<![CDATA[Hemodialysis is a critical treatment for patients with end-stage chronic kidney disease (CKD). Although it is associated with a higher risk of mortality and complications compared to other renal treatment choices, many patients prefer hemodialysis as their renal replacement therapy. Traditional parametric survival models often struggle to capture the complexity of the distribution due to multimodal survival data in heterogeneous populations, such as hemodialysis patients. To overcome this, mixture models provide greater flexibility by combining several distributions to better reflect latent survival patterns. This study investigates mortality risk factors by applying a Bayesian Weibull mixture proportional hazards model to 151 hemodialysis patients from one hospital in Surabaya, Indonesia, in 2024, with data extracted from medical hospital records. The Expectation-Maximization and No-U-Turn Sampler (EM-NUTS) approach was used to estimate model parameters and latent class memberships. The Expectation-Maximization (EM) approach, originally developed for control charts and time series, was adapted for survival analysis to address latent heterogeneity. Exploratory analysis reveals multimodal survival distributions, suggesting distinct risk groups. Model selection using the Bayesian Information Criterion and the Akaike Information Criterion identified two latent groups with distinct risk profiles. In the first group, male gender and diabetes significantly increased mortality risk, while hypertension had a smaller effect. In the second group, hypertension was the dominant risk factor, with less influence from gender and diabetes. These findings emphasize the importance of accounting for population heterogeneity in the survival analysis of hemodialysis patients. The use of advanced Bayesian mixture models with EM-NUTS estimation provides robust tools for uncovering hidden subgroups and improving risk stratification, allowing for more personalized treatment strategies in CKD care.]]>
Copyrights © 2026
