Optimizing an automated sleep detection algorithm using wrist-worn accelerometer data for individuals with chronic...

To optimize a wrist-worn accelerometer-based, automated sleep detection methodology for chronic pain populations.

A cohort of 16 patients with chronic pain underwent free-living observation for one week before participating in an Interdisciplinary Pain Management Program. Patients wore ActiGraph GT9X devices and maintained a sleep diary, documenting their nightly bedtimes and wake times. To derive sleep quality measures from accelerometry data, the Tudor-Locke sleep detection algorithm was employed. However, this algorithm had not been validated for chronic pain patients. Therefore, a sensitivity analysis of the algorithm's parameters was conducted, identifying a set of parameters which maximized the agreement between sleep periods identified by the algorithm and sleep periods identified by participant's sleep logs, which were considered ground truth. Sleep measures derived when using the optimized parameters were then compared against sleep measures derived using the default parameters.

Our optimized parameter set achieved a mean sleep detection agreement of 67% with participant's sleep logs, while the default parameter set achieved a mean agreement of 50%. Statistically significant differences were observed between sleep measures from the optimal and default parameter sets (P < .001). These findings suggest an optimized parameter set should be favored for chronic pain populations.

The Tudor-Locke algorithm provides automated sleep detection for accelerometry data; however, caution must be exercised when applying the algorithm to populations beyond its validated scope. In this manuscript, we provide an empirically optimized parameter set for applying this algorithm to adults with chronic pain.