Background/Purpose: Previous studies have estimated physical and mental states using wearable device data such as from smartwatches [1–3]. This study aimed to identify digital biomarkers for RA to support telemedicine and clinical care.

Methods: Patients with RA who consented to participate in the study at the 3 sites wore a wristwatch-type wearable device for 3 months, and their RA-related patient-reported outcomes were prospectively recorded daily on a smartphone application which the authors had developed [1]. Attending physicians examined the participants every month for 3 months and the clinical data were also collected. From the wearable device, data on number of steps, metabolic equivalents (METs), heart rate-related data, and sleep- related data were obtained. Relationship between wearable device data and clinical disease activity index (CDAI) were evaluated, as well as the relationship between inflammation blood biomarkers (C-reactive protein (CRP), erythrocyte sedimentation rate (ESR)) and CDAI as controls. Correlation was analyzed univariately. Machine learning (ML) techniques, which included Lasso regression, Ridge regression, Random Forest, and XGBoost, were employed to model disease activity indices such as CDAI and simplified disease activity index (SDAI), using daily data from the wearable device. We also implemented a ML classifier to detect elevated CRP ( >0.14 mg/dL), and validated its performance with ROC analysis.

Results: A total of 129 patients successfully completed the 3-month observation period. At the start of the study, the mean age of patients was 55±13 years with a mean disease duration of 8.5±10.3 years. The mean CDAI and CRP levels were 13.5±10.7 and 0.39±0.71 mg/dL, respectively. Statistical analysis revealed that an increased number of daily steps was associated with lower disease activity (CDAI) (r = -0.28; 95% CI [-0.46, -0.047]; P = 0.018). In contrast, elevated heart rate during sleep (r = 0.23; 95% CI [0.03, 0.41]; P = 0.025), increased sympathetic activity (r = 0.33; 95% CI [0.09, 0.53]; P = 0.009), and a trend toward higher overall heart rate (r = 0.32; 95% CI [0.12, 0.50]; P = 0.002) were correlated with higher CDAI scores. Additionally, inflammation markers including CRP and ESR showed moderate correlations with CDAI (CRP: r = 0.33, 95% CI [0.24 to 0.42] P < 0.001) (ESR: r = 0.32, 95% CI [0.23 to 0.41]; P < 0.001). ML approaches demonstrated predictive ability, achieving an AUC of 0.90 for CDAI remission and a median absolute error (MAE) of 5.04 in predicting CDAI values. Similarly, performance for SDAI was strong, with an AUC of 0.82 and MAE of 5.12.

Conclusion: Metrics from wearable devices, including step counts and heart rate, were significantly associated with RA disease activity, supporting their potential as digital biomarkers. These correlations were comparable to those of traditional markers like CRP and ESR. Moreover, ML models using wearable data accurately predicted disease activity and inflammation, demonstrating their utility in practice and telemedicine.References: [1] Izumi K et al. Ann Rheum Dis 2021;80:1099-1100. [2] Izumi K et al. Front Psychiatry 2021;12:611243. [3] Izumi K et al. Ann Rheum Dis 2024;83:1587-1588. Acknowledgement: M HK and KI are contributed equally.

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/wrist-worn-wearable-device-data-can-be-a-new-digital-biomarker-for-disease-activity-in-rheumatoid-arthritis-a-multicenter-single-arm-prospective-study/