Rheumatic diseases are chronic, immune-mediated conditions characterized by significant heterogeneity in presentation and disease course. However, current clinical approaches often rely on snapshot-based assessments that fail to capture the complex longitudinal evolution of these conditions. To address these limitations and support the implementation of precision medicine, we present the design for the Rheumatic Digital Twin, a novel, modular conceptual framework intended to integrate heterogeneous multimodal data, ranging from electronic health records and clinical notes to imaging and omics, into a dynamic, computational representation of the patient journey. Our theoretical architecture addresses challenges related to data silos and variable availability of data modalities through a multistage approach that envisions the use of domain-specific foundation models to independently process distinct data modalities. To effectively model the temporal progression inherent in chronic diseases, the proposed design utilizes Transformer architectures, leveraging self-attention mechanisms to treat patient events, such as lab results or medication changes, as sequential data tokens. We describe how these unimodal representations would subsequently be fused via joint embedding techniques to construct a shared, multimodal representational space. Envisioned to function analogously to a recommender system, the Rheumatic Digital Twin framework is modeled to map patients into a latent space where proximity reflects clinical and biological similarity. By identifying “nearest neighbors,” historical patients with comparable trajectories, the system aims to enable in silico cohorting, theoretically allowing clinicians to forecast key clinical events, predict treatment responses, and identify likely disease courses based on the outcomes of similar peers.

Objectives: To investigate the association between joint involvement pattern (JIP) subgroups and treatment responses to conventional synthetic disease-modifying antirheumatic drugs (csDMARDs) and biological disease-modifying antirheumatic drugs (bDMARDs), and to compare the impact of JIP subgroups with other clinical parameters in treatment-naïve patients with early rheumatoid arthritis (RA). Methods: An individual patient data meta-analysis was conducted using 2 randomised controlled trials, NOrdic Rheumatic Diseases Strategy Trials And Registries (NORD-STAR) and Behandel-Strategieën (BeSt), including 1250 treatment-naïve patients with early RA. JIP subgroup assignment was based on 4 previously identified subgroups defined by baseline clinical characteristics, primarily joint involvement in the 66/68 joint scheme. Treatment outcomes were measured using the longitudinal Clinical Disease Activity Index (CDAI) and other disease activity indices through week 48. Associations of the JIP subgroups and other clinical predictors were evaluated using a mixed-model analysis. Results: Patients with a hand-dominant JIP (JIP-Hand) showed significantly better CDAI scores after treatment (Beta for CDAI = −1.4 [95% CI, −2.3 to −0.55]; p = .0016), whereas those with a polyarthritis pattern (JIP-Poly) exhibited worse outcomes (Beta = 0.95 [95% CI, 0.064-1.8]; p = .035). Female sex was also associated with worse CDAI scores (Beta = 1.2 [95% CI, 0.40-2.0]; p = .0031), whereas anticitrullinated protein antibodies did not show a significant association (Beta = 0.19 [95% CI, −0.69 to 1.1]; p = .67). When compared across groups, csDMARDs and combined bDMARDs were similarly effective in the respective JIP subgroups (interaction p > .10). Conclusions: In early RA, csDMARD and bDMARD treatments resulted in the greatest improvement in disease activity in JIP-Hand and the least improvement in JIP-Poly.

In the last decade, genome-wide association studies (GWAS) have identified tens of thousands of common variants associated with a wide array of complex traits and diseases. Integration of GWAS with molecular data has informed the development of statistical tools for causal gene discovery. In this paper, we give an overview of commonly used causal inference methods and discuss the strengths and limitations of colocalization, Mendelian randomization (MR) and network-based approaches. Colocalization is often used to assess whether the genetic association signals for two traits arise from the same causal variant, thereby strengthening inferred causal associations. MR was developed to tackle issues of confounding and reverse causality, providing a rigorous approach to causal inference and demonstrating improved false discovery rates. Unlike MR, network-based analyses employ a discovery approach and model complex relationships between multiple variables. All causal inference methods are, to varying degrees, susceptible to spurious associations due to genetic confounding, pleiotropy and linkage disequilibrium. Here, we discuss the latest developments in the field of causal gene inference and limitations of these methods. We give an overview of interplay between different approaches as well as practical applications with reference to published examples in context of heart disease.

