Pharmacological personalization of JAK inhibitors in rheumatoid arthritis: A multi-omics and AI-based scoping review and evidence-gap map.
TL;DR
Current multi-omics and AI/ML evidence is insufficient to guide routine selection of a specific JAK inhibitor in rheumatoid arthritis and a "data × methods × JAKi" map identifies both emerging signals and the links that must be strengthened before MO/MM and AI/ML can support clinical decision-making in RA.
Abstract
INTRODUCTION Personalization of Janus kinase inhibitor (JAKi) therapy in rheumatoid arthritis (RA) remains an unresolved clinical task. Multi-omics/multimodal (MO/MM) data and artificial intelligence/machine learning (AI/ML) may support treatment-response prediction, but their JAKi-specific application has not been systematically mapped.
Methods
We conducted a JBI scoping review with PRISMA-ScR reporting; the protocol was registered on OSF. Major bibliographic databases and additional sources were searched through 7 July 2025 without language or publication-status restrictions. We included direct JAKi studies using AI/ML or integrated MO/MM data and contextual non-JAKi RA studies combining MO/MM with AI/ML. Risk of bias and reporting completeness were assessed using PROBAST+AI and TRIPOD+AI. Findings were synthesized narratively.
Results
Eighteen publications were included, four of which were conference abstracts. The corpus comprised three direct JAKi AI/ML studies, three JAKi-associated MO/MM studies without AI/ML, eleven contextual non-JAKi AI/ML studies, and one mixed JAKi/TNFi study. AI/ML was applied in 15/18 publications. None of the direct JAKi AI/ML studies performed independent external validation. Reported AUROCs ranged from 0.656 to 1.00, with the highest estimates arising under internal or incompletely reported validation. Calibration and explainability were infrequently reported.
Conclusion
Direct evidence for AI/ML-guided JAKi personalization remains sparse and heterogeneous and does not support selection of a specific JAKi or routine individualized treatment decisions. Progress requires adequately powered JAKi-specific cohorts, class-separated analyses, harmonized measurements, transparent preprocessing, and independent external validation. The "data × methods × JAKi" map identifies both emerging signals and the links that must be strengthened before MO/MM and AI/ML can support clinical decision-making in RA. TAKE-HOME MESSAGE Current multi-omics and AI/ML evidence is insufficient to guide routine selection of a specific JAK inhibitor in rheumatoid arthritis. Independent external validation, calibration, and prospective assessment of clinical utility are required before these approaches can support treatment decisions.
