Abstract 4145608: Calibration and Discrimination of Cardiovascular Risk Scores in Women with Breast Cancer in the Women’s Health Initiative

Introduction: Women with breast cancer (BC) are at increased risk for cardiovascular disease (CVD). Currently, there is no standardized approach to assess CVD risk in women with BC, despite recommendations for risk assessment. We estimated and compared the performance of established ASCVD risk scores (Framingham Risk Score [FRS], Atherosclerotic Cardiovascular Disease [ASCVD] Risk Estimator, and Life’s Essential 8 [LE8]) in a cohort of postmenopausal women with BC. Methods: Women who were diagnosed with BC (stage I-III) and free of CVD from the Women‘s Health Initiative were included in this study. The primary outcome was 10-year risk of ASCVD, defined as a composite of adjudicated coronary heart disease, myocardial infarction, or stroke. For each risk score, we examined calibration and discrimination. Calibration was assessed by calculating bias, slope, standard error of the estimate (SEE), and root mean square deviation (RMSD). Discordance was defined as the percent difference between expected and observed events. Discrimination was evaluated using the c-statistic. We reassessed calibration and discrimination after recalibrating the FRS and ASCVD scores. Results: In 6,946 participants (mean age BC diagnosis: 70 years; 89.7% White), the 10-year incidence of ASCVD was 8.6%. FRS underestimated ASCVD risk by 31.1%, whereas ASCVD and LE8 scores overestimated the risk by 53.7% and 9.3%, respectively (Hosmer-Lemeshow P<0.001 for all scores) (Figure). Discordance was consistent throughout the risk continuum for FRS and ASCVD. However, for LE8, discordance was minimal for those at moderate- to high- risk and overestimated in those at low-risk. The ASCVD risk score had the highest c-statistic of 0.76, compared to 0.62 and 0.71 for the FRS and LE8, respectively (Figure). After recalibration, the discordance for FRS and ASCVD was reduced where the FRS overestimated the risk by 3.7% and the ASCVD overestimated the risk by 37.3%. There were no changes in c-statistics. Conclusions: In this analysis, LE8 had the best calibration, whereas ASCVD had the best discrimination. The ASCVD risk score had a c-statistic of 0.76, suggesting it may be a practical tool to aid in 10-year CVD risk stratification in women with BC.