Periodic Reporting for period 1 - STRATA-FIT (Stratification of Rheumatoid Arthritis: CompuTational models to personalise mAnagement strategies for difFIcult-to-Treat disease)
Période du rapport: 2023-05-01 au 2024-10-31
Rheumatoid arthritis (RA) is the most common chronic inflammatory joint disease, affecting ~1% of adults (~22M in the EU). It accounts for ~10,000 disability-adjusted life years (DALYs) and societal costs of ~€55B annually. RA causes joint pain and stiffness, with potential extra-articular complications such as sicca symptoms and lung disease. Cardiovascular disease and cancer are common co-morbidities, contributing to disability, reduced quality of life, and increased mortality.
While effective treatments can reduce symptoms, joint damage, and improve outcomes, a subgroup—Difficult-to-Treat RA (D2T RA)—remains unresponsive to multiple therapies and accounts for ~20% of RA patients. This group drives a large share of RA’s socio-economic burden. Recently, the EULAR Task Force, led by Europe’s foremost rheumatology organisation, identified D2T RA as an urgent unmet need.
D2T RA is multifactorial: drug non-response, intolerance, misdiagnosis, comorbidities, non-adherence, and psychosocial factors all contribute. Predicting treatment response remains limited, and after first-line methotrexate, therapy decisions rely on trial-and-error. This may expose patients to ineffective or harmful treatments. A personalised approach, combining pharmacological and non-pharmacological strategies, is urgently needed.
To address this, two leading members of the EULAR Task Force (UMCU, MUV) have partnered with four top European rheumatology centres (KI, RS, iMM, LMU/KUM), an innovative SME (MDW), and the EULAR-PARE patient network to form the STRATA-FIT consortium. This unique initiative aims to identify and stratify D2T RA patients into clinically meaningful phenotypes based on distinct clinical profiles.
Moreover, early identification of RA patients at risk of progressing to D2T RA could enable preventive strategies and more effective treatment selection from the outset.
While effective treatments can reduce symptoms, joint damage, and improve outcomes, a subgroup—Difficult-to-Treat RA (D2T RA)—remains unresponsive to multiple therapies and accounts for ~20% of RA patients. This group drives a large share of RA’s socio-economic burden. Recently, the EULAR Task Force, led by Europe’s foremost rheumatology organisation, identified D2T RA as an urgent unmet need.
D2T RA is multifactorial: drug non-response, intolerance, misdiagnosis, comorbidities, non-adherence, and psychosocial factors all contribute. Predicting treatment response remains limited, and after first-line methotrexate, therapy decisions rely on trial-and-error. This may expose patients to ineffective or harmful treatments. A personalised approach, combining pharmacological and non-pharmacological strategies, is urgently needed.
To address this, two leading members of the EULAR Task Force (UMCU, MUV) have partnered with four top European rheumatology centres (KI, RS, iMM, LMU/KUM), an innovative SME (MDW), and the EULAR-PARE patient network to form the STRATA-FIT consortium. This unique initiative aims to identify and stratify D2T RA patients into clinically meaningful phenotypes based on distinct clinical profiles.
Moreover, early identification of RA patients at risk of progressing to D2T RA could enable preventive strategies and more effective treatment selection from the outset.
In the initial phase of the STRATA-FIT Project, a comprehensive codebook and data handling manual were developed to harmonize data across centres in alignment with FAIR principles. The partners Medical Data Works (MDW) and Karolinska Institute (KI) played a key role in ensuring seamless integration into the federated learning system, which enabled the first statistical analysis on real data and ensured the security of nodes through dedicated administrators. Under the leadership of the University Medical Centre of Utrecht (UMCU), a Statistical Analysis Plan (SAP) was designed to address the project’s core research questions. SAP aims to stratify and predict difficult-to-treat rheumatoid arthritis (D2T RA), paving the way for more effective treatment strategies. In parallel, MDW facilitated the training in federated learning, establishing a solid foundation for the key analyses planned for 2025. Medical University of Vienna (MUV) made substantial progress in advancing the Systematic Literature Review (SLR) on interventions for D2T RA subgroups. This work was supported by four dedicated PhD students, ensuring a collaborative and well-resourced effort. The partner SU laid the groundwork for future research by drafting the plan for the upcoming pilot study, and also finalized the concept for the next pilot study, reinforcing the project's forward-looking momentum. Our partners at the Ludwig Maximilians University of Munich (LMU) focused on evaluating biobanking methods. We have also developed several project-specific solutions to support adherence to the regulatory roadmap and to promote best practices in data handling and modelling. Meanwhile, our partner Gulbenkian Institute For Molecular Medicine (GiMM) supported the project in achieving significant communication milestones, including hosting a collaborative research session ahead of the EULAR 2024 Congress and strengthening the project’s presence on social media to document and disseminate its progress. Finally, the coordination led by UMCU has played a central role in project management. It has facilitated collaboration and communication across the consortium, established the structure for regular meetings, provided guidance on project rules and regulations, and supported the preparation of deliverables and the submission of the first periodic report, ensuring the smooth operation of the project as a whole.
A key innovation in STRATA-FIT is the use of federated learning (FL), a privacy-preserving method of decentralized data analysis. This allows the project to securely integrate and analyze health data from different centers without transferring sensitive patient data to a central location. Instead, FL enables the analysis of local data while keeping it secure, protecting patient privacy, and ensuring compliance with legal and ethical standards. This is facilitated through the health train approach, where a “train” (or analysis) moves between data stations, allowing for iterative machine learning without compromising data security.
In RP1, WP2, WP3 and WP7 were highly active in order to establish the federated learning infrastructure. Agreements between MDW and project partners were established, and the initial steps of the FL infrastructure have been set up. Data harmonization and necessary pre-processing of partner-specific RA data to enable concerted analyses also took place during RP1.
Relevant hardware were identified and the necessary software was installed, and their proper functionality as local nodes in the federated learning framework was ensured. Finally, partners have locally made the data available and partner MDW has executed initial federated test analyses based on each of the datasets made available. Further analysis will take place in the coming months of the project.
In RP1, WP2, WP3 and WP7 were highly active in order to establish the federated learning infrastructure. Agreements between MDW and project partners were established, and the initial steps of the FL infrastructure have been set up. Data harmonization and necessary pre-processing of partner-specific RA data to enable concerted analyses also took place during RP1.
Relevant hardware were identified and the necessary software was installed, and their proper functionality as local nodes in the federated learning framework was ensured. Finally, partners have locally made the data available and partner MDW has executed initial federated test analyses based on each of the datasets made available. Further analysis will take place in the coming months of the project.