Toward causal artificial intelligence for biologic treatment response in rheumatoid arthritis: current evidence and...

Rheumatoid arthritis (RA) is a heterogeneous autoimmune disease with substantial variation in response to biologic therapy. Artificial intelligence (AI) is increasingly used to predict treatment response, but most existing models are association-based and offer limited interpretability for clinical decision-making. To critically review the role of causal artificial intelligence in predicting and explaining biologic treatment response in RA and assess its potential to improve interpretability and individualized treatment selection. A narrative review with an integrative approach was conducted, as described by Gasparyan et al. Studies on causal inference, explainable AI, and biologic treatment-response modeling in RA were identified through searches of PubMed/MEDLINE, Scopus, Web of Science, and the Directory of Open Access Journals, updated to May 2026, with backward citation screening. The RA literature is dominated by predictive and explainable machine learning studies rather than integrated causal AI systems. Recent scoping reviews identified 24 RA-specific studies and 74 RA studies in a broader inflammatory arthritis literature, with AUC values 0.54 to 0.92. External validation is uncommon. A smaller body of work has begun to estimate individualized treatment effects or emulate target trials, suggesting a bridge toward causal reasoning. Causal AI is a promising translational direction for predicting and explaining biologic treatment response in RA. The evidence base remains limited; empirical validation, stronger handling of confounding, and usable transparent models are needed. Priorities include narrow drug coverage, sequential treatment decisions, multimodal data integration, fairness, and shared decision making.