EXTERNAL VALIDATION OF OFFICE AND LABORATORY-BASED CARDIOVASCULAR RISK SCORES FOR INCIDENT CARDIOVASCULAR DISEASE AND MORTALITY IN 118,564 ADULTS IN MEXICO

Résumé fourni par la source

ASCVD/CVD Risk Assessment Cardiovascular disease (CVD) is the leading cause of death in Mexico, and risk prediction models are crucial for CVD prevention. While several models are endorsed in international guidelines, most were developed for non-Latin American populations, highlighting the need for validation in Mexican populations. This study aims to assess the external validity of various office-based and laboratory-based CVD risk models in Mexico. We conducted an external validation using data from a prospective study of 118,564 adults aged >40 years. We evaluated office-based and laboratory-based models for predicting fatal events (NFOB-F, NFLBF, FLB) and fatal plus non-fatal events (NFOB, NFLB) over 10 years of follow-up. Events were identified through death registries and self-reports in a resurvey. Model performance was assessed using discrimination (C-statistics, AUC), calibration (mean, weak and moderate calibration), and clinical utility (Decision Curve Analysis). We identified 6,758 (4.4%) total CVD events, including 6,314 (4.1%) fatal events and 461 (0.3%) non-fatal events in resurveyed participants (n=8,061). The highest AUROC values were observed for SCORE2 in the NFLB-F subset (0.821, 95%CI 0.794–0.829) and Globorisk in the FLB subset (0.818, 95%CI 0.800–0.836). In contrast, the Framingham model consistently performed worse across all subsets. Uno’s C-statistic did not significantly differ from the AUC, and Harrel’s C-statistic produced nearly identical results. Calibration assessments showed low mean calibration for most models, with WHO models performing better. Weak calibration was poorest for the Framingham model across subpopulations. Moderate calibration revealed that most models performed well for predicting risks up to 5% or 7.5%, but overpredicted higher-risk cases. Finally, clinical net benefit analyses indicated that the clinical utility of these models, when compared to chronological age, was small. Overall, the models showed variable predictive capacity, with some models demonstrating better performance in certain subsets. The study validated CVD risk models in Mexico, finding variable performance, with some models showing better predictive capacity in specific subsets.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.