Improving mid-life healthy adult TST and SWS through personalised interventions using AI and smart devices to objectively assess individual sensitivity to behavioural and environmental factors which influence sleep: Abstracts from the 17th World Sleep Congress, Volume 115, Supplement 1

Introduction: 77% of UK mid-life adults suffer from short sleep – i.e. <7 hours’ total sleep time (TST). This has a negative impact on their health and wellbeing. The only treatment for those who are not chronic insomniacs is ‘sleep hygiene’. While sleep hygiene has had a medium effect in healthy-adult sleep tests, the research used subjective measures and lacked behaviour change techniques (BCTs) like self-monitoring. Individual sensitivity varies greatly on each dimension: e.g. one person’s sensitivity to evening light can be 58x that of another (Philips et al, 2019). Slow wave sleep (SWS) may be critical for mid-life adults because it supports clearance of beta-amyloid from the brain, which reduces the risk of developing Alzheimer’s Disease. There is limited research into how to increase healthy-adult SWS.

Materials and Methods: A sleep-health-intervention system was built, using smart-device and digital-diary data and closed-loop machine learning (ML) to provide personalised intervention recommendations and tracking through a digital interface. The recommendation combines a sleep-research rules-based approach with a ML assessment of an individual’s sensitivity to 16 sleep-hygiene behavioural and environmental factors. Supporting rationale, combining the individual’s data with sleep-research insights, is explained using messaging and images accessible to the target audience. Once an individual selects up to 3 goals, the tracker provides daily feedback as to whether they are meeting these goals, using 10 high-impact digital BCTs to support change.

The system and supporting protocols were built in collaboration with 35 healthy 40-55-year-olds (no chronic insomnia, OSA, severe anxiety or depression). First