A new scoping review suggests that machine learning models may help predict treatment response to biological and targeted synthetic disease-modifying antirheumatic drugs in rheumatoid arthritis, potentially supporting more personalized treatment strategies in clinical practice.
Biological and targeted synthetic DMARDs have changed outcomes for many rheumatoid arthritis patients, but responses remain highly variable. Identifying which patients are most likely to benefit from a particular treatment remains a major challenge. Machine learning approaches offer a potential solution by integrating complex clinical, laboratory, and patient-reported data to generate predictive models, but evidence in this area remains fragmented.
Evidence-based mapping
To address this gap, researchers conducted a scoping review according to PRISMA-ScR guidelines to analyze studies that applied machine learning techniques to predict treatment response to biological or targeted synthetic DMARDs in rheumatoid arthritis. A search of PubMed, MEDLINE, and Embase identified 24 eligible studies published up to March 2024.
Most studies relied on real-world data sources, particularly national or regional disease registries, while others used electronic health records. Sample sizes ranged from fewer than 50 patients to more than 7,000 patients, reflecting significant heterogeneity in study design and data availability.
Model performance and key predictors
A variety of machine learning techniques were employed, with boosted decision trees, random forests, support vector machines, and regularized regression models being used most frequently. Commonly evaluated outcomes include remission, low disease activity, and treatment nonresponse.
Model performance varied across studies, with reported area under the curve values ranging from 0.54 to 0.92, with an average AUC of 0.71. Boosted trees and neural networks tended to have the strongest predictive performance. Common predictors include baseline disease activity, inflammatory biomarkers, functional status, and patient-reported outcomes.
However, external validation was uncommon, with fewer than 1 in 5 studies reporting, limiting confidence in generalizability. Although most studies were assessed as having low to moderate risk of bias, the quality of reporting and consistency of methodology varied widely.
Barriers to clinical translation
The authors conclude that machine learning has the potential to predict treatment response in rheumatoid arthritis, but significant barriers remain to implementing these tools into routine clinical practice. Further standardization of outcome definitions, increased reporting transparency, and robust external validation are required to ensure reliability and clinical utility.
As interest in precision medicine continues to grow, future research integrating machine learning into prospective studies and clinical trials may help transform these predictive models into practical decision support tools for rheumatologists.
reference
Eliaha EB et al. Machine learning to predict treatment response to biological and targeted synthetic disease-modifying antirheumatic drugs in rheumatoid arthritis: A scoping review. BMC Rheumatol. 2025;DOI: 10.1186/s41927-025-00584-x.
