Abstract
Abstract
Fatigue is a burdensome symptom in inflammatory arthritis (IA), yet meaningful longitudinal change remains poorly characterized in real-world settings. To characterize fatigue trajectories, quantify clinically meaningful improvement/worsening, and identify baseline predictors of fatigue outcomes in a digital IA cohort. We analyzed Midaia digital application data from patients with rheumatoid arthritis, psoriatic arthritis, axial spondyloarthritis, or mixed IA with baseline Brief Fatigue Inventory (BFI) data. Fatigue trajectories were modelled via cluster-robust longitudinal regression, with secondary landmark analyses classifying BFI improvement/worsening using a prespecified pragmatic 1.0-point BFI change threshold at weeks 12 (W12), 24 (W24), and 52 (W52). Predictors were examined using logistic regression, machine learning with area under the curve (AUC) and model interpretation, and inverse probability weighting (IPW) for loss to follow-up. The cohort (2,660 patients; 3,436 valid BFI observations; 485 with follow-up) was 78.3% female, mean age 45.6 ± 12.9 years, baseline BFI 5.2 ± 2.0. The unadjusted BFI slope was − 0.017 points/week (95% CI − 0.025 to − 0.008;
p
< 0.001; corresponding to − 0.20 points by W12, below the MCID); the IPW-adjusted slope was attenuated to − 0.003 points/week, suggesting that the crude estimate was sensitive to differential follow-up. At W12, 26.8% improved and 25.6% worsened, with similar proportions at W24 and W52. Higher baseline BFI predicted W12 improvement (OR 1.54) and lower worsening (OR 0.62); higher baseline Patient Global Assessment of Disease Activity predicted worsening (OR 1.02). Machine-learning discrimination was modest (AUC 0.651–0.720). Diet/nutrition engagement was associated with higher improvement (OR 1.78) and lower worsening (OR 0.45). Fatigue trajectories in IA were heterogeneous. Landmark improvement/worsening proportions were unchanged after IPW, whereas mean improvement was attenuated. Baseline fatigue severity dominated prediction; action-plan associations are hypothesis-generating.