Objective features of sedentary time and light activity differentiate people with low back pain from healthy controls: a pilot study

BACKGROUND CONTEXTPhysical inactivity has been described as both a cause and a consequence of low back pain (LBP) largely based on self-reported measures of daily activity. A better understanding of the connections between routine physical activity and LBP may improve LBP interventions.PURPOSEIn this study, we aim to objectively characterize the free-living physical activity of people with low back pain in comparison to healthy controls using accelerometers, and we aim to derive a set of LBP-specific physical activity minutes thresholds that may be used as targets for future physical activity interventions.STUDY DESIGNCross-sectional.PATIENT SAMPLEA total of 22 low back pain patients and 155 controls.OUTCOME MEASURESAccelerometry derived physical activity measures.METHODSTwenty-two people with LBP were compared to 155 age and gender-matched healthy controls. All subjects wore an ActiGraph accelerometer on the right hip for 7 consecutive days. Accelerometry-based physical activity features (count-per-minute CPM) were derived using Freedson's intervals and physical performance intervals. A random forest machine learning classifier was trained to classify LBP status using a leave-one-out cross-validation procedure. An interpretation algorithm, the SHapley Additive exPlanations (SHAP) algorithm was subsequently applied to assess the feature importance and to establish LBP-specific physical activity thresholds.RESULTSThe LBP group reported mild to moderate disability (average ODI=18.5). The random forest classifier identified a set of 8 features (digital biomarkers) that achieved 88.1% accuracy for distinguishing LBP from controls. All of the top distinguishing features were related to differences in the sedentary and light activity ranges (<800 CPM), whereas moderate to vigorous physical activity was not discriminative. In addition, we identified and ranked physical activity thresholds that are associated with LBP prediction that can be used in future studies of physical activity interventions for LBP.CONCLUSIONSWe describe a set of physical activity features from accelerometry data associated with LBP. All of the discriminating features were derived from the sedentary and light activity range. We also identified specific activity intensity minutes thresholds that distinguished LBP subjects from healthy controls. Future examination on the digital markers and thresholds identified through this work can be used to improve physical activity interventions for LBP treatment and prevention by allowing the development of LBP-specific physical activity guidelines.