RT Journal Article
SR Electronic
T1 Estimating the performance of three cardiovascular disease risk scores: the Estonian Biobank cohort study
JF Journal of Epidemiology and Community Health
JO J Epidemiol Community Health
FD BMJ Publishing Group Ltd
SP 272
OP 277
DO 10.1136/jech-2017-209965
VO 73
IS 3
A1 Saar, Aet
A1 Läll, Kristi
A1 Alver, Maris
A1 Marandi, Toomas
A1 Ainla, Tiia
A1 Eha, Jaan
A1 Metspalu, Andres
A1 Fischer, Krista
YR 2019
UL http://jech.bmj.com/content/73/3/272.abstract
AB Background We aim to investigate the predictive ability of PCE (Pooled Cohort Equations), QRISK2 and SCORE (Systematic COronary Risk Estimation) scoring systems for atherosclerotic cardiovascular disease (ASCVD) risk prediction in Estonia, a country with one of the highest ASCVD event rates in Europe.Methods Seven-year risk estimates were calculated in risk score–specific subsets of the Estonian Biobank cohort. Calibration was assessed by standardised incidence ratios (SIRs) and discrimination by Harrell’s C-statistics. In addition, a head-to-head comparison of the scores was performed in the intersection of the three score-specific subcohorts.Results PCE, QRISK2 and SCORE risk estimates were calculated for 4356, 7191 and 3987 eligible individuals, respectively. During the 7-year follow-up, 220 hard ASCVD events (PCE outcome), 671 ASCVD events (QRISK2 outcome) and 94 ASCVD deaths (SCORE outcome) occurred among the score-specific subsets of the cohort. While PCE (SIR 1.03, 95% CI 0.90 to 1.18) and SCORE (SIR 0.99, 95% CI 0.81 to 1.21) were calibrated well for the cohort, QRISK2 underestimated the risk by 48% (SIR 0.52, 95% CI 0.48 to 0.56). In terms of discrimination, PCE (C-statistic 0.778) was inferior to QRISK2 (C-statistic 0.812) and SCORE (C-statistic 0.865). All three risk scores performed at similar level in the head-to-head comparison.Conclusion Of three widely used ASCVD risk scores, PCE and SCORE performed at acceptable level, while QRISK2 underestimated ASCVD risk markedly. These results highlight the need for evaluating the accuracy of ASCVD risk scores prior to use in high-risk populations.