Abstract
INTRODUCTION: Most current clinical risk prediction scores for cardiovascular disease prevention use a composite outcome. Risk prediction scores for specific cardiovascular events could identify people who are at higher risk for some events than others informing personalized care and trial recruitment. We sought to predict risk for multiple different events, describe how those risks differ, and examine if these differences could improve treatment priorities. METHODS: We used participant-level data from five cohort studies. We included participants between 40 and 79 years old who had no history of myocardial infarction (MI), stroke, or heart failure (HF). We made separate models to predict 10-year rates of first atherosclerotic cardiovascular disease (ASCVD), first fatal or nonfatal MI, first fatal or nonfatal stroke, new-onset HF, fatal ASCVD, fatal MI, fatal stroke, and all-cause mortality using established ASCVD risk factors. To limit overfitting, we used elastic net regularization with alpha = 0.75. We assessed the models for calibration, discrimination, and for correlations between predicted risks for different events. We also estimated the potential impact of varying treatment based on patients who are high risk for some ASCVD events, but not others. RESULTS: Our study included 24,505 people; 55.6% were women, and 20.7% were non-Hispanic Black. Our models had C-statistics between 0.75 for MI and 0.85 for HF, good calibration, and minimal overfitting. The models were least similar for fatal stroke and all MI (0.58). In 1,840 participants whose risk of MI but not stroke or all-cause mortality was in the top quartile, we estimate one blood pressure-lowering medication would have a 2.4% chance of preventing any ASCVD event per 10 years. A moderate-strength statin would have a 2.1% chance. In 1,039 participants who had top quartile risk of stroke but not MI or mortality, a blood pressure-lowering medication would have a 2.5% chance of preventing an event, but a moderate-strength statin, 1.6%. CONCLUSION: We developed risk scores for eight key clinical events and found that cardiovascular risk varies somewhat for different clinical events. Future work could determine if tailoring decisions by risk of separate events can improve care.