Decision support systems (DSS) for predicting hypertensive events using real-world telemonitoring data.

Area of Science:

• Cardiovascular disease management
• Health informatics
• Machine learning in healthcare

Background:

• Telemonitoring generates vast amounts of patient data.
• Timely analysis of telemonitoring data is crucial for proactive clinical decision support.
• Limited literature exists on applying machine learning (ML) to telemonitoring for adverse event prediction.

Purpose of the Study:

• To propose an end-to-end decision support system (DSS) framework using ML for proactive intervention.
• To predict the probability of adverse hypertensive events within 7 days using real-world telemonitoring data.
• To evaluate different ML models and optimize human-computer hybrid system development for hypertension management.

Main Methods:

• Trained and evaluated Logistic Regression, Random Forest (RF), XGBoost, and Fusion Neural Network models.
• Utilized tenfold cross-validation and metrics including AUCPR, AUCROC, and F1 score.
• Employed SHapley Additive exPlanations (SHAP) for feature importance analysis and compared daily risk score calculation methods.

Main Results:

• ML models achieved comparable AUCPR (≈75%) and AUCROC (≈87%), with XGBoost performing slightly better.
• An optimal 10-day prediction input window was identified across all models.
• Averaging results across models yielded the best composite daily risk score; key features included SBP and DBP variance.

Conclusions:

• Developed and evaluated an ML-based DSS framework for predicting adverse hypertensive events.
• Demonstrated the effectiveness of ML models in analyzing telemonitoring data for proactive alerts.
• Highlighted the importance of transparent reporting and rigorous evaluation for successful clinical decision support systems.