Interpretable machine learning models for long-term blood pressure prediction using longitudinal clinical data and SHAP analysis

While hypertension is one of the leading modifiable risk factors for cardiovascular disease, achieving adequate blood pressure control in routine clinical practice remains challenging. This is because blood pressure is influenced by multiple longitudinal clinical factors. Although machine learning (ML) approaches have increasingly been applied to the prediction of blood pressure, many existing models lack sufficient interpretability for clinical implementation. We retrospectively analyzed longitudinal clinical data from 2,544 hypertensive outpatients treated at Kan-Etsu Hospital between 2010 and 2020. Linear regression, Lasso, Ridge, and ElasticNet models were developed to predict future systolic blood pressure (SBP) and diastolic blood pressure (DBP) using demographic information, longitudinal blood pressure measurements, comorbidities, and antihypertensive medication data. Predictive performance was evaluated using root mean squared error (RMSE) and mean absolute error (MAE), while model interpretability was assessed using SHapley Additive exPlanations (SHAP) analysis and UpSet analysis. Predictive performance was comparable across all linear models, with an SBP RMSE of 17.8-17.9 mmHg and a DBP RMSE of 12.0-12.2 mmHg. SHAP analysis demonstrated that SBP difference and SBP rate change strongly contributed to SBP prediction, whereas age-related variables were dominant contributors to DBP prediction. UpSet analysis demonstrated partial overlap of important features across models, indicating the robustness and reproducibility of feature selection independent of regularization strategy. These findings suggest that interpretable linear ML models may provide a framework for understanding factors associated with future blood pressure trajectories while maintaining transparency and explainability, supporting the potential utility of explainable artificial intelligence approaches in hypertension management.