JMIR Publications

ABSTRACT

Background:

Treat-to-target strategies have substantially improved outcomes in rheumatoid arthritis (RA), yet induction strategies remain guided by population-level evidence that does not adequately account for inter-individual heterogeneity in disease biology, treatment response and patient context. Artificial intelligence (AI) offers an opportunity to support individualized prediction in RA.

Objective:

To review the current evidence on AI applications for treatment response and remission prediction in RA and to propose a conceptual research-oriented framework for evaluating AI-enabled prediction of clinical deep remission (CliDR) as a stringent modeling endpoint to support individualized induction strategies.

Methods:

A narrative review of literature on AI applications in RA was conducted. Evidence on remission prediction, treatment response trajectories, multimodal data integration was analyzed qualitatively to identify methodological trends, performance limitations endpoint heterogeneity and translational gaps.

Results:

CliDR, defined by absence of swollen and tender joints and normalization of inflammatory markers, is proposed as a candidate prediction endpoint for AI model for RA induction therapy due to its biologically coherent and objective definition. Existing AI models for RA treatment outcome prediction demonstrate moderate accuracy and limited external validation, reflecting challenges related to heterogenous endpoints, small sample sizes and observational data structure. An AI-enabled research framework for RA induction therapy is outlined, spanning baseline phenotyping, treatment prioritization, early treatment response trajectory modeling and longitudinal monitoring with emphasis on evaluation, interpretability and clinician oversight.

Conclusions:

AI offers a potential pathway towards more individualized RA induction strategies. Rigorous validations are required before AI-enabled prediction tools can inform routine RA care.