Toward Individualized Prediction of Response to Methotrexate in Early Rheumatoid Arthritis: A Pharmacogenomics-Driven...

Area of Science:

• Rheumatology
• Pharmacogenomics
• Machine Learning

Background:

• Early rheumatoid arthritis (RA) treatment response varies significantly.
• Identifying predictors of methotrexate response is crucial for effective management.
• Personalized medicine approaches are needed to optimize RA therapy.

Purpose of the Study:

• To evaluate machine learning (ML) models for predicting methotrexate response in early RA.
• To investigate the utility of clinical and genomic biomarkers in treatment prediction.
• To assess the potential of ML in guiding RA treatment decisions.

Main Methods:

• Utilized demographic, clinical, and genomic data from 643 early RA patients.
• Included 160 single-nucleotide polymorphisms (SNPs) associated with RA or methotrexate metabolism.
• Applied supervised ML methods with cross-validation on training and validation cohorts.

Main Results:

• ML models combining clinical factors and SNPs predicted treatment response with 76% accuracy in the validation cohort.
• Key predictors included specific SNPs (rs12446816, rs13385025, rs113798271, rs2372536) and baseline Disease Activity Score (DAS28).
• Achieved an area under the receiver operating curve of 0.84 in the training cohort.

Conclusions:

• Pharmacogenomic biomarkers and baseline DAS28 scores effectively predict methotrexate response in early RA.
• ML-driven prediction of treatment response shows promise for optimizing RA therapy selection and escalation.
• This approach can aid clinicians in making informed treatment choices for RA patients.