JMIR Publications

ABSTRACT

Background:

Sleep problems are common among autistic individuals; however, reliable sleep assessment methods suitable for everyday life are lacking. Remote measurement technologies (RMT), including wearable sensors, passive smartphone data, and brief active self-reports, offer low-burden, scalable approaches. However, their feasibility has not been investigated in autistic adolescents and adults.

Objective:

This study assessed the feasibility of a 28-day multimodal RMT protocol for sleep measurement in autistic and nonautistic participants, examined differences in passive sleep features and active sleep quality scores between groups, and explored associations between them.

Methods:

Autistic and nonautistic participants, aged 14-35 years, completed a 28-day multimodal observation protocol involving Fitbit devices, RADAR-base passive sensing and active reporting apps. Passive sleep features were extracted using Fitbit sleep staging and steps, and smartphone accelerometer, ambient light, and app usage data. The features comprised sleep onset and offset time, sleep preparation period, wake after sleep onset, number of awakenings, latency to arising, total sleep time, sleep efficiency, and sleep-stage proportions. The feasibility assessment considered modality-specific data availability and the number of eligible analysis days (defined as those with both an identifiable primary sleep period and a sleep quality score) and examined correlations with participant characteristics. Linear mixed-effects regression models evaluated group differences and correlations between passive measures and active sleep quality scores. We used agglomerative clustering to explore whether nights could be meaningfully grouped based on passive sleep features, and whether these groups associated with active reports.

Results:

Feasibility analyses were based on 34 autistic and 39 nonautistic participants who enrolled in the study. Median Fitbit wear time, passive smartphone data availability, and active sleep rating availability were similar across groups and exceeded 75%. However, only 447 of 952 autistic participant-days (47.0%) and 645 of 1092 nonautistic participant-days (59.1%) were usable. Among autistic participants, greater tactile sensitivity was associated with lower availability of the primary sleep period and fewer eligible days. Autistic participants reported lower sleep quality than nonautistic participants (β = .447, P = .011), had shorter total sleep time (β = .408, P = .005), and had shorter sleep preparation periods (β = .444, P = .005). In both groups, higher sleep efficiency, longer total sleep duration, greater proportion of REM sleep, and later sleep offset time were associated with higher sleep quality ratings. Agglomerative clustering yielded three passive sleep feature profiles associated with significantly different active ratings.

Conclusions:

This study demonstrates favourable feasibility of multimodal RMT sleep assessment for most autistic and nonautistic participants, whilst also identifying specific challenges for some. Passive sleep features and derived profiles corresponded with active daily sleep quality ratings, supporting their utility for further refinement and adaptation in pursuit of low-burden, ecologically valid sleep assessment for autistic populations.