External validation of the Global Registry of Acute Coronary Events (GRACE) risk score: insights from the FORCE-ACS registry

Abstract Background Acute coronary syndrome (ACS) patients are a heterogeneous group with a varying risk of in-hospital and long-term mortality. Clinical risk scores play an important role in estimating these risks. Among the available risk scores, the Global Registry of Acute Coronary Events (GRACE) risk score is most used in routine clinical practice. Given the temporal improvements in both the in-hospital and long-term survival for ACS patients, continued validation of the GRACE risk score in contemporary patient cohorts is warranted. Moreover, the performance of the GRACE risk score in important subgroups with specific risk profiles has received limited attention. Purpose To assess the performance of the GRACE risk score for predicting in-hospital and one-year mortality in a contemporary, unselected cohort of ACS patients. Methods The study population consisted of ACS patients enrolled in the FORCE-ACS registry. To assess model discrimination, c-statistics were computed from the area under the receiver operator characteristic curve. Calibration was visually assessed by plotting observed versus predicted mortality and tested using the Hosmer-Lemeshow goodness-of-fit test. Indices of calibration and discrimination were also assessed in subgroups based on sex, age, type of ACS and bleeding risk (according to the PRECISE-DAPT score). Results In total, 2,587 ACS patients (median age 68 years, 28.4% women) who were enrolled in the FORCE-ACS registry between January 2015 and June 2018 were used for model validation. The in-hospital and one-year mortality rates were 2.4% and 6.3%, respectively. The discriminative ability of the GRACE risk score was good for in-hospital mortality (c-statistic 0.87, 95% CI: 0.82–0.91) and one-year mortality (c-statistic 0.82, 95% CI: 0.78–0.85) (Figure 1). The GRACE risk score predicted one-year mortality less well in patients at high risk of bleeding (c-statistic 0.68 vs. 0.78 in patients at low risk of bleeding, p=0.04). We did not observe statistically significant differences in discrimination for predicting in-hospital or one-year mortality between other subgroups. In the overall cohort, calibration of the GRACE risk score for in-hospital morality was adequate (Hosmer-Lemeshow test: p=0.11), but the GRACE risk score overestimated the in-hospital mortality risk in patients <75 years and patients at low risk of bleeding (Figure 2). The GRACE risk score overestimated the one-year mortality risk (Hosmer-Lemeshow test: p<0.01) in the overall cohort, particularly in patients <75 years, men and patients at low risk of bleeding (Figure 2). Conclusion The GRACE risk score remains a useful tool for predicting in-hospital and one-year mortality, although absolute risk at one year might be overestimated in contemporary ACS patients. Funding Acknowledgement Type of funding sources: Other. Main funding source(s): The FORCE-ACS registry is supported by grants from ZonMw, the St. Antonius Research Fund and AstraZeneca Figure 1. Discrimination of the GRACE risk scoreFigure 2. Calibration of the GRACE risk score