External Validation and Head-to-Head Comparison of Cuore and Score2 Cardiovascular Risk Scores in an Italian Primary Care-Administrative Cohort.
To externally validate and compare CUORE and SCORE2 for 10-year prediction of first fatal or non-fatal myocardial infarction or stroke in an Italian primary care cohort, and to quantify the impact of administrative endpoint definition and model recalibration.
We studied 38,101 adults in ATS Milan (baseline 2007-2009) with primary care risk factors linked to hospital discharge and mortality registries for 10-year follow-up. The endpoint was first fatal or non-fatal myocardial infarction or stroke, ascertained using a strict (primary diagnosis/procedure fields) and a broad (any diagnosis/procedure field) administrative definition. Discrimination was assessed with IPCW time-dependent AUC and calibration with decile plots, calibration-in-the-large and calibration slope, comparing uncalibrated and recalibrated implementations. SCORE2 AUC ranged 0.713-0.720, whereas CUORE ranged 0.730-0.741. Broad ascertainment increased event rates (SCORE2: 5.6% vs 4.4%; CUORE: 6.8% vs 5.0%) without materially changing AUC. SCORE2 showed more stable calibration, whereas CUORE showed calibration slopes <1.
In this real-world Italian cohort, CUORE and SCORE2 provided robust ordinal risk stratification and moderate discrimination at 10 years. Absolute risk estimates were sensitive to administrative endpoint mapping and recalibration strategy, supporting transparent endpoint reporting and local calibration assessment before applying risk thresholds in practice.
We studied 38,101 adults in ATS Milan (baseline 2007-2009) with primary care risk factors linked to hospital discharge and mortality registries for 10-year follow-up. The endpoint was first fatal or non-fatal myocardial infarction or stroke, ascertained using a strict (primary diagnosis/procedure fields) and a broad (any diagnosis/procedure field) administrative definition. Discrimination was assessed with IPCW time-dependent AUC and calibration with decile plots, calibration-in-the-large and calibration slope, comparing uncalibrated and recalibrated implementations. SCORE2 AUC ranged 0.713-0.720, whereas CUORE ranged 0.730-0.741. Broad ascertainment increased event rates (SCORE2: 5.6% vs 4.4%; CUORE: 6.8% vs 5.0%) without materially changing AUC. SCORE2 showed more stable calibration, whereas CUORE showed calibration slopes <1.
In this real-world Italian cohort, CUORE and SCORE2 provided robust ordinal risk stratification and moderate discrimination at 10 years. Absolute risk estimates were sensitive to administrative endpoint mapping and recalibration strategy, supporting transparent endpoint reporting and local calibration assessment before applying risk thresholds in practice.