To develop a predictive model for undiagnosed hypertension (UHTN) in older adults based on five modifiable factors [eating behaviors, emotion, exercise, stopping smoking, and stopping drinking alcohol (3E2S) using machine learning (ML) algorithms.
The supervised ML models [random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB)] with SHapley Additive exPlanations (SHAP) prioritization and conventional statistics (χ2 and binary logistic regression) were employed to predict UHTN from 5,288 health records of older adults from ten primary care hospitals in Thailand.
The χ2 analyses showed that age and eating behavior were the predicting features of UHTN occurrence. The binary logistic regression revealed that taking food supplements/vitamins, using seasoning powder, and eating bean products were related to normotensive and hypertensive classifications. The RF, XGB, and SVM accuracy were 0.90, 0.89, and 0.57, respectively. The SHAP identified the importance of salt intake and food/vitamin supplements. Vitamin B6, B12, and selenium in the UHTN were lower than in the normotensive group.
ML indicates that salt intake, soybean consumption, and food/vitamin supplements are primary factors for UHTN classification in older adults.