Predicting Health Insurance Claims in Thailand Using Bayesian Multilevel Logistic Regression and Socioeconomic Indicators

Abstract

Socioeconomic disparities may influence the probability of health insurance claims and the financial sustainability of insurance systems. This study developed a Bayesian hierarchical logistic model to predict the probability that an individual had at least one recorded health insurance claim during the study period. Secondary individual-level data covering 2020–2023 were obtained from the National Statistical Office of Thailand and linked with province-level healthcare indicators from the Ministry of Public Health and regional economic indicators. The initial merged dataset contained 52,870 observations. After excluding records with missing outcomes, incomplete predictors, duplicate identifiers, and unmatched geographic information, the final analytic sample comprised 48,320 individuals from 77 provinces. Individuals were modeled as nested within provinces, and posterior distributions were estimated using Hamiltonian Monte Carlo. The Bayesian hierarchical model achieved an area under the curve of 0.862, exceeding logistic regression, random forest, and gradient boosting by 0.081, 0.039, and 0.031, respectively. It also produced the lowest root mean square error of 0.093 and Brier score of 0.072. Higher income, education, employment, healthcare access, and regional gross domestic product were associated with lower claim probability, whereas higher health expenditure was associated with greater claim probability. Posterior convergence and sensitivity analyses indicated stable estimates across alternative prior specifications. These findings demonstrate that Bayesian hierarchical modeling can improve claim-risk prediction while explicitly quantifying parameter and predictive uncertainty. The model may support risk classification, resource allocation, and equity-oriented health insurance policy in Thailand.