A Bayesian binary quantile regression approach with adaptive lasso for hypertension modeling: evidence from bootstrap consistency analysis

Abstract

Hypertension remains one of the most prevalent and serious non-communicable disease worldwide due to its strong association with cardiovascular, and renal complications. This study aimed to develop a robust statistical model to identify significant predictors of hypertension status using the Bayesian Binary Quantile Regression with Adaptive Lasso (BBALQR) method. Given the complexity of health data, BBALQR offers a flexible and robust alternative to traditional models, especially in the presence of outliers and binary responses. The dataset comprises 653 patient records from Arosuka Hospital in Solok, West Sumatra, Indonesia. The proposed model was evaluated across five quantile levels \(\tau\) = 0.05, 0.25, 0.55, 0.75 and, 0.95, and parameter stability was assessed through Bootstrap resampling with 10, 25, and 50 replications. Among the tested quantiles, \(\tau\) = 0.05 yielded the lowest mean squared error (MSE), and greater Accuracy value and Press’s Q, indicating the best model performance. Furthermore, Bootstrap analysis confirmed the consistency and reliability of parameter estimates within the 95% confidence interval. Significant predictors of hypertension include age, body weight, cholesterol, and blood sugar levels. The findings suggest that BBALQR, combined with Bootstrap validation, offers a reliable modeling framework for binary health outcomes with non-normal and heteroscedastic data characteristics.