WEARABLE-DERIVED METRICS WERE ASSOCIATED WITH AIS SCORES AND DEMONSTRATED UTILITY IN MACHINE LEARNING-BASED CLASSIFICATION: THESE FINDINGS SUPPORT THE POTENTIAL OF WEARABLE-DERIVED METRICS AS DIGITAL BIOMARKERS OF INSOMNIA IN PATIENTS WITH RA
Author(s)
Misako Higashida-Konishi, MD1, Keisuke Izumi, MD2, Shuntaro Saito, MD3, Hiroki Tabata, MD1, Satoshi Hama, MD1, Tatsuhiro Oshige, MD1, Yutaka Okano, MD1, Hisaji Oshima, MD1, Katsuya Suzuki, MD1, Jiro Sakamoto, MA4, Toshikazu Fukami, MA4, Kazumichi Minato, MA4, Nobuhiko Kajio, MD5, Yasushi Kondo, MD3, Hiroaki Taguchi, MD5, Yuko Kaneko, MD3.
1Division of Rheumatology, Department of Medicine, NHO Tokyo Medical Center, Tokyo, Japan, 2NHO Tokyo Medical Center, Keio University, TechDoctor, Inc., Tokyo, Japan, 3Division of Rheumatology, Department of Internal Medicine, Keio University School of Medicine, Tokyo, Japan, 4TechDoctor, Inc., Tokyo, Japan, 5Department of Rheumatology, Kawasaki Municipal Hospital, Kanagawa, Japan.
1Division of Rheumatology, Department of Medicine, NHO Tokyo Medical Center, Tokyo, Japan, 2NHO Tokyo Medical Center, Keio University, TechDoctor, Inc., Tokyo, Japan, 3Division of Rheumatology, Department of Internal Medicine, Keio University School of Medicine, Tokyo, Japan, 4TechDoctor, Inc., Tokyo, Japan, 5Department of Rheumatology, Kawasaki Municipal Hospital, Kanagawa, Japan.
OBJECTIVES: Insomnia, which is more prevalent in patients with rheumatoid arthritis (RA) than in the general population, significantly impairs quality of life. The Athens Insomnia Scale (AIS) is a validated patient-reported outcome measure for assessing insomnia severity. Wearable devices can objectively monitor physiological and behavioral parameters. This study evaluated associations between wearable-derived metrics and AIS scores in patients with RA.
METHODS: A prospective observational study enrolled adult patients with RA who wore a study-provided wrist-worn device (Google Fitbit). Associations between AIS scores and wearable-derived metrics were examined using Pearson and partial correlation analyses adjusted for potential confounding variables. Machine learning (ML) models based on wearable-derived metrics were developed to classify higher and lower AIS scores. Model performance was evaluated using five-fold cross-validation, with AUC as the primary performance metric. SHapley Additive exPlanations (SHAP) analysis was used to identify influential features.
RESULTS: A total of 129 patients were enrolled, of whom 107 completed the questionnaires and were included in the final analysis. The median AIS score was 5 (interquartile range, 3-8). AIS scores were significantly correlated with wearable-derived parameters, including daytime mean heart rate (r=0.30, p<0.05), sleep duration excluding wake periods (r=−0.46, p<0.05), and a sleep heart rate variability (HRV) parameter (r= 0.49, p<0.05). ML models based on wearable-derived metrics demonstrated moderate-to-good discriminative performance for identifying insomnia severity, with AUC values of 0.66 for the cutoff-based model (<6 vs. ≥6 , based on the validated AIS threshold for insomnia) and 0.79 for the quartile-based model (lowest vs. highest quartile). SHAP analysis identified daytime HRV parameters as the most influential features.
CONCLUSIONS: Wearable-derived metrics were associated with AIS scores and demonstrated utility in ML-based classification. These findings support the potential of wearable-derived metrics as digital biomarkers of insomnia in patients with RA. (MHK and KI contributed equally.)
METHODS: A prospective observational study enrolled adult patients with RA who wore a study-provided wrist-worn device (Google Fitbit). Associations between AIS scores and wearable-derived metrics were examined using Pearson and partial correlation analyses adjusted for potential confounding variables. Machine learning (ML) models based on wearable-derived metrics were developed to classify higher and lower AIS scores. Model performance was evaluated using five-fold cross-validation, with AUC as the primary performance metric. SHapley Additive exPlanations (SHAP) analysis was used to identify influential features.
RESULTS: A total of 129 patients were enrolled, of whom 107 completed the questionnaires and were included in the final analysis. The median AIS score was 5 (interquartile range, 3-8). AIS scores were significantly correlated with wearable-derived parameters, including daytime mean heart rate (r=0.30, p<0.05), sleep duration excluding wake periods (r=−0.46, p<0.05), and a sleep heart rate variability (HRV) parameter (r= 0.49, p<0.05). ML models based on wearable-derived metrics demonstrated moderate-to-good discriminative performance for identifying insomnia severity, with AUC values of 0.66 for the cutoff-based model (<6 vs. ≥6 , based on the validated AIS threshold for insomnia) and 0.79 for the quartile-based model (lowest vs. highest quartile). SHAP analysis identified daytime HRV parameters as the most influential features.
CONCLUSIONS: Wearable-derived metrics were associated with AIS scores and demonstrated utility in ML-based classification. These findings support the potential of wearable-derived metrics as digital biomarkers of insomnia in patients with RA. (MHK and KI contributed equally.)
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MT44
Topic
Medical Technologies, Patient-Centered Research, Real World Data & Information Systems
Topic Subcategory
Digital Health
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)