Comparative validation of the AHA PREVENT equation with US and European cardiovascular risk equations among 892,000 adults

Abstract Hypothesis and Purpose An American Heart Association scientific advisory group recently developed the Predicting Risk of cardiovascular disease EVENTs equations (PREVENT). PREVENT was validated among US adults, but its performance for predicting 10-year risk of atherosclerotic cardiovascular disease (ASCVD) has not been compared with the pooled-cohort equations (PCEs) or Systematic COronary Risk Evaluation 2 (SCORE2) equations in populations outside the US. We sought to compare these equations using data from Clalit Health Services, a health maintenance organization comprising the majority of the Israeli population. Population Studied and Sample Size We collected data from 975,665 Israeli adults aged 40-79 years with continuous enrollment throughout 2013 and no prior ASCVD (defined as hospital diagnoses or permanent outpatient diagnoses of myocardial infarction or stroke). We excluded 82,602 (8.5%) with missing equation inputs, and 1,371 (0.14%) with disenrollment during followup for a final study population of 891,692 adults (91.4%). Study Design and Methods We collected laboratory and examination data required for estimating ASCVD risk between 2011-2013 (Table 1). We then calculated 10-year ASCVD risk estimates using the PCEs, PREVENT, and SCORE2 equations, compared with observed 10-year incidence of ASCVD, and compared against risk thresholds used to recommend initiation of statin therapy. Outcomes The PCEs and SCORE2 had higher discriminative accuracy relative to PREVENT (C statistics of 0.726 and 0.728 versus 0.714). The PCEs were associated with the highest mean ASCVD risk (10.3% versus 4.9% for PREVENT and 5.8% for SCORE2), best calibration slopes (0.95 versus 1.58 and 1.70), and highest proportion of individuals meeting risk-based thresholds recommended by the European Society of Cardiology (21% versus 3.1% and 5.5%; Table 2). Using only hospital diagnoses to define ASCVD resulted in improved calibration of PREVENT and SCORE2 and overestimation by the PCEs (data not shown). All pairwise differences were significant after Bonferroni correction. Conclusions Our data show noninferior calibration and discrimination of the PCEs relative to newer US and European risk equations. These results were sensitive to the different approaches for defining ASCVD, with substantial implications for guideline-based care. Further validation and potential recalibration should be considered prior to global implementation of newer ASCVD risk equations.