Prediction of Incident Hypertension Within the Next Year: Prospective Study Using Statewide Electronic Health Records...

Area of Science:

• Cardiovascular Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Hypertension is a prevalent, costly condition with severe health consequences like cardiovascular disease (CVD) and stroke.
• Effective management is challenging, necessitating improved predictive tools.

Purpose of the Study:

• To develop and prospectively validate a machine learning-based risk prediction model for incident essential hypertension within one year.
• To identify key predictors of developing hypertension.

Main Methods:

• Utilized electronic health records (EHRs) from the Maine Health Information Exchange for retrospective and prospective cohorts.
• Employed the XGBoost machine learning algorithm for feature selection and model building.
• Generated a predictive risk score for each individual.

Main Results:

• The 1-year hypertension risk model achieved high accuracy (AUCs of 0.917 and 0.870).
• Identified key risk factors including type 2 diabetes, lipid disorders, CVD, mental illness, clinical utilization, and socioeconomic determinants.
• High-risk individuals were often elderly with multiple chronic conditions, particularly those on mental disorder medications.

Conclusions:

• A prospective, validated 1-year risk prediction model for essential hypertension was developed using statewide EHR data.
• The real-time predictive model is deployed in Maine, offering potential for improved hypertension interventions and care.
• The model highlights disparities in social determinants of hypertension risk.