A machine learning-based framework for predicting hypertension using serum hematological factors

Hypertension (HTN) is a leading global cause of cardiovascular disease (CVD) and all-cause mortality, underscoring the need for early detection and intervention. This study aimed to develop a machine learning (ML)-based predictive framework for HTN using routinely available hematologic and clinical parameters. We analyzed data from the Mashhad Stroke and Heart Atherosclerotic Disorder (MASHAD) study (2010-2020). From an initial 9,704 participants, 4,923 individuals (3,033 without and 1,890 with incident HTN) were included after applying predefined inclusion/exclusion criteria and preprocessing. Predictors included hematologic biomarkers (WBC, RBC, MCV, PLT, RDW, PDW, NLR) and clinical/demographic variables (age, sex, smoking, BMI, GFR, physical activity). Missing values (< 10% for most variables) were addressed via median imputation for continuous features. We employed multiple ML algorithms (XGBoost, LightGBM, logistic regression, decision trees, random forests, gradient boosting, k-nearest neighbors, naive Bayes, ExtraTrees, AdaBoost) to develop and compare models. Following model training and performance-based ranking, XGBoost emerged as the top-performing classifier, achieving a ROC-AUC of 0.66 (95% CI: 0.63-0.69). Feature importance analysis identified age, BMI, RBC count, and MCV as the most influential predictors, with consistent rankings across other models. All models exhibited stable performance across training and test sets, indicating minimal overfitting. Our findings demonstrate an ML-driven framework exploring potential signals in HTN risk using routinely available clinical and hematologic data. While the predictive performance reflects the inherent challenge of forecasting a multifactorial disease from baseline variables, this work establishes a transparent, interpretable exploratory benchmark and identifies key modifiable predictors for future validation and enhancement in more comprehensive datasets.