Bayesian Accelerated Failure Time Model for Zero-Inflated Survival Data With Application to Liver Cirrhosis

Analyzing count data, particularly in survival analysis, is challenging when there are many zeros, and some observations are right-censored. This study tackles zero-inflated survival event data (ZISED), specifically focusing on liver cirrhosis cases in Nigeria. Current models fall short in effectively capturing ZISED’s unique structure. To address this, we propose a flexible Bayesian accelerated failure time (AFT) modeling framework. We investigate both parametric and semiparametric AFT models—specifically nonpartition AFT (NPAFT) and partition AFT (PAFT)—for Weibull, log-normal, and log-logistic distributions. Using simulations based on various initial coefficient values and sample sizes and a liver cirrhosis dataset, we compared model performance. Results indicate that the log-normal PAFT model provides the best fit for these complex datasets. We identified key covariates influencing survival including age, liver size, hepatitis type, time, hypertension, education level, sex, marital status, diet, drug use, and liver damage. Overall, this study advances ZISED modeling by introducing a flexible Bayesian AFT framework. It also provides valuable insights into factors affecting survival among liver cirrhosis patients. This supports better clinical decision-making and public health interventions.