Researchers within Mayo Clinic’s Center for Individualized Medicine and Division of Rheumatology have developed a first-of-its-kind machine learning algorithm that can predict rheumatoid arthritis disease activity in a patient. The algorithm analyzes biochemical metabolites ― the product of the body’s metabolism ― in blood.
“Having fast, reliable and scalable measures for predicting the clinical course of disease activity is an important unmet need for patients with rheumatoid arthritis,” says Jaeyun Sung, Ph.D., a computational biologist within the Center for Individualized Medicine and co-senior author of the study. Dr. Sung develops computational analytical approaches to understand the intricate relationship between microbial organisms and human metabolic and immune health.
The study, which was published in Arthritis Research & Therapy, lays the groundwork for monitoring rheumatoid arthritis disease progression and systemic inflammation using blood samples alone. The findings provide direction for the potential future development of clinical laboratory tests and digital diagnostics to further enable precision medicine for rheumatoid arthritis patients.
Rheumatoid arthritis is a chronic, autoimmune disorder characterized by joint inflammation and pain that can eventually lead to bone and cartilage erosion, joint deformity, and loss in mobility. This complex disease affects nearly 1.3 million people in the U.S.
Dr. Sung says the study sheds light on why symptoms differ significantly among rheumatoid arthritis patients, which in turn makes it is so difficult to treat.
“We turned to the blood because it could potentially provide a treasure-trove of novel biomarkers for assessing not only disease activity, but also clinical subgroups, risk factors and predictors of treatment response that complement current standard laboratory tests,” Dr. Sung explains.
John Davis III, M.D., a clinical rheumatologist in Mayo Clinic’s Division of Rheumatology with a specialty interest in inflammatory arthritis, says a patient’s rheumatoid arthritis disease activity is not notable through symptoms alone. He says providing interpretable predictions could enhance the clinical treatment of rheumatoid arthritis. Dr. Davis is co-senior author of the study.
“Our study highlights the importance of investigating which biochemical functions are altered during the onset and progression of the disease,” Dr. Davis says. “To this end, metabolomics platforms can present unique opportunities for discovering novel biomarkers.”
Harnessing artificial intelligence to predict rheumatoid arthritis disease activity
For the study, Dr. Sung and Dr. Davis performed a metabolomic analysis with their machine learning algorithm on 128 plasma samples from patients with rheumatoid arthritis. Metabolomics is the large-scale study of small molecules — known as metabolites — within cells, biofluids and tissues. Metabolite levels in the blood are influenced by various factors, including genetics, inflammation, diet and even the gut microbiome.
The researchers identified 33 metabolites that stratify patients of two contrasting disease activity groups. Notably, a number of these identified metabolites were seen in previous studies to have pro-inflammatory or anti-inflammatory effects in rheumatoid arthritis.
“We found that metabolites in the blood were different between patients with higher and lower disease activity,” explains Dr. Sung. “So basically, this implies that depending on the disease activity of a patient’s RA — whether they’re in remission or on the other end of the spectrum, possibly suffering from much pain — they have different biochemicals floating around in their blood. Some were molecules you don’t want much of in your body, but some were those you want kept up. This observation led us to wonder whether biochemical blood profiles predict a patient’s disease activity score.”
Next the investigators found 51 metabolites that were significantly associated with the Disease Activity Score-28, or DAS28-CRP, which is the standard index that measures disease activity in patients with rheumatoid arthritis. These metabolites were identified after controlling for age, sex and medication history.
“Metabolomics can screen for nearly 1,000 metabolites, which is way too many to keep track of,” Dr. Davis says. “By narrowing down this list into a panel of 51 metabolites, we obtained reasonable prediction of DAS28-CRP on both discovery and validation cohorts. Our study is the first to leverage biochemical features from a plasma metabolomic profile to predict quantitative disease activity.”
Teaming up to enhance patient outcomes
The researchers say that the study highlights the essential partnership between computational biologists and clinicians to solve complex problems in medicine. They emphasize the importance of creating data-driven tools that serve as a reliable companion to a clinician instead of a technology that replaces their judgment.
Drs. Sung and Davis hope their findings can inspire future studies into how inflammation and pain in rheumatoid arthritis are coupled to physiological metabolism. Moreover, this work offers a promising glimpse into diagnosing rheumatoid arthritis disease activity solely through blood, with the overall aim to make accurate assessments faster, cheaper and minimally invasive.
“The blood provides a great window into understanding disease, especially in cases where biopsies of inflamed tissue are not easily accessible,” says Dr. Sung. “What if we can learn everything we need to know about the patient from a single drop of blood? Then, together with the patient, the clinician can examine the patient’s digital dashboard and spot any early warning signs or areas that can be improved through diet or medication. At Mayo, we’re on the right path toward advancing individualized medicine for patients with RA.”
Funding
The work was supported in part by Mayo Clinic’s Center for Individualized Medicine, and Mark E. and Mary A. Davis to Mayo Clinic’s Center for Individualized Medicine.
