Associations Between Smartphone-Derived Behavioral Data and Rheumatoid Arthritis Flares

Rheumatoid arthritis (RA) features sporadic symptoms that intensify during flares, significantly affecting the quality of life. This study aimed to (1) characterize flare frequency and severity, (2) assess if short-term changes in patient-reported outcomes (PROs) signal RA flares, and (3) examine the relationship between passive smartphone data and self-reported flares.

Participants from FORWARD Databank completed PROs in two phases: conditional (flare questions triggered by PRO changes) and fixed (biweekly flare assessments). Passive smartphone data, including mobility and communication patterns, were collected alongside PROs and flares (binary outcome). Adjusting for demographic and seasonal confounders, we assessed associations between smartphone data, PROs, and flares using logistic generalized estimating equation models, multivariate analyses with backward selection, and kappa statistics.

The study included 292 adults with RA. In the conditional phase, 71% reported greater than or equal to one flare over 441 days (2.9 per participant), while 76% reported flares in the fixed phase over 172 days (3.7 per participant). Flares were linked to worse PROs. Increased mobility and longer texts were associated with fewer flares, whereas slower reaction times and shorter texts were associated with more flares. Flares were less common in the summer. Lower mobility radius (odds ratio [OR] 0.88), younger age, workdays, and lower educational level were associated with flare in the conditional phase. Worse patient global (OR 1.25) and pain (OR 1.30) were associated with flaring in the fixed phase.

Integrating PROs with passive smartphone data demonstrates novel associations with flare occurrence and highlights the potential for future predictive modeling. These findings suggest that, with further validation, personalized algorithms may one day support the earlier recognition of flares and improved disease management.