Abstract
Background
Sleep disturbance is common among people living with dementia as well as their caregivers. Non‐contact video technology can be used to characterise such disturbances as well as quantifying sleep quality by measuring the number of sleep body positions (poses). Such an approach may be beneficial for home‐based longitudinal clinical monitoring of sleep pattern changes and disturbances at all stages of dementia. Here we present our pilot results of a personalised data‐driven method applied to video data for quantification of sleep disturbance comparing older and younger participants.
Method
Data were collected in two separate studies which included an overnight 10‐12 hour laboratory sleep recording from thirteen older (65‐80 years, 9 male:4 female) and eleven younger (18‐34 years, 7 male:4 female) participants in a dedicated sleep facility. A data‐driven analysis using Principal Component Analysis and k‐means clustering was applied to infrared video data extracted from a clinical polysomnography (PSG) system. The data‐driven analysis automatically determined statistically significant groupings or clusters of unique body poses for each individual. Pose number, number of pose transitions, pose duration, and pose transition duration were computed for each participant.
Result
The number of data‐driven poses in older and younger participants was remarkably similar with 15.2±3.9 and 15.8±1.8 (mean±SD) poses per participant, respectively. However, the older group had a higher number of pose transitions (33.0±8.82) compared to the younger group (23.9±6.83) (p = 0.03). A significant 20% difference (p = 0.03) in the average duration of each body position was observed, with 62.3±7.1 minutes and 78.9±15.9 minutes for the older and younger groups, respectively (see Figure 1 and Figure 2). Pose transition duration was 19.2±7.25 seconds and 14.6±4.32 seconds for the older and younger groups, respectively where they were not significantly different.
Conclusion
Although the number of body positions did not vary significantly between the two cohorts, the older group changed body position more frequently and it took them longer to do so. Data‐driven automated analysis of video‐based sleep monitoring holds significant promise for quantifying age‐related and inter‐individual differences in sleep behavior.