Introduction: Chronic low back pain (cLBP) is a common musculoskeletal condition associated with reduced physical activity (PA). Accurate assessment of PA is essential for understanding the impact of cLBP on overall health and physical function. While wrist-worn accelerometers are gaining popularity due to improved patient compliance, there is no standardized method for processing wrist-worn data. Consequently, PA data from wrist-worn accelerometers cannot be directly compared with data from the traditionally used hip-worn accelerometers. For instance, wrist-worn sensors have been shown to overestimate PA compared to hip-worn sensors, however this has not been evaluated in the cLBP population. Thus, there is a need for a specific comparison of accelerometry data to determine the most accurate method for measuring PA from the wrist in the cLBP population. This study aims to (1) evaluate the accuracy of PA outcomes from wrist-worn sensors in comparison to hip-worn sensors in individuals with cLBP, and (2) develop a processing method to harmonize the physical activity outcomes from the wrist-worn and hip-worn accelerometers.
Methods: This study utilizes data from the ongoing comeBACK study (1U19AR076737-01, NIH Back Pain Consortium-BACPAC), which seeks to inform a precision medicine approach to treating cLBP. Over 400 individuals with cLBP were recruited, and this analysis includes data from 180 participants (mean age 53.0 years, 54.5% female) who wore both wrist-worn and hip-worn Actigraph (GT3X+) accelerometers concurrently for seven consecutive days. Physical activity metrics, including step count, total activity counts, sedentary time (SED), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA), were derived from both devices. To harmonize PA outcomes from the wrist- and hip-worn accelerometers, we applied a percentile matching method.
Results: In this cLBP population, a non-linear relationship was found between step counts from wrist and hip accelerometers, with the wrist device overestimating steps by an average of 3,500 steps per day. Established wrist-worn sensor cutpoints underestimate SED by an average of 25 minutes and overestimate LPA by 66 minutes. The novel cutpoints proposed in this study through percentile-matching reduced the SED underestimation to 7 minutes and reduced the LPA overestimation to 11 minutes. Harmonizing MVPA estimates between wrist-worn and hip-worn devices remains challenging due to the different movement patterns captured by wrist and hip placements.
Discussion: Our results highlight significant disparities between PA sensors simultaneously placed on the wrist and hip of people with cLBP, emphasizing the importance of cautious data interpretation across different wear locations. We quantified sensor bias and proposed calibration methods to support accurate comparisons of data from these two different wear locations. These findings are valuable for researchers and clinicians aiming to comprehensively assess PA in individuals with cLBP and for those seeking to harmonize data from diverse sensor placements.