An explainable machine learning clinical decision support system for tailored blood pressure care in adults: integrating arterial stiffness to mitigate overtreatment risks

Abstract Background Hypertension is a leading modifiable risk factor for cardiovascular disease, yet the optimal systolic blood pressure (SBP) target for an individual adult remains uncertain: fixed, aggressive targets can cause cerebral hypoperfusion, orthostatic hypotension, falls and cognitive harm, whereas the optimal target varies with biological vascular aging. Explainable machine learning could tailor blood-pressure care by integrating a marker of vascular aging, but such a clinical decision support system has not been evaluated. Methods Using the Framingham Heart Study Offspring and Third Generation cohorts (baseline 1998–2011), we developed and internally validated an explainable machine learning clinical decision support system that predicts the SBP target associated with the lowest risk of major adverse cardiovascular events (MACE). Eight algorithms were compared; features were selected with the variance inflation factor and least absolute shrinkage and selection operator (LASSO) regression; the best model was interpreted with SHapley Additive exPlanations (SHAP) and evaluated by discrimination, calibration and decision curve analysis. Results After exclusions, 3912 participants were in the development cohort and 1658 in the internal validation cohort. LASSO selected eight predictors including carotid-femoral pulse wave velocity (cf-PWV), a marker of vascular aging. Random forest performed best (development area under the receiver operating characteristic curve [AUC] 0.74; validation AUC 0.73). SHAP identified cf-PWV as the dominant feature, with a threshold effect near 9.2 m/s above which the recommended target was relaxed. Decision curve analysis showed net benefit across a broad range of threshold probabilities. Conclusions Integrating cf-PWV into an explainable machine learning framework supported tailored SBP targets for adults with clinically useful performance, offering an objective, vascular-aging-based approach to mitigate overtreatment risks. The web-based calculator translates the model into a shared-decision tool that keeps clinicians in the loop.