Th132 - Subjective Fatigue Assessed by FACIT-F and BFI Questionnaires Correlates with Objective Wearable-Derived Data in Patients with Rheumatoid Arthritis

Keisuke Izumi – NHO Tokyo Medical Ceneter, TechDoctor Inc., Keio University School of Medicine; Shuntaro Saito – Keio University School of Medicine; Hiroki Tabata – NHO Tokyo Medical Center; Satoshi Hama – NHO Tokyo Medical Center; Tatsuhiro Oshige – NHO Tokyo Medical Center; Yutaka Okano – NHO Tokyo Medical Center; Hisaji Oshima – NHO Tokyo Medical Center; Katsuya Suzuki – NHO Tokyo Medical Center; Yasumasa Mashimo – TechDoctor Inc.; Jiro Sakamoto – TechDoctor Inc.; Toshikazu Fukami – TechDoctor Inc.; Kazumichi Minato – TechDoctor Inc.; Nobuhiko Kajio – Kawasaki Municipal Hospital; Yasushi Kondo – Keio University School of Medicine; Hiroaki Taguchi – Kawasaki Municipal Hospital; Yuko Kaneko – Keio University School of Medicine

Abstract Text: 【Background/Purpose】Fatigue profoundly impacts quality of life in rheumatoid arthritis (RA), yet assessment relies on subjective patient-reported outcome (PRO) measures (PROMs) such as functional assessment of chronic illness therapy-fatigue (FACIT-F) and brief fatigue inventory (BFI). These questionnaires are prone to recall bias and fail to capture real-time fluctuations. This study aimed to clarify the association between wearable-derived data and fatigue severity via correlation analysis and machine learning.【Methods】This prospective observational study enrolled patients with RA who continuously wore a Fitbit. FACIT-F/BFI scores were correlated with 7-day preceding wearable data ( < 80% wear time excluded), adjusting for covariates. Binary classification models were constructed using wearable data to predict severe (lowest quartile) vs. non-severe (highest quartile) PRO status, excluding intermediate quartiles. Five-cross validation and SHAP analysis were performed to identify key features and assess model robustness.【Results】Our study enrolled 107 patients with RA. FACIT-F scores indicated that No patients reported no fatigue (FACIT-F>40), and 1.9% had mild fatigue (FACIT-F 31–40). Moderate fatigue (FACIT-F 21–30) was observed in 19.6% of patients, while severe fatigue (FACIT-F≤21) was present in 80.4%. BFI scores showed that 26.4% of patients reported no fatigue (BFI=0), 54.7% had mild fatigue (BFI 1–3), 17.9% had moderate fatigue (BFI 4–6), and only 1.0% had severe fatigue (BFI 7–10). Correlations with an absolute correlation coefficient |r|≥0.4 are shown below. FACIT-F scores were correlated with resting heart rate (r=−0.40) and nocturnal HRV (r=−0.40). BFI scores were correlated with daytime HRV (r=0.42) and nocturnal HRV (r=0.43). FACIT-F binary classification model showed performance: accuracy 0.774, precision 0.703, recall 0.789, F1-score 0.742, ROC-AUC 0.88. BFI binary classification model showed performance: accuracy 0.773, precision 0.668, recall 0.828, F1-score 0.735, ROC-AUC 0.82. 【Conclusion】A correlation was observed between self-reported fatigue and heart rate variability (HRV) and activity metrics from wearable data. The binary classification model achieved excellent performance, demonstrating that wearable devices serve as superior real-time fatigue biomarkers. M HK and KI are contributed equally.