Background/Purpose: The Expanded Risk Model for Rheumatoid Arthritis (ERS-RA) incorporates traditional CV risk factors and RA-related measures of disease activity and has demonstrated better performance compared to a traditional cardiovascular (CV) risk factor model. Additional external validations are needed to better understand the utility of ERS-RA in different patient groups with RA.

Methods: We used data from the longitudinal Brigham and Women’s Hospital Rheumatoid Arthritis Sequential Study (BRASS). We obtained CV endpoints from three data sources: 1) self-reported CV events in BRASS, assessed annually, 2) linkage to Medicare/Medicaid claims, and 3) linkage to electronic health records (EHR). We used validated algorithms to define CV events in claims. We searched for and adjudicated CV events in EHRs with diagnosis and procedure codes. Due to uncertainty in accuracy of self-reported CV events, we included events from claims and EHR in primary analyses (narrow definition) and integrated self-reported events as a third source (broad definition). We defined follow-up as time from enrollment until first CV event or last study visit and truncated at 10 years. For the narrow CV definition, participants were censored at the visit prior to their self-reported CV event. We restricted to participants aged 20-80 years at enrollment and excluded participants with prevalent CV at baseline. We used multiple imputations with fully conditional specification to impute missing covariates. Model performance was assessed by evaluating discrimination (C-statistic) and visual inspection of calibration plots. Additionally, model performance was evaluated by dichotomizing predicted 10-year risk as < 7.5% vs. ≥7.5% and < 10% vs. ≥10%.

Results: We identified 1,332 participants in BRASS without evidence of CV events at baseline. The cohort had a mean age of 54 years; 16% were male; 8% current smokers; 6% diabetes; 22% hyperlipidemia; and 30% hypertension. RA duration was < 5 years in 37%; and 71% were rheumatoid factor positive. The mean follow-up was 81 (SD 41) months [approximately 6.8 (3.4) years]. We observed 71 (5%) participants with a CV event (narrow definition) and 166 (12%; broad definition).  We observed c-statistics of 0.659 (broad definition) and 0.810 (narrow definition) (Table 1). Calibration curves (Figure 1) indicated slight overprediction of CV risk using the narrow definition, with tighter calibration for lower predicted risk values. Conversely, using the broad definition, we observed slight underprediction of CV risk and less precision. Using a cut-point of < 7.5% we observed sensitivity (0.542 vs. 0.789), specificity (0.714 vs. 0.708), positive predictive value (PPV, 0.212 vs. 0.132), and negative predictive value (NPV, 0.916 vs. 0.984) for broad and narrow definitions, respectively. Results were similar for a cut-point of 10% predicted risk.

Conclusion: In this cohort of individuals with established RA the ERS-RA generally demonstrated good discrimination and calibration for CV events defined using validated algorithms in claims and EHR; however, its precision was lower when self-reported CV events were included.

Table 1. Model Performance of the ERS-RA for Broad and Narrow CV Definitions

Figure 1. Calibration Curves for Broad and Narrow CV Definitions

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/ers-ra-as-a-tool-for-cardiovascular-risk-prediction-in-established-rheumatoid-arthritis-an-external-validation/