Researchers have used machine learning to identify changes to the immune system associated with juvenile idiopathic arthritis (JIA). Understanding these changes may lead to better diagnosis and treatment options for this disease, a study suggests.
JIA is a heterogeneous disease, meaning that any two JIA patients are unlikely to have the same experience. JIA can be divided into subgroups, but it’s not clear whether these groups reflect actual differences in how the disease progresses, as opposed to just differences in the observable end result.
Researchers in this study set out to systematically analyze the immune system of JIA patients in hopes that a more detailed understanding of the exact changes that take place in JIA could help develop better classification systems. These, in turn, might lead to earlier and more accurate diagnoses, as well as better treatment.
The team recruited 85 patients with JIA — who were classified via four distinct systems — and 43 healthy volunteers to serve as controls. Patients with JIA were grouped according to their treatment into either untreated patients with JIA (no medication or non-steroidal anti-inflammatory drugs), steroid-treated patients with JIA (oral steroids or methotrexate or leflunomide, or a combination), or biologic-treated patients with JIA (abatacept, adalimumab, canakinumab, etanercept, tocilizumab, with or without additional steroid, leflunomide or methotrexate treatment).
They also looked at 16 patients with non-JIA inflammatory diseases, such as lupus, aiming to identify features unique to JIA.
The investigators used flow cytometry to analyze participants’ immune systems. This technique allows researchers to identify individual cells expressing particular markers — such as the markers that distinguish between different types of immune cells, like B-cells and T-cells.
Unsurprisingly, they found that JIA patients’ immune systems were behaving differently from those of controls, with JIA patients having generally more pro-inflammatory cells. There were some differences among JIA subtypes and between JIA and other inflammatory diseases — although there was also a lot of overlap.
“While all JIA subsets demonstrated overall similar changes, [systemic] sJIA in particular was a consistent outlier within the JIA diseases with the most pronounced immune deviation,” the researchers wrote. “sJIA has long been considered an entity distinct from other, more common, JIA subtypes, mediated by abnormalities in the innate immune system with features of an autoinflammatory disease.”
Interestingly, the researchers noted that JIA patients tended to have fewer of the cells responsible for producing interferon gamma, a chemical messenger that can drive inflammation. This is particularly noteworthy because other researchers have proposed models where interferon drives the progression of JIA, but the low levels of these cells suggest that it may not be so linear.
“Thus, while [interferon gamma] has been proposed as a key proinflammatory cytokine in JIA, our data in the peripheral blood are more consistent with impeded production of [interferon gamma],” the researchers wrote.
The researchers then fed their data into a variety of machine learning algorithms to see whether these algorithms could “learn” to discriminate between JIA patients and healthy patients based only on changes in the immune system.
These algorithms were 90% accurate at sorting out JIA patients from healthy individuals, and the investigators were optimistic that such studies could act as a proof-of-concept for eventually using these techniques for diagnosis and determining what treatments to give patients — although actually implementing such strategies effectively will require a lot more research and data.
“Findings could be translated into simple assays that provide added value to the clinician developing a personalised treatment plan for the patient. Such a data-driven personalised medicine approach would greatly improve the appropriate treatment selection for patients,” the researchers concluded.