JMIR Publications

ABSTRACT

Background:

Difficult-to-treat (D2T) rheumatoid arthritis (RA) remains a major clinical challenge, with limited tools to personalize management strategies in routine care.

Objective:

The STRATA-FIT project aims to develop and validate computational models to identify and stratify D2T RA into clinically meaningful phenotypes, predict which patients are at high risk of developing D2T RA, and evaluate a patient-tailored treatment strategy supported by an artificial intelligence (AI)-based decision aid.

Methods:

STRATA-FIT (Stratification of Rheumatoid Arthritis: compuTational models to personalize mAnagement strategies for difFIcult-to-Treat disease) is a European multicenter real-world (RW) cohort study using a federated learning (FL) approach complemented with a prospective cross-over clinical pilot study evaluating the developed personalized treatment strategies for D2T RA patients compared to usual care. Six expert rheumatology centers across Europe (Netherlands, Portugal, Germany, Sweden, Austria, Hungary) will contribute to the cohort, using data from electronic health records (EHRs) and two established RA registries. Approximately 14,000 RA patients with up to 15 years of follow-up are included in the RW cohort, while 263 patients with D2T RA will be enrolled in the prospective clinical pilot study.

Results:

For the RW cohort, the primary outcome is identification and phenotypic stratification of D2T RA according to the EULAR definition. For the clinical pilot study, the primary outcome is change in health-related quality of life (EQ-5D-5L) over 12 months; secondary outcomes include feasibility and usability of the AI-based decision aid, disease activity, physical function, radiographic progression, adherence, and work productivity. Stage I establishes a federated RW RA cohort using the privacy-preserving Personal Health Train infrastructure and three computational models. Stage II translates the identification and stratification models into a proof-of-concept decision aid evaluated in ACCESS-STRATA, comprising a 12-month observational and 12-month intervention phase. Biomarker analyses on biobanked samples will further explore mechanisms underlying D2T RA. The federated learning infrastructure has been implemented in five of the six participating centers. As of September 2026, 12 participants have been enrolled. This protocol describes the STRATA-FIT study design, computational modeling strategy, and planned analyses.

Conclusions:

STRATA-FIT will provide evidence on the use of computational models to identify D2T RA, stratify patients into clinically relevant phenotypes, and predict which patients are at increased risk of developing D2T RA. The prospective ACCESS-STRATA study will provide pilot evidence on the feasibility and potential clinical utility of translating these models into an AI-supported decision aid for personalized management. Findings will inform further model refinement and validation and the design of larger studies evaluating implementation and clinical impact in routine rheumatology care. Clinical Trial: NL-OMON58205 (ACCESS-STRATA-I)