influencing patient outcomes. The dataset, derived from the medical records of stroke patients admitted to Labuang Baji Hospital between January 2022 and June 2024, included variables such as LOS, discharge outcomes, sex, age, stroke type, hypertension, hypercholesterolemia, and diabetes mellitus. The analysis revealed that stroke type was a significant determinant of hospitalization outcomes. Specifically, ischemic stroke patients exhibited faster recovery times than those with hemorrhagic strokes, with a hazard ratio of 1.892, representing an 89% greater likelihood of recovery. Additionally, stroke patients across all districts treated at Labuang Baji Hospital demonstrated similar average recovery rates and discharge durations.