FRI0057 A MODEL FOR QUANTIFYING THE EFFECT OF INFLAMMATION ON CARDIOVASCULAR DISEASE RISK PREDICTION IN RA PATIENTS

Abstract

Background:Patients with rheumatoid arthritis (RA) are at increased risk for cardiovascular disease (CVD)[1]. Quantifying the effect of inflammation on CVD risk is important because rheumatologists can reduce inflammation with effective RA medications. A new score has been developed for predicting the risk for a CVD event (MI, stroke or CV death) in RA patients. It combines serological measures of inflammation (the multi-biomarker disease activity [MBDA] score, a measure of RA disease activity; and three individual biomarkers [TNF-RI, MMP-3 and leptin]), with age and four conventional CVD risk factors (smoking, hypertension, diabetes and history of a high- risk CVD condition)[2]. To gain insight into the potential effect that treating inflammation may have on the CVD risk score, it would be useful to know how the score is affected by the level of inflammation.Objectives:Explore the quantitative contribution of inflammation to CVD risk score in individual RA patients.Methods:To quantify the effect of inflammation on the CVD risk score across a range of MBDA scores, a commercial dataset of 177,486 RA patients with ≥2 MBDA tests between October 2010 and June 2019 was split 2:1 into training and validation datasets. Curves showing variation in the CVD risk score across the spectrum of all possible MBDA scores (1-100) were generated for canonical patient types differing in the number of conventional risk factors (0 to 4) and age (45, 55, 65, 75, 85 years). To generate these curves, the contributions of TNF-RI, MMP-3 and leptin to the CVD risk score were treated in aggregate (denoted the molecular score) and estimated using a linear regression model of the difference in molecular scores vs. the difference in MBDA scores. This model for the molecular score was fit in the training dataset, then in the full dataset, with dataset (training or validation) and the interaction between dataset and change in MBDA score included as additional predictor variables. The method was considered validated if the F-test for the interaction variable was not significant at the 0.05 level.Results:The model for estimating the molecular score from the MBDA scores was validated and shown to fit the data well (Figure 1). The estimated molecular score was applied to the CVD risk score algorithm to generate curves that show how CVD risk score varies with MBDA score for several distinct patient types. These curves demonstrate that the predicted 3-year CVD risk increases continuously and markedly with increasing level of inflammation, as represented by the MBDA score (Figure 2). Age and the number of conventional risk factors also affected the predicted CVD risk, with older patients (Figure 2a) and those with more conventional risk factors (Figure 2b) being at higher risk for a CVD event.Conclusion:The level of CVD risk predicted by a new prognostic test for RA patients depends not only on conventional risk factors, which are relatively time invariant, but also varies greatly due to inflammation, which can potentially be reduced with RA treatment.