Diagnostic Accuracy of Cardiovascular Disease Prediction Models: A Systematic Review and Meta‐Analysis of Validation Studies

Résumé fourni par la source

Background and Aim: Cardiovascular diseases (CVD) are widely accepted to be the most serious health care problem in the world. We performed a diagnostic test accuracy systematic review and meta-analysis to establish the real-world performance of established CVD risk prediction tools, providing high-certainty evidence to optimize primary prevention strategies in resource-constrained healthcare systems worldwide. Methods: Databases including PubMed, Scopus, and Web of Science were systematically searched for studies published from January 1, 2013, to December 30, 2024. A total of 58 cohort studies involving 7.5 million participants were ultimately included. Data on true positives, false positives, true negatives, and false negatives were extracted to compute pooled sensitivity, specificity, positive and negative likelihood ratios, diagnostic odds ratios (DOR), and area under the receiver operating characteristic curve (AUC). Summary receiver operating characteristic (SROC) curves were constructed, publication bias was evaluated using Deeks' funnel plot asymmetry test, and clinical utility was assessed through Fagan nomograms. Analyses were performed in STATA version 18. Results: Across 58 studies, 47,276 CVD events occurred in men and 33,931 in women. Pooled sensitivity and specificity were 0.79 (95% CI: 0.77-0.89) and 0.66 (95% CI: 0.50-0.79) for the Framingham Risk Score (FRS); 0.74 (95% CI: 0.68-0.80) and 0.79 (95% CI: 0.43-0.95) for the ACC/AHA model; 0.77 (95% CI: 0.57-0.89) and 0.69 (95% CI: 0.43-0.89) for SCORE; and 0.85 (95% CI: 0.81-0.88) and 0.80 (95% CI: 0.79-0.81) for COX models. The highest pooled AUCs were observed for SCORE (0.81, 95% CI: 0.77-0.84) and COX (0.89, 95% CI: 0.89-0.92), indicating superior discrimination. Conclusions: Based on pooled AUCs, the most reliable threshold-independent metric, SCORE and COX demonstrated superior performance, followed by FRS. While sensitivity/specificity and DOR provide classification insights, AUC offers the most robust basis for cross-model comparison. Future research should focus on validating and comparing established models, adapting them locally, and integrating novel predictors, rather than developing redundant ones.

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