504 Personalizing cardiovascular risk prediction for SLE patients. (14th December 2022)

• Record Type: Journal Article
• Authors: Choi, May Y; Weber, Brittany; Guan, Hongshu; Yoshida, Kazuki; Li, Daniel; Eldrodt, Jack; Stevens, Emma; Cai, Austin; Everett, Brendan; Costenbader, Karen H
• Is Part Of: Lupus science & medicine. Volume 9(2022)Supplement 3
• Page Start: A22
• Page End: A23
• Publication Date: 2022-12-14
• Subjects: Systemic lupus erythematosus -- Periodicals; 616.772005
• DOI: 10.1136/lupus-2022-lupus21century.21
• Languages: English
• ISSNs: 2398-8851

Abstract

Abstract : Objective: The risk of cardiovascular disease (CVD), including myocardial infarction (MI) and stroke, is increased in SLE patients and is underestimated by current prediction algorithms designed for the general population including the 10-year atherosclerotic cardiovascular disease (ASCVD) risk score. The American College of Cardiology/American Heart Association now considers systemic inflammatory diseases such as SLE as risk enhancers for CVD. The purpose of this study was to develop an SLE-specific prediction tool to provide a more accurate estimate of CVD risk by including both traditional and SLE-related CVD risk factors. Methods: We included SLE patients enrolled in the Brigham and Women's Hospital SLE Cohort and collected one-year baseline data on traditional CVD risk factors, demographic and clinical features from the electronic medical record at cohort enrollment. Disease activity was rated using a modified physician global assessment (PGA) tool and SLE-related variables including autoantibodies, complement levels, and SLE manifestations were also collected. All subjects were required to have one or more visits for SLE during the baseline period. A up to ten-year follow-up period for CVD events began day +1 at end of baseline period (index date). The primary outcome was first major adverse cardiovascular events (MACE) defined as composite of first myocardial infarction (MI), stroke, or cardiac death, in the follow-up period. These were identified by ICD-9/10 codes and adjudicated by medical record review by board-certified cardiologists as either definite or probable events (not meeting all the criteria for MI or stroke definition). The secondary outcome was boarded to include first event of: carotid artery occlusion or stenosis, transient ischaemic attack, atrial fibrillation/flutter, heart failure, peripheral vascular disease, or angina pectoris. We excluded subjects with CVD events prior to the index date. Three Cox regression risk prediction models that categorized patients into low risk <7.5% risk, moderate risk 7.5-20%, and >20% risk over 10 years were derived: 1) primary outcome with definite/probable events, 2) combined model 1 and secondary outcomes, and 3) primary outcome with definite events only. We performed least absolute shrinkage and selection operator (LASSO) regression for variable selection and required one of the candidate predictors to be the 10-year ASCVD risk score. We assessed model performance using integrated time-dependent area under the curve, Harrell's C statistic, optimism corrected C- statistic, integrated discrimination (IDI), and net reclassification index (NRI) using bootstrap resampling. Results: We included 1243 patients; 93.0% female and mean age of 41.6 (SD 13.3) years. There were 90 definite and probable MACEs (46 MIs, 36 strokes, and 19 cardiac deaths) and 211 secondary events over the follow-up period. The variables selected included: ASCVD risk score, disease activity (PGA at most recent baseline visit), disease duration, creatinine level, presence of anti-dsDNA, anti-RNP, lupus anticoagulant, anti-Ro60/SSA, and low C4 (table 1). Models 1 (primary outcomes with definite and probable events) and 3 (primary outcomes with definite events only) performed similarly and outperformed model 2 (combination of model 1 and secondary events) (table 2). Model performance improved in comparing risk predicted by ASCVD risk score alone vs. ASCVD risk score combined with selected SLE variables by LASSO regression for models 1 and 2, particularly at year 1. For these models, the number of SLE patients who were classified as high risk (>20%) more than doubled when selected SLE variables were added to the ASCVD model compared to the ASCVD model alone (table 3). The ten-year IDI and NRI were significant in the improvement direction. Conclusion: Our novel SLE-specific cardiovascular risk prediction scores enhanced the performance of the traditional ASCVD risk algorithm and identified a greater number of SLE patients (at least two-fold) at high-risk for CVD events over 10 years. These models will need to be validated in a larger and more diverse population of SLE patients.