Machine learning predicts response to pediatric arthritis drugs

The application of machine learning by scientists at the University of Manchester could one day allow doctors to target children and young people with arthritis who are most likely to benefit from first-line treatment.

Methotrexate is the first-line drug given for juvenile idiopathic arthritis (JIA), but it is only effective or well tolerated in half of the children and adolescents who receive it.

Patients who fail to respond to drugs may have to wait longer before receiving second-line treatment and may suffer from severe joint pain and other symptoms that are often devastating to children and their families. It has a significant impact.

the study, published in diary e-biomedicineThis will facilitate more precise studies on the identification of predictors of response to methotrexate, such as biomarkers, which may lead to better prediction of expected outcome after initiation of medication.

It found that one in eight children and adolescents who started methotrexate had some symptoms, although they showed improvement in the inflammatory features of the disease.

They also showed that 16 percent of children taking methotrexate may see slower improvement in disease activity over time than other children.

Lead author Dr Stephanie Shope-Worrall said: ‘Giving methotrexate to children for whom methotrexate is not effective not only wastes health services’ time, money and effort, but also exposes children to potential side effects. This will result in unnecessary exposure to

“But now machine learning is opening the door to understand which aspects of a child’s illness are ameliorated by the drug, and which children should immediately start other treatments alongside or instead of methotrexate. It is now possible to predict what will happen.

“Furthermore, this study shows how clinical trials that only look at drug ‘response’ or ‘non-response’ for childhood-onset arthritis miss the point.

“This oversimplification results in drugs being labeled as ‘effective’ when important symptoms such as pain remain, or when there is significant improvement in one aspect of this complex disease. They may even be labeled as ‘ineffective.’

The research team accessed data from four national cohorts of children and young people who started treatment before January 2018.

Components of the Juvenile Arthritis Disease Activity Score are recorded at the start of treatment and over the next year, including the number of joint swellings, physician perception of the disease, patient and parent health reports, and inflammation blood test results. it was done.

They used machine learning to identify and predict clusters of patients with different disease patterns after methotrexate treatment. The clusters are then compared to existing treatment response measures.

Rapid improvers (11%), slow improvers (16%), improvement-relapse (7%), and persistent disease (44%) were identified from 657 children and adolescents tested in 1,241 patients. it was done.

Two other clusters, which they termed Persistent Physician Global Ratings (8%) and Persistent Parent Global Ratings (13%), were characterized by improvement in all but one activity score feature.

Dr. Shoop-Worrall further added, “The long-term effects of this delayed disease control require further investigation. Our study also highlights pediatric clustering as a basis for stratified treatment decisions. demonstrates the usefulness of machine learning techniques.”

“This study builds on existing research on methotrexate treatment response and confirms that response is not bivariate and can vary widely depending on different characteristics of the disease within an individual.

“Current trials of methotrexate in JIA classify patients into responders and non-responders.

“That misclassification can undermine research that seeks to identify predictors of response, such as biomarkers.”