Abstract
Objective
Rheumatoid arthritis is a heterogeneous disease. Data‐driven approaches, from synovial histology to joint involvement patterns (JIPs), have sought to define clinically meaningful subgroups. Whether these subgroups represent stable phenotypes or transient disease states remains unclear. We aimed to develop a novel analytical method to measure longitudinal stability of these subsets.
Methods
We analyzed two randomized controlled trials, NORD‐STAR (NOrdic Rheumatic Diseases Strategy Trials And Registries; n=630), BeSt (Behandel‐Strategieën, n=445) and one real‐world cohort (Reumazorg Zuidwest Nederland (RZWN), n=252), all treated with DMARDs under a treat‐to‐target strategy. Patients were assigned to previously defined JIP‐subgroups based on shared clinical features. Disease progression was then tracked over one year along the two principal axes of JIP variation: overall severity (oligoarticular vs. polyarticular) and anatomical distribution (hand vs. foot). Longitudinal stability was assessed by measuring how phenotypically similar each patient remained to others sharing their initial joint pattern, quanitified using silhouette‐like scores.
Results
Effective treatment led to convergence toward mild disease. Nevertheless, patients sharing similar baseline JIPs maintained clustering together over time significantly more than expected by chance in BeSt (
P
=0.018–0.038), NORD‐STAR (
P
=0.004–0.017), and RZWN (
P
Conclusion
Patients with similar baseline joint involvement patterns maintain phenotypic similarity over time, suggesting JIP subgroups reflect distinct, stable phenotypes rather than transient disease states. Our tool is available to study the longitudinal stability of any subset.