Background/Purpose: Patients with rheumatic diseases (RMD) face an increased cardiovascular (CV) mortality compared to the general population. Still, risk stratification based on traditional cardiovascular risk factors fail prediction in these patients. Particular metabolites and lipid homeostasis are of high interest as both are described to play a crucial role in cardiovascular disease (CVD) development.

In this pilot study, we aimed to identify a metabolite profile in patients with systemic lupus erythematodes (SLE), ANCA-associated vasculitis (AAV) or psoriatic arthritis (PsA) that identifies patients at cardiovascular risk, which in future can be adapted and used independently from traditional risk factors.

Methods: Plasma samples of patients with PsA (n=21), AAV (n=16) and SLE (n=16) were analyzed with a combined approach using targeted LC-MS/MS-analysis for low-concentrated lipid mediators, non-targeted LC-HRMS analysis for abundant lipids and the Biocrates MxP Quant 500 Kit a total of 685 metabolites and lipids could be detected. Compounds of interest for discrimination of patients with CV risk were selected using machine-learning with random forest algorithm, whereas 70% of the data was used for training and 30% for testing using stratified randomization. Prior training a Spearman correlation matrix and recursive feature elimination were applied to reduce the number of metabolites.

Included patients had a comparable low disease activity at inclusion with a DAS28-CRP of 2.31+0.9 (PsA), a BVAS of 2.37 +1.45 (AAV) and a SLEDAI-2K of 2.37+1.96 (SLE). 31 of the 53 included patients had a traditional CV risk factor or a CVD, with a proportion of 22.58% male in patients with CVD and 36.36% in patients without CVD.

Patients with CVD had a mean age of 60.83 years (yrs) +11.58 and a mean RMD duration of 16.83+12.82 yrs, compared to non-CVD patients with 43.81yrs+9.98 of mean age and 15.61+25.36 yrs of disease duration.

Results: Our machine-learning model can classify patients with and without CVD or CV risk with an accuracy of 86.67% and an ROC-AUC of 0.963 (77.78% sensitivity, 100% specificity). This model based on 14 metabolites including different amino acids, kynurenine, hexoses, acylcarnitine, palmitoyl ethanolamine, phospholipids and trigylcerides.

Conclusion: In this pilot study, we identified a panel of metabolites that can stratify patients with PsA, AAV and SLE based on CVD and traditional CV risk factors with a high sensitivity and specificity. The identified panel and a machine-learning model will be validated in larger cohorts to assess a risk stratification that is independent from traditional CV risk factors. Furthermore, the role of the selected metabolites in the pathophysiology of CVD might help to better understand the complex interaction of pathways leading to CVD development in patients with RMD.

« Back to ACR Convergence 2021

ACR Meeting Abstracts - https://acrabstracts.org/abstract/metabolomics-as-an-innovative-tool-for-cardiac-risk-stratification-in-patients-with-rheumatic-diseases-results-from-a-pilot-study/