Comparative validation of PCE, Framingham Risk Score, Globorisk, and PREVENT for 10-year cardiovascular risk prediction in the UK Biobank

Background Cardiovascular disease (CVD) remains the leading cause of global mortality. Accurate and generalizable risk prediction models are essential for effective primary prevention. Widely used models such as the Pooled Cohort Equations (PCE), developed to estimate risk of atherosclerotic cardiovascular disease (ASCVD), the Framingham Risk Score (FRS), Globorisk, and the more recent Predicting Risk of cardiovascular disease EVENTs (PREVENT) model have demonstrated variable performance across populations. Objective To assess and compare the discrimination, calibration, and clinical utility of the PCE, FRS, Globorisk, and PREVENT models for predicting 10-year cardiovascular risk in a large contemporary UK cohort. Methods A total of 183,124 participants without prior cardiovascular disease were selected from the UK Biobank cohort and followed for up to 10 years for incident cardiovascular events. Risk scores were computed using baseline clinical variables as specified by each model: PCE (2013 ACC/AHA Pooled Cohort Equations, Goff et al.), FRS (2008 general cardiovascular risk model, D’Agostino et al.), Globorisk (UK-specific calibration, Hajifathalian et al., 2015), and PREVENT (2023 AHA equations, Khan et al.). Model performance was evaluated using receiver operating characteristic (ROC) analysis, area under the curve (AUC) with 95% confidence intervals, calibration analysis, sensitivity, specificity, positive and negative predictive values, Decision Curve Analysis (DCA), and Net Reclassification Improvement (NRI). Results Over the 10-year follow-up period, 10,560 incident cardiovascular events were recorded (5.8% event rate). PCE demonstrated the best overall discrimination (AUC = 0.717, 95% CI: 0.712–0.723), closely followed by FRS (AUC = 0.715, 95% CI: 0.709–0.721). PREVENT achieved intermediate performance (AUC = 0.668, 95% CI: 0.662–0.674), while Globorisk showed the lowest discrimination (AUC = 0.605, 95% CI: 0.599–0.611). PCE demonstrated the best calibration across risk strata. Decision Curve Analysis indicated that PCE provided the highest net clinical benefit across most decision thresholds. NRI analysis confirmed that no single model provided consistent reclassification benefit over others. Model performance declined across all models in older adults (> 59 years). Conclusion Sex-stratified analysis showed that PCE and PREVENT discriminate similarly within each sex (AUC 0.672 vs. 0.673 in men; 0.709 vs. 0.701 in women), indicating that the pooled discrimination gap between these two models is driven substantially by pooling across sexes rather than a consistent superiority of PCE. Considering all four models together, PCE and FRS demonstrated the best pooled discrimination and calibration, with PCE showing the highest net clinical benefit on Decision Curve Analysis. PREVENT, though designed with broader applicability, showed intermediate pooled performance. Globorisk demonstrated the lowest discriminatory ability in this cohort. Performance declines in older adults across all models underscore the need for age-specific recalibration and population-tailored risk prediction frameworks. Clinical trial number Not applicable.