Machine learning approaches to identify influential factors associated with hypertension and prehypertension among rural adults in Bangladesh

Background: Hypertension poses a significant public health challenge in Bangladesh, particularly among rural populations with limited access to healthcare. Thus, this study aimed to identify the influential factors associated with hypertension and prehypertension in rural Bangladesh using sophisticated machine learning (ML) methods. Methods: A total of 1603 respondents were selected in this study using a multistage random sampling from a cross-sectional survey. Five commonly used sophisticated ML algorithms were used. The predictive performance of these models was evaluated using standard validation metrics. Influential variables were identified and ranked using SHapley Additive exPlanations (SHAP) technique. Results: The prevalence of hypertension and prehypertension was 30.9% and 40.8%, respectively. The XGB model outperformed other ML models in predicting hypertension (accuracy: 74.3%, ROC: 75.8%), while the LR model was better at predicting prehypertension (accuracy: 59.2%, ROC: 54.1%). Top factors in predicting hypertension were older age, being overweight or obese, having a past or no smoking history, reporting no chronic disease, and having a family history of hypertension, whereas top factors for pre-hypertension were current smoking status, absence of cardiovascular disease, being in a younger or middle-aged group, having no family history of hypertension, and current employment. Conclusion: Three out of ten people in rural areas were hypertensive, while two out of five were prehypertensive. The ML models had the potential to predict hypertension. The current findings highlight an urgent need for strengthened national and regional public health initiatives to improve hypertension detection, awareness, and management in rural Bangladesh.