The Predicting Risk of cardiovascular Events (PREVENT) model, developed by the American Heart Association, had a moderate discrimination capacity for the incidence of atherosclerotic cardiovascular disease (ASCVD) across four geographically distinct academic health care systems in the U.S., yet calibration metrics varied widely across health care systems, sociodemographics and underlying cardiometabolic comorbidities, according to a study published July 14 in JACC.

The discrimination and calibration of the PREVENT equations were evaluated by So Mi Cho, PhD, et al., through a retrospective review of data from the electronic health records of 270,320 patients from four integrated U.S. health care systems: Mass General Brigham (MGB), Mount Sinai Health, Penn Medicine and Vanderbilt University Medical Center. The patients were between 30 and 79 years old and none had ASCVD diagnosed between 2010 and 2014.

Calibration was assessed using the Greenwood-Nam-d'Agostino test and the time-dependent Harrell's C-index was used to assess discordance and mean calibration and discrimination, and the first ASCVD event over the 10-year follow-up was observed.

Results showed that the mean estimated 10-year ASCVD risk was 5% in the MGB patients, 6% in the Mount Sinai patients, 6% in the Penn patients and 5% in the Vanderbilt patients. Their mean age was about 55 years and about half of the patients were women.

Notably, the PREVENT model underestimated the observed incidence rate in MGB, Mount Sinai and Vanderbilt patients, but more closely mirrored the empirical rate in Penn. Overall, PREVENT yielded a moderate discrimination C-index in all four health care systems. Calibration differed by sex, with greater underestimation among women in MGB patients (discordance −81%) and Vanderbilt patients (discordance −55.5%) but among men in Mount Sinai patients (discordance −41%).

The ability of PREVENT to predict risk in patients stratified by race and ethnicity varied across the health care systems. Compared with the pooled cohort equations, PREVENT demonstrated better overall calibration in Penn patients, but worse in MGB and Mount Sinai patients. Additionally, discrimination was better among patients without diabetes or taking antihypertensives.

"The novel PREVENT framework demonstrated heterogeneous calibration and discrimination abilities by individual- and population-level sociodemographics, risk distributions and practice settings," write the authors. "These results highlight the need for individualized interpretation and local evaluation and calibration of available risk prediction tools for optimal clinical decision-making."

"...the [AHA] PREVENT equations represent an important step forward in cardiovascular risk prediction," write Yuan Lu, ScD, FACC, and Khurram Nasir, MBBS, MPH, MSc, FACC, in an accompanying editorial comment. They note that this external validation study underscores a persistent gap between model innovation and practical implementation. "…Closing the gap between prediction and prevention will require collaboration across disciplines – and a commitment to ensuring that all patients receive timely, tailored, and data-informed cardiovascular care."