A machine learning approach to predict hypertension using cross-sectional & two years follow up data from a health & demographic cohort of Assam, North East India

Background & objectives Hypertension affects a sizable section of the world population and is being recognised as a growing problem. Its prediction using machine learning (ML) algorithms, will add to its control and prevention. The objective of the present investigation was to check the applicability of ML approaches in the prediction and detection of hypertension. Methods We included 53,301 participants at baseline from a health and demographic surveillance system in Dibrugarh, Assam (Dibrugarh-HDSS). We constructed two models, one at baseline and the other after two years of follow-up. Of the total participants (baseline: 29,402; follow up: 4,400), 70 per cent were randomly selected to fit seven popular classification models namely decision tree classifier (DTC), random forest classifier (RFC), support vector machine (SVM), linear discriminant analysis (LDA), logistic regression, Ada-boost classifier, and XG boost classifier. The data from the remaining 30 per cent were used to evaluate the performance of the models. Results In the baseline data, the Ada-boost classifier could identify hypertension with a maximum accuracy score of 87.02 per cent (CI: 86.01-88.03). The maximum area under the curve (AUC) score of 98.37 per cent (CI: 97.36-99.38) was obtained under RFC. For the prediction of risk at two years, the maximum average accuracy score of 77.57 per cent (CI: 76.6-78.54) was achieved under X-G Boost followed by RFC (77.2%, CI: 76.15-78.25) and a maximum AUC of (85.82%, CI: 84.88-86.76) was obtained under RFC. Interpretation & conclusions In both the identification and prediction of hypertension, RFC was found to be better than the other classifiers. 'Waist circumference' followed by 'body mass index' (BMI) were found to have maximum relative importance in the identification of hypertension, while in the case of two-year risk prediction, the baseline 'systolic blood pressure' (SBP), diastolic blood pressure (DBP), and 'BMI' had the maximum relative importance. The findings revealed the potential of predictive models in accurately identifying high-risk individuals, enabling timely interventions, and optimising clinical decision-making.