Bayesian binary quantile regression adaptive Lasso for modelling hypertension: a bootstrap based consistency analysis

Hypertension remains a major public health challenge in developing countries, with its prevalence continuing to rise and risk factors varying across populations. Traditional logistic regression often fails to capture distributional heterogeneity in health outcomes, thereby motivating the use of more flexible approaches. This study applied Bayesian Binary Quantile Regression with adaptive Lasso (BALBqr) to hypertension data from 653 patients at Arosuka Solok Hospital, Indonesia. The method was further integrated with bootstrap resampling as a consistency check for posterior estimates. The Bayesian quantile framework enabled the evaluation of covariate effects across different quantiles of the conditional distribution, while adaptive Lasso improved interpretability by addressing multicollinearity and ensuring parsimonious variable selection. Bootstrap validation complemented posterior inference by assessing the stability of parameter estimates within bootstrap confidence intervals, with the optimal replication identified at 250 iterations based on the smallest bias relative to the posterior mean. Simulation results demonstrated that the proposed approach consistently outperformed conventional methods in terms of bias, RMSE, and MAE across various error distributions. Based on this study, age, body weight, cholesterol, triglycerides, and blood sugar levels were found to be statistically significant predictors of hypertension status at the 0.95 quantile.