ASGARD risk scores for long-term, individualized cardiovascular surveillance of low-risk patients with asymptomatic non-severe aortic valve stenosis: development and internal validation

Abstract

Abstract Background Routine echocardiography of asymptomatic patients with non-severe aortic valve stenosis (AS) mainly focuses on the valve condition at 1-2 year intervals but has limited ability to predict the risk of other cardiovascular comorbidities. Thus, there is a compelling need to replace routine echocardiography for low-risk patients with more individualized cardiovascular surveillance for a safe period. Aim First, to develop and validate separate risk stratification models for AS-related events (AVE) and major cardiovascular events (MCE) to identify low-risk AS patients who can safely postpone routine echocardiography replaced by tailored risk stratification and adequate risk factor management. Second, to estimate optimal follow-up intervals according to risk stratification for each outcome. Methods We analyzed serial measurements from 69% of 1579 patients with mild-to-moderate AS who were asymptomatic and event-free one year after inclusion in the randomized, multicenter Simvastatin and Ezetimibe in Aortic Stenosis (SEAS) trial. We developed two ASGARD risk scores using multivariable Cox proportional hazard regression analyses, designed as landmark analyses starting at year-1 and with a clinically relevant 2-year follow-up. The ASGARD scores estimated the risk of AVE (composite of aortic valve replacement or hospitalization with heart failure) and MCE (composite of myocardial infarction, percutaneous coronary intervention, coronary artery bypass graft, stroke, and all-cause mortality). We estimated the prognostic models with backward elimination, selecting the lowest Akaike information criterion values and p<0.05. There were 7 predictors for AVE and 6 for MCE among current demographic, electrocardiograms, laboratory data, and echocardiographic measures one year earlier (baseline) (Figures 1A, 2A). We defined optimal follow-up for risk score quartiles at rates ≤5% visualized on Kaplan-Meier plots. We assessed model performance in 31% of patients held out for internal validation. Results ASGARD development models for both AVE and MCE risk demonstrated overall consistent predictive performance in the internal validation cohort, as evidenced by c-statistics, calibration-in-the-large, calibration slope, Brier score, and goodness-of-fit (all p>0.05) (Figure 1B-C, 2B-C). ASGARD predicted AVE risk similar to a new Vmax measurement (area under the curve [95% CI], 0.82 [0.78-0.87] vs. 0.85 [0.82-0.89) (Figure 1B), and MCE risk better than Framingham cardiovascular risk score (0.76 [0.70-0.82] vs. 0.68 [0.62-0.75]) (Figure 2B). The one-year event rates were ≤5% for the 2 lowest risk score quartiles (Figure 1D, 2D). Conclusion The ASGARD risk scores show promise for long-term, individualized cardiovascular surveillance of low-risk patients with non-severe AS without unnecessary echocardiography. Further studies are warranted for external model validation, continuous optimization, and exploring the potential for implementation in clinical practice.Figure 1Figure 2