Optimizing actigraphic estimates of polysomnographic sleep features in insomnia disorder

STUDY OBJECTIVES: Actigraphy is a useful tool for estimating sleep, but less accurately distinguishes sleep and wakefulness in patients with insomnia disorder (ID) than in good sleepers. Specific algorithm parameter settings have been suggested to improve the accuracy of actigraphic estimates of sleep onset or nocturnal sleep and wakefulness in ID. However, a direct comparison of how different algorithm parameter settings affect actigraphic estimates of sleep features has been lacking. This study aimed to define the optimal algorithm parameter settings for actigraphic estimates of polysomnographic sleep features in people suffering from ID and matched good sleepers.

METHODS: We simultaneously recorded actigraphy and polysomnography without sleep diaries during 210 laboratory nights of people with ID (n = 58) and matched controls (CTRL) without sleep complaints (n = 56). We analyzed cross-validation errors using 150 algorithm parameter configurations and Bland-Altman plots of sleep features using the optimal settings.

RESULTS: Optimal sleep onset latency and total sleep time (TST) errors were lower in CTRL (8.9 ± 2.1 and 16.5 ± 2.1 min, respectively) than in ID (11.7 ± 0.8 and 29.1 ± 3.4 min). The sleep-wake algorithm, a period duration of 5 min, and a wake sensitivity threshold of 40 achieved optimal results in ID and near-optimal results in CTRL. Bland-Altman plots were nearly identical for ID and controls for all common all-night sleep features except for TST.

CONCLUSION: This systematic evaluation shows that actigraphic sleep feature estimation can be improved by using uncommon parameter settings. One specific parameter setting provides (near-)optimal estimation of sleep onset and nocturnal sleep across ID and controls.