Scholarlink

Background <p>Biologic and targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs) have improved outcomes in rheumatoid arthritis (RA). However, heterogeneity in treatment response remains a significant challenge. Machine learning (ML) may enable improved prediction, but the comprehensive review of ML applications in RA is fragmented and limited. This scoping review synthesizes the literature on ML methods for predicting treatment response to b/tsDMARDs in RA.</p> Methods <p>Following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Extension for Scoping Reviews (PRISMA-ScR) guidelines, we systematically searched PubMed, MEDLINE, and Embase (from databases’ inception through March 2024). Using the Covidence online platform, two reviewers independently screened titles, abstracts, and full texts for eligibility. Studies were included if they applied ML methods in predicting treatment response to b/tsDMARD in RA. We provided a qualitative synthesis of databases used, study design, population, outcomes, predictors, and model validation. Risk of bias was assessed using Quality in Prognosis Studies (QUIPS), and reporting quality was evaluated using TRIPOD guidelines.</p> Results <p>Of 294 citations reviewed, 24 studies met the inclusion criteria. Most used real-world data from registries (<i>N</i> = 12, 50%), followed by electronic health records (<i>N</i> = 4, 17%). Study sample sizes ranged from 39 to 7,300 (Median = 494). ML models—especially boosted trees, random forests, support vector machines, and regularized regression—were most frequently applied. Study outcomes included remission, low disease activity, and treatment non-response. Common baseline predictors were disease activity, biomarkers, functional status, and patient-reported measures. AUCs ranged from 0.54 to 0.92 (Mean = 0.71), with boosted trees and neural networks often performing best. External validation was rare (<i>N</i> = 7, 17.5%), and most studies showed a low-to-moderate risk of bias (<i>N</i> = 32, 80%).</p> Conclusion <p>ML methods are increasingly used to predict RA treatment response, but vary widely in methodology and performance. Standardization, external validation, and transparent reporting are critical for advancing clinical application.</p> Clinical trial number <p>Not Applicable (NA).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning for predicting treatment response to biologic and targeted synthetic disease-modifying antirheumatic drugs in rheumatoid arthritis: a scoping review

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月11日

查看全文
  • Ehiremen Bennard Eriakha,
  • Yu Han,
  • Mai Li,
  • Jieni Li,
  • Yinan Huang

摘要

Background

Biologic and targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs) have improved outcomes in rheumatoid arthritis (RA). However, heterogeneity in treatment response remains a significant challenge. Machine learning (ML) may enable improved prediction, but the comprehensive review of ML applications in RA is fragmented and limited. This scoping review synthesizes the literature on ML methods for predicting treatment response to b/tsDMARDs in RA.

Methods

Following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Extension for Scoping Reviews (PRISMA-ScR) guidelines, we systematically searched PubMed, MEDLINE, and Embase (from databases’ inception through March 2024). Using the Covidence online platform, two reviewers independently screened titles, abstracts, and full texts for eligibility. Studies were included if they applied ML methods in predicting treatment response to b/tsDMARD in RA. We provided a qualitative synthesis of databases used, study design, population, outcomes, predictors, and model validation. Risk of bias was assessed using Quality in Prognosis Studies (QUIPS), and reporting quality was evaluated using TRIPOD guidelines.

Results

Of 294 citations reviewed, 24 studies met the inclusion criteria. Most used real-world data from registries (N = 12, 50%), followed by electronic health records (N = 4, 17%). Study sample sizes ranged from 39 to 7,300 (Median = 494). ML models—especially boosted trees, random forests, support vector machines, and regularized regression—were most frequently applied. Study outcomes included remission, low disease activity, and treatment non-response. Common baseline predictors were disease activity, biomarkers, functional status, and patient-reported measures. AUCs ranged from 0.54 to 0.92 (Mean = 0.71), with boosted trees and neural networks often performing best. External validation was rare (N = 7, 17.5%), and most studies showed a low-to-moderate risk of bias (N = 32, 80%).

Conclusion

ML methods are increasingly used to predict RA treatment response, but vary widely in methodology and performance. Standardization, external validation, and transparent reporting are critical for advancing clinical application.

Clinical trial number

Not Applicable (NA).