Background/Purpose: Cardiovascular (CV) disease risk prediction models, that were originally developed for use in the general population, have been found to have suboptimal performance in patients with rheumatoid arthritis (RA) and to underestimate CV disease risk. The purpose of this study was to identify new protein biomarkers that could be added to the classic Framingham Risk Score model to improve capacity for coronary artery calcification (CAC) prediction. Our secondary aim was to quantify the improvement in the prediction of CAC with the assessment of these circulating biomarkers.

Methods: 561 patients with RA were included in this study. CAC by computed tomography was assessed using the Agatston method. 168 soluble protein-based potential candidate CV biomarkers were selected from literature screens and bioinformatics databases. 8 biomarkers were selected after algorithm training, 4 cytokines: Osteopontin, cartilage glycoprotein-39, cystatin C, chemokine (C-C motif) ligand 18; and 3 MRM (Spectroscopy Multiple Reaction Monitoring) biomarkers: SerD1_NFGYTLR PON1_IQNILTEEPK , Clusterin_IDSLLENDR. 6 models were designed: FRS (model 1); FRS + RA related factors (model 2); FRS + cytokines (model 3); FRS + RA related factors + cytokines (model 4), FRS + cytokines + MRM (model 5); FRS + RA related factors + cytokines + MRM (model 6). Increase in calibration between models was calculated through logistic regression using model 1 as the reference. Net reclassification index (NRI) and integrated discrimination improvement (IDI) and calibration of the models were calculated. Final reliability was performed using a 5-fold cross-validation of the final model.

Results: FRS score was statistically significantly associated with AUC values for both CAC >100 (0.784 [95% confidence interval -CI- 0.743-0.824]) and CAC >300 Agatston units (0.808 [95%CI 0.762-0.854]) categories. When FRS AUC was compared with the other 5 models, a statistically significant difference was found between model #3 vs. the FRS reference model (AUC 0.823 [95% CI 0.783-0.862] vs. 0.784 [95% CI 0.743-0.824], p=0.027) in the CAC >100 Agatston units subset. However, no differences were found between the other models or within the same model in the >300 Agatston units analysis. In the RA population with CAC higher than 300 Agatston units, model #3 (0.086 [95% CI 0.016-0.157], p=0.016), and model #4 (95% CI 0.093 [0.014-0.016], p=0.025), were found to have statistically significant NRIs. All models in the CAC > 100 Agatston units analysis disclosed a significantly higher integrated discrimination improvement (IDI) value when compared to the FRS reference model. Similarly, in the CAC > 300 Agatston units model, IDI was significantly higher in model #3, model #4 and model #5 when compared to the reference FRS model. Models calibrations were found to be optimal throughout the study. Internal cross-validation of model #4 through pseudo-R-squared was found to be optimal with a value of 0.237 and 0.218, respectively, for the prediction of CAC higher than 100 and higher than 300 Agatston units.

Conclusion: The addition of novel biomarkers related to RA and CV pathophysiological pathways to traditional CV risk scores might improve its prediction capacity.

« Back to 2017 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/novel-biomarkers-for-the-prediction-of-subclinical-coronary-artery-atherosclerosis-in-patients-with-rheumatoid-arthritis/