Machine-Learning Algorithm Predictive of RA Disease Activity

doctor holding blood testing sample
Plasma metabolic profiling analyses was employed to determine metabolites linked to RA disease activity.

Researchers from the Mayo Clinic have developed a machine-learning algorithm based on biochemical metabolites in the blood that predicts disease activity and measures the inflammatory status of patients with rheumatoid arthritis (RA). The research was described in a paper published in Arthritis Research & Therapy.

Advancements in metabolomic research of small-molecular metabolites in RA have facilitated the discovery of biomarkers for risk factors, clinical subgroups, and treatment response predictors. Metabolomics refers to the large-scale analysis of metabolites, or small molecules, that reside within cells and tissues.

Despite advancements in metabolomics, precision medicine in RA is currently challenged by a lack of evidence-based knowledge regarding blood metabolites that reflect disease activity in the disease.

To address this largely unmet need, researchers performed ultra-high-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) on a discovery cohort comprising 128 plasma samples from 64 patients with RA. The investigators also performed UPLC-MS/MS on a validation cohort consisting of 12 samples from 12 patients with RA.

The mean ages of study participants in the discovery cohort were 62.7±10.5 years at outpatient visit 1 and 63.5±10.6 years at outpatient visit 2. The mean age of the validation cohort was 67.8±10.6 years.

Once the metabolomic profiles from the samples were assessed, the investigators then used mixed-effects regression models to identify metabolites between groups of patients with lower disease and higher disease activity scores. Disease activity was assessed with the Disease Activity Score-28 using C-reactive protein (DAS28-CRP). Lower RA disease activity was defined by a DAS28-CRP score of 3.2 or lower, while higher disease activity was defined by a DAS28-CRP score of 3.2 or higher.

The researchers found 33 metabolites that were differentially abundant between the lower (n=76) and higher (n=52) disease activity groups (P <.05). Glucuronate and hypoxanthine were 2 metabolites that were significantly increased in patients with higher disease activity, while the remaining 31 metabolites were significantly increased in lower disease activity.

In addition, the investigators identified a total of 51 metabolites within the discovery cohort that were associated with DAS28-CRP (P <.05). A generalized linear model (GLM) that was constructed based on these metabolites demonstrated higher prediction accuracy (mean absolute error [MAE]±SD, 1.51±1.77) when compared with a GLM lacking feature selection (MAE±SD, 2.02±2.21).

In the validation cohort, the researchers found a stronger association between predicted and actual DAS28-CRP in the model with feature selection (Spearman’s ρ=0.69; 95% CI, 0.18-0.90) vs the model without feature selection (Spearman’s ρ=0.18; 95% CI, −0.44-0.68).

The researchers identified 8 metabolites that were not associated with any treatment use: 6-bromotryptophan, bilirubin (E, E), biliverdin, glucuronate, N-acetyltrypto- phan, N-acetyltyrosine, serine, and trigonelline. According to the investigators, these findings “strongly suggest key metabolic pathways and modules potentially contributing to, or serving as indicators of, RA pathogenesis independent of confounding treatment effects.”

Other metabolites identified by the researchers were significantly associated with CRP patient groups: mannose, beta-hydroxyisovalerate, (14 or 15)-methylpalmitate (a17:0 or i17:0), erucate (22:1n9), 10-undecenoate (11:1n1), and N-acetylcitrulline were higher in the high-CRP group, while abundances of serine and linoleoylcarnitine (C18:3) were lower in the high-CRP group.

A limitation of this study was the small number of samples from each disease activity group, both in the discovery as well as the validation cohort.

The researchers suggest the findings from the study further support “the key role of high-throughput metabolomic technologies in identifying blood-borne biochemical signatures and metabolic pathways reflective of RA progression and systemic inflammation.”


Hur B, Gupta VK, Huang H, et al. Plasma metabolomic profiling in patients with rheumatoid arthritis identifies biochemical features predictive of quantitative disease activity. Arthritis Res Ther. 2021;23(1):164. doi:10.1186/s13075-021-02537-4