Abstract
BACKGROUND: Despite clinical need, personalized risk scores for established atherosclerotic cardiovascular disease (ASCVD) are scarce.
OBJECTIVES: The objective of the study was to develop and validate 10-year residual risk scores for cardiovascular death in patients with established ASCVD.
METHODS: Prospective observational cohort study. Models developed and validated in U.K. Biobank (UKB) (baseline 2006-2010; follow-up through 2021) and externally validated in Mass General Brigham cohort (MGB) (baseline 2007; follow-up through 2018). Analyzed on October 2022-February 2024. Eligible participants had established ASCVD. Guideline-based clinical factors plus additional biomarkers and self-reported health ratings.
PRIMARY OUTCOME: 10-year cardiovascular death. Elastic-net Cox and gradient-boosted tree models. C statistics and goodness-of-fit assessed in holdout set.
RESULTS: UKB: 32,994 participants (mean age 61; 11,727 [35.5%] women; 2,660 [8.0%] cardiovascular deaths), with 9,899 (30%) randomly selected as a holdout validation set. MGB: 54,969 patients (mean age 71; 22,738 [41.4%] women; 6,927 [12.6%] cardiovascular deaths). Median follow-up: 10 years (IQR 10-10) in UKB; 9.4 years (5-10) in MGB. We developed 2 residual risk scores, RRS16 and RRS24, incorporating 16 and 24 algorithmically selected factors. RRS16 used routinely available factors; RRS24 also included self-reported health and additional biomarkers. RRS16 achieved C statistics of 0.752 (95% CI: 0.736-0.768) in UKB and 0.750 (0.744-0.756) in MGB, outperforming the 2018 the American Heart Association (AHA) guideline model (0.658 [0.642-0.674] in UKB, 0.580 [0.574-0.586] in MGB). RRS24 achieved 0.784 (0.768-0.800) in UKB. Both models well calibrated (P > 0.1). RRS16 calculator: https://mora.bwh.harvard.edu/rrs16/.
CONCLUSIONS: RRS16 and RRS24 outperformed AHA guideline model in estimating the residual risk in patients with established ASCVD. Both are clinically applicable but require further validation in diverse populations.