Musculoskeletal complaints account for 30% of GP consultations, with many referred to rheumatology clinics via letters. This study developed a Machine Learning (ML) pipeline to prioritize referrals by identifying rheumatoid arthritis (RA), osteoarthritis, fibromyalgia, and patients requiring long-term care. Using 8044 referral letters from 5728 patients across 12 clinics, we trained and validated ML models in two large centers and tested their generalizability in the remaining ten. The models were robust, with RA achieving an AUC-ROC of 0.78 (CI: 0.74–0.83), osteoarthritis 0.71 (CI: 0.67–0.74), fibromyalgia 0.81 (CI: 0.77–0.85), and chronic follow-up 0.63 (CI: 0.61–0.66). The RA-classifier outperformed manual referral systems, as it prioritised RA over non-RA cases (P < 0.001), while the manual referral system could not differentiate between the two. The other classifiers showed similar prioritisation improvements, highlighting the potential to enhance care efficiency, reduce clinician workload, and facilitate earlier specialized care. Future work will focus on building clinical decision-support tools.

Objectives: Patients with rheumatoid arthritis (RA) display different trajectories towards improvement of disease. We aimed to disentangle the heterogeneity of RA disease trajectories from the first clinical visit onwards using graph-based pseudotime analysis. Methods: We studied early patients with RA over 1.5 years in 2 data sets: Leiden (Netherlands), n = 1237, with 5017 visits, and Towards a Cure for Early Rheumatoid Arthritis (TACERA) (United Kingdom), n = 243, with 750 visits. We created a pipeline for time-independent clustering of clinical and haematologic features to identify disease states. Sequence analyses of these states defined the trajectories. We studied the predictability of the trajectories with baseline features. Results: Clustering identified 8 disease states with localised inflammation (joints) and systemic inflammation (erythrocyte sedimentation rate [ESR] or leucocytes) as the main discriminating factors. The disease state sequences consisted of 4 trajectories, which we independently replicated in TACERA: A, high ESR; B, rapid progression from many inflamed joints towards remission; C, high leucocytes; and D, many inflamed joints with poor prognosis. Systemic vs local inflammation patterns showed moderate predictability at baseline (sensitivity of 71% and precision of 0.73 for trajectory A, although lower precision of 0.52 for trajectory B), while other trajectories were less predictable. Trajectories C and D had strong resemblance with B at baseline but deteriorated into less favourable trajectories. Patients in trajectory A were more often female and on average older. The trajectories were not explained by time till disease-modifying antirheumatic drug, baseline disease activity, or symptom duration. The suboptimal trajectories coincided with worse patient-reported outcomes, even when the inflammation was mainly systemic. Conclusions: We identified 4 distinct trajectories in early RA, differentiating RA into localised vs systemic inflammation. Our results highlight potential differences in disease pathology and opportunities for further targeted treatment. Inevitably, patterns without linkage to our selected features could not be detected.

Rheumatoid arthritis (RA) is a heterogeneous disease with variable symptoms, prognosis, and treatment response, necessitating refined patient classification. We applied multimodal deep learning and clustering to identify distinct RA phenotypes using baseline clinical data from 1,387 patients in the Leiden Rheumatology clinic. Four Joint Involvement Patterns (JIP) emerged: foot-predominant arthritis, seropositive oligoarticular disease, seronegative hand arthritis, and polyarthritis. Findings were validated in clinical trial data (n = 307) and an independent secondary care cohort (n = 515). Clusters showed high stability and significant differences in remission rates (P = 0.007) and methotrexate failure (P < 0.001). JIP-hand patients had superior outcomes (particularly in ACPA-positive patients) versus JIP-foot (HR:0.37, P < 0.001) and JIP-poly (HR:0.33, P = 0.005), independent of baseline disease activity and clinical markers. Synovial histology analysis (n = 194) revealed distinct inflammatory patterns across clusters, hinting at different underlying biological mechanisms. These validated RA phenotypes based on joint involvement patterns may enable targeted research into disease mechanisms and personalized treatment strategies.

Advances in artificial intelligence (AI) are transforming patient stratification in rheumatology. In 2025, three landmark studies demonstrated how multimodal AI approaches spanning clinical, molecular and longitudinal data can uncover distinct disease subtypes and predict therapeutic response, advancing the field towards precision rheumatology.