External validation and recalibration of office- and laboratory-based cardiovascular risk scores for prediction of 10-year risk of fatal cardiovascular disease in 112,262 adults in Mexico City

BACKGROUNDCardiovascular disease (CVD) is the leading cause of mortality in Latin America, yet most CVD risk prediction models were developed in high-income countries with limited validations in these settings. OBJECTIVESTo externally validate CVD risk prediction models for 10-year fatal CVD, and recalibrate the Globorisk-fatal model in Mexican population. METHODSWe analyzed 112,262 adults [≥]40 years from the Mexico City Prospective Study. Outcomes were restricted to fatal CVD, including myocardial infarction (MI) and stroke, censored at 10 years. CVD risk was estimated using the laboratory- and office-based Framingham, Globorisk, Globorisk-LAC, WHO, and SCORE2 equations. Discrimination was assessed with Harrells c-statistic and AUROCs, calibration with mean estimates, slopes, and calibration curves, and overall performance with Brier scores. Sex-specific recalibration of the Globorisk-fatal model was performed using observed 10-year risks. RESULTSDuring 10 years of follow-up, 2,429 fatal CVD events were recorded (1,667 MI, 762 stroke). All models showed good discrimination, with c-statistics ranging from 0.761-0.805 in men and 0.797-0.831 in women. The Globorisk-fatal model had the highest c-statistic in women (0.831, 95%CI 0.821-0.841), and the laboratory-based WHO-MI model in men (0.805, 95%CI 0.783-0.827). Despite this, all equations consistently overestimated CVD risk, particularly in women. Calibration analyses revealed systematic overprediction at higher risk levels, more pronounced in men. Recalibration of the Globorisk-fatal model improved agreement between predicted and observed risks, reducing overestimation. CONCLUSIONSCVD risk models showed good discrimination but consistently overestimated risk in this Mexican cohort. The recalibrated Globorisk-fatal model improves risk estimation of fatal CVD in Mexican adults.