External Validation of the American Heart Association PREVENT Cardiovascular Disease Risk Equations

Abstract

Importance: The American Heart Association's Predicting Risk of Cardiovascular Disease Events (PREVENT) equations were developed to extend and improve on previous cardiovascular disease (CVD) risk assessments for the purpose of treatment initiation and patient-clinician communication.

Objective: To assess prognostic capabilities, calibration, and discrimination of the PREVENT equations in a study sample representative of the noninstitutionalized, US general population.

Design, setting, and participants: This prognostic study used data from the National Health and Nutrition Examination Survey (NHANES) 1999 to 2010 data cycles. Participants included adults for whom 10-year follow-up data were available. Data curation and analyses took place from December 2023 through May 2024.

Main outcomes and measures: Primary measures were risk estimated by the PREVENT equations, as well as risk estimates from the previous Pooled Cohort Equations (PCEs). The primary outcome was composite CVD-related mortality at 10 years of follow-up. Additional analyses compared the PREVENT equations against the PCEs. Model discrimination was assessed with receiver-operator characteristic curves and Harrell C statistic from proportional hazard regression; model calibration was determined as the slope of predicted versus observed risk.

Results: The study cohort, accounting for NHANES complex survey design, consisted of 172.9 million participants (mean age, 45.0 years [95% CI, 44.6-45.4 years]; 52.1% women [95% CI, 51.5%-52.6%]). In analyses adjusted for the NHANES survey design, a 1% increase in PREVENT risk estimates was statistically significantly associated with increased CVD mortality risk (hazard ratio, 1.090; 95% CI, 1.087-1.094). PREVENT risk scores demonstrated excellent discrimination (C statistic, 0.890; 95% CI, 0.881-0.898) but moderate underfitting of the model (calibration slope, 1.13; 95% CI, 1.06-1.21). PREVENT risk models performed statistically significantly better than the PCEs, as assessed by the net reclassification index (0.093; 95% CI, 0.073-0.115).

Conclusions and relevance: In this prognostic study of the PREVENT equations, PREVENT risk estimates demonstrated excellent discrimination and only modest discrepancies in calibration. These findings provided evidence supporting utilization of the PREVENT equations for application in the intended population as suggested by the American Heart Association.

Figure 1.. Kaplan-Meier Survival Curve Stratified by Predicting Risk of Cardiovascular Disease Events (PREVENT) Risk Score Category

The risk score categories are defined as low (<5.0%), low-moderate (5.0% to <7.5%), moderate-high (7.5% to <10.0%), and high (≥10.0%) risk. Dotted lines denote 95% CIs.

Figure 2.. Receiver-Operator Characteristic Curves Demonstrating Sensitivity and Specificity of Predicting Risk of Cardiovascular Disease Events Score for Cardiovascular (CV) and Competing Risk (CR) Mortality in the Overall Cohort

Dotted line denotes the expected performance of a perfectly random classifier.

Figure 3.. Calibration Assessed as the Slope of the Observed vs Predicted Risk Percentages According to Predicting Risk of Cardiovascular Disease Events Score

Solid lines and error bars denote the mean line of regression and 95% CIs. Dotted line denotes perfect calibration (ie, a slope of 1.0).