Clinical trial–derived risk model may not generalize to real-world patients with acute coronary syndrome

Background Accurate risk stratification can guide clinical decision-making in the management of acute coronary syndromes (ACS). However, the applicability of risk models to the general ACS population remains unclear. The purpose of this study was to validate and compare a modified international clinical trial and a registry-based risk model in a contemporary, less selected ACS population. Methods In the prospective, observational Canadian ACS Registry, 4627 patients with ACS were enrolled from 51 centers. Baseline patient data were recorded on standardized case report forms. We evaluated risk models derived from the Platelet glycoprotein IIb/IIIa in Unstable angina: Receptor Suppression Using Integrilin Therapy (PURSUIT) and the Global Registry of Acute Cardiac Events (GRACE) predicting in-hospital death among patients with non–ST-elevation ACS. Model discrimination was measured by the c-statistic, and calibration was assessed graphically and by the Hosmer-Lemeshow goodness-of-fit test. Results In-hospital mortality rates were 2.4% overall and 1.5% among the patients with non–ST-elevation ACS (n = 2925; 63.2%) in our validation cohort. Both the in-hospital PURSUIT and GRACE risk models showed similar and good prognostic discrimination (c-statistics = 0.84 and 0.83, respectively; P = .69 for difference). The GRACE model also demonstrated good calibration (Hosmer-Lemeshow P = .40). In contrast, calibration in the PURSUIT model was poor (Hosmer-Lemeshow P < .001), with consistent overestimation of risks. Conclusions Both the PURSUIT and GRACE models demonstrated good discrimination for in-hospital mortality rates in the Canadian ACS Registry. However, the GRACE risk model, derived from a less selected population, provided superior calibration in risk assessment across the spectrum of ACS. Our findings underscore the potential importance of risk model validation in the general ACS population rather than a clinical trial population to establish its generalizability before integration into clinical practice.