Wearable Movement Data as a Potential Digital Biomarker for Chronic Pain: An Investigation Using Deep Learning

Abstract

Objective: Chronic pain is known to affect approximately a quarter of the United States population. Despite the prevalence of chronic pain, current chronic pain measures are often reliant on self reports and other similar subjective data. The purpose of this study was to examine chronic pain in a longitudinal and naturalistic setting for the potential to enhance chronic pain patient care by capturing more realistic and consistent data in everyday settings. Methods: The study sample consisted of a National Health and Nutrition Examination Survey (NHANES) sample from 2003–2004 with 4,240 participants that had an average age of 49.57(SD = 18.38) years. The NHANES sample had movement intensity data collected each minute over 7 days by wearing a hip mounted Actigraph accelerometer. This data was used to determine if chronic pain could be predicted by movement intensity. Results: A deep learning model for time series (i.e., convolutional long short-term memory network model) determined that participants’ chronic pain could be predicted with daily actigraphy movement intensity (AUC validation = 0.60, AUC test = 0.57). Additionally, participants with chronic pain had, on average, lower physical activity intensity (124.09 counts per minute, 95% CI: 112.58–135.59) than participants without chronic pain (152.29 counts per minute, 95% CI: 147.87–156.72). Conclusions: The results of this study demonstrated that passive wearable measures of physical activity could serve as a potential biomarker for chronic pain. Chronic pain could be predicted using daily activity movement intensity, which supports the value of passive data collection for assessment of chronic pain.

Keywords

• deep learning
• chronic pain
• actigraphy
• digital tools
• physical activity
• passive sensing