Predicting Systolic and Diastolic Blood Pressure Response Using Machine Learning: A 96-Feature Analysis in Hypertensive Patients with Comorbidities

Abstract

Hypertension represents a complex condition that substantially increases the global cardiovascular disease burden and related deaths. This study compares three tree-based machine learning approaches-Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)-using 96 multi-domain features to predict reductions in both systolic and diastolic blood pressure following antihypertensive treatment in patients with varying comorbidity profiles. Our approach utilizes paired t-test analyses to examine blood pressure changes before and after medication across different patient categories, while employing comprehensive decision tree visualisation to create interpretable decision pathways that identifying predictive associations between medications and blood pressure outcomes. Analysis of 160 patients indicated significant blood pressure improvements in all studied patient groups, with systolic blood pressure reductions showing statistical significance (p = 0.001) and diastolic blood pressure changes demonstrating similar significance levels (p = 0.02). The Decision Tree method showed optimal performance for systolic blood pressure prediction, recording 93% F1-score and 83% AUC values, whilst Random Forest demonstrated excellence performance in diastolic blood pressure prediction with 98% F1-score and 92% AUC. XGBoost performed less effectively than the other two algorithms across metrics. Through decision tree analysis, we identified strong predictive associations between diuretics and ACE inhibitors with systolic blood pressure reduction, whilst nitrate compounds and combined medication regimens showed significant predictive relationships with diastolic blood pressure decrease. The machine learning models successfully integrated diverse patient characteristics across multiple domains, including demographics, clinical parameters, lifestyle factors, and socioeconomic determinants. Our findings from this 160-patient cohort demonstrate the clinical utility of interpretable machine learning models for medication response prediction, providing valuable insights that can guide personalized antihypertensive therapy selection and inform clinical decision-making through data-driven treatment approaches.