Real-World Pharmacotherapy-Driven Cardiovascular Risk Prediction Using Interpretable Machine Learning and Jordanian...

Area of Science:

• Machine learning applications in healthcare
• Cardiovascular disease risk prediction
• Health informatics

Background:

• Cardiovascular disease (CVD) is a leading global cause of death, particularly in low- and middle-income countries.
• Existing risk models often lack accuracy and generalizability in these regions.
• Electronic health records (EHRs) offer a rich data source for improving risk prediction.

Purpose of the Study:

• To develop an interpretable machine learning model for cardiovascular risk prediction.
• To integrate pharmacotherapy data into cardiovascular risk assessment.
• To utilize national EHR data from Jordan for model development.

Main Methods:

• Retrospective cohort study of ~600,000 individuals from the Hakeem EHR system (2018-2022).
• Integration of demographic, clinical, blood pressure, laboratory, and medication data.
• Benchmarking, optimization, and calibration of machine learning models, including SHAP analysis for interpretability.

Main Results:

• High prevalence of hypertension (50.2%), hyperlipidemia (54.9%), and diabetes (47.9%) observed.
• The JoRisk model, a calibrated seed-bagged gradient boosting model, achieved strong performance (ROC-AUC 0.844).
• Key predictors included antihyperlipidemic therapy, blood pressure variability, age, and sex.

Conclusions:

• JoRisk provides a calibrated, interpretable machine learning framework for short-term cardiovascular risk prediction.
• The model effectively uses routinely available EHR variables and pharmacotherapy data.
• JoRisk serves as a scalable decision-support tool for risk stratification in resource-constrained healthcare systems.