Performance of PREVENT Cardiovascular Risk in Electronic Health Record-Based Clinical Practice.

Area of Science:

• Cardiology
• Public Health
• Health Informatics

Background:

• The American Heart Association introduced the PREVENT equations in 2023 as a race-free, sex-specific model for cardiovascular disease (CVD) risk prediction.
• Initial validation demonstrated strong performance, but reliability with missing data was unclear.

Purpose of the Study:

• To evaluate the discrimination and calibration of the PREVENT equations using electronic health record (EHR) data.
• To assess the robustness of the PREVENT equations when faced with missing laboratory and vital sign data.

Main Methods:

• A retrospective cohort study utilized EHR data from Duke University Health System (March 2014-December 2024).
• Two cohorts were analyzed: a 'relaxed' cohort with imputed missing data and a 'strict' cohort with complete records.
• Models included published PREVENT equations, locally fitted Cox models, neural networks, and recalibrated PREVENT models.

Main Results:

• The PREVENT equations demonstrated strong discrimination (C-index ~0.77) in both cohorts, indicating robustness to missing data after imputation.
• Calibration was better in the strict cohort, suggesting some risk underestimation in the relaxed cohort.
• Local adaptations minimally impacted discrimination but modestly improved calibration.

Conclusions:

• The PREVENT equations exhibit strong discrimination and generalizability, performing reliably even with missing data when imputation is used.
• These findings support the use of PREVENT equations for accurate CVD risk identification and ranking in clinical practice.