Clinical–psychosocial archetypes predict short-term outcomes in inflammatory arthritis: an unsupervised segmentation study

BACKGROUND: Heterogeneity in inflammatory arthritis (IA) outcomes limits the effectiveness of non-individualized treatment approaches, and digital health platforms can capture psychosocial and behavioral signals that may stratify responses beyond diagnosis or baseline severity. OBJECTIVE: To identify clinically interpretable patient clusters and evaluate their associations with 12-week outcomes in IA, testing whether cluster membership adds information beyond demographics, diagnosis, and baseline symptom burden. METHODS: We retrospectively analyzed the use of a CE-certified rheumatology application among adult patients. Baseline clinical and psychosocial variables (Patient's Global Assessment of Disease Activity (PGADA) and Patient's Global Assessment of Pain Intensity, sleep quality, social support, distress, fatigue, activity, diet/fasting) were winsorized (1st/99th), z-scaled, and imputed by median/mode for features only; outcomes were complete-case. Unsupervised k-means (k = 5) was selected based on silhouette, gap, and consensus diagnostics. The primary validation outcome was remission (12-week PGADA ≤ 20 mm for patients with baseline PGADA ≥ 40 mm), with distributional changes in PGADA and percentage change as secondary endpoints. RESULTS: Among 2,924 patients, five clusters were identified (size range 17.7-22.9% of the cohort). The 12-week remission rate was 7.0%, with the "resilience" profile (characterized by better sleep, stronger social support, and lower distress) showing the highest probability of remission and the most favourable PGADA distribution. In contrast, distress-dominant clusters (characterized by poor sleep and weak support) showed the lowest remission rates and minimal improvement. The median ΔPGADA% was 8.3% (IQR - 8.2% to 32.0%). In adjusted analyses, the cluster signal persisted beyond baseline severity; percentage-change estimates were attenuated for clusters with lower baseline PGADA. CONCLUSION: Cluster-level phenotypes derived from routinely collected app data align with short-term clinical outcomes, highlighting sleep, social support, and distress as modifiable factors that may influence short-term outcomes. Programs should emphasize the quality of activity and recovery (not just volume), particularly for patients with high distress and poor sleep. Future work should evaluate cluster-informed, multicomponent interventions in prospective studies. Key Points • Clinical-psychosocial archetypes derived from routinely collected app data (symptoms, sleep, social support, distress, lifestyle) were strongly associated with 12-week remission and PGADA change, beyond diagnosis and baseline severity. • Distress-dominant archetypes with poor sleep and weak social support had the lowest remission rates and minimal improvement, indicating that unaddressed psychological burden and sleep problems can blunt the benefits of otherwise appropriate pharmacological care. • Resilient archetypes, with better sleep, stronger social support, lower distress, and healthier lifestyle patterns, showed the most favourable outcomes, supporting a stratified care model in which digital tools help identify high-risk patients and prioritise targeted behavioral, psychosocial, and recovery-focused interventions rather than simply prescribing more physical activity.