AI-Enhanced Tools Help Personalize Rheumatoid Arthritis Treatment
Machine learning-enhanced algorithms may help clinicians select the best medications for each patient, speeding up the long journey to find effective relief for a crippling disorder.
By John Halamka, M.D., president, Mayo Clinic Platform, and Paul Cerrato, senior research analyst and communications specialist, Mayo Clinic Platform
One of the major problems that rheumatoid arthritis (RA) patients and their physicians must contend with before starting a drug regimen is deciding which of the many available agents to choose from. To date, it’s been a trial-and-error process, which means patients often endure months of continued pain and possible adverse reactions only to find they have been put on an ineffective medication. Part of the solution to this dilemma can be found in the field of pharmacogenomics.
Pharmacogenomics explores the ways in which a person’s genetic makeup influences their response to specific medications. A mutation in the liver cytochrome P-450 enzyme system, for instance, can alter the way a person metabolizes a variety of drugs, which in turn can cause the drug to be toxic, or ineffective, depending on the specific genetic variant. In a study involving RA patients, Mayo Clinic investigators looked at 160 single nucleotide polymorphisms (SNPs) that have been linked to the disease or to the metabolism of the drug methotrexate. They then combined that data with patients’ clinical characteristics, age, sex, smoking history and rheumatoid factor--a biomarker for the disease--to develop a machine learning based algorithm. The purpose of the algorithm was to predict which patients were most likely to improve on the drug.
Elena Myasoedova, M.D., Ph.D, the lead investigator and a rheumatologist at Mayo Clinic, found that the algorithm’s ability to predict a response to methotrexate enabled them to identify which patients were most likely to benefit from this medication during the first three months of treatment. Their research, published in Arthritis Care & Research, evolved from a long-term marriage between AI and pharmacogenomics. “This approach began by developing tools to predict drug treatment outcomes in major depressive disorder, but we are delighted to see that it can potentially be applied widely, in this case to the drug therapy of rheumatoid arthritis,” says pharmacogenomics leaders at Mayo Clinic Liewei Wang, M.D., PH.D. and Richard Weinshilboum, M.D. The data used in the published study was the result of a collaboration between Mayo Clinic and the PhArmacogenetics of Methotrexate in Rheumatoid Arthritis (PAMERA) Consortium.
Elena Myasoedova and her colleagues employed random forest modeling to develop their predictive algorithm, trained it on 336 UK patients, including rheumatoid factor positivity, smoking status, a variety of sociodemographic variables, and the relevant pharmacogenomics data. They used the disease activity score with 28 joint counts (DAS28) to measure the clinical impact of their model. To ensure the algorithm’s accuracy, they trained it with 5 repeats and 10-fold cross validation using the training cohort. They found “the area under the receiver operating curve of 0.84 (p=0.05) in the training cohort and achieved a prediction accuracy of 76% (p=0.05) in the validation cohort (sensitivity 72%, specificity 77%).”
While more research is needed to understand how these findings can be used in practice, the results hold promise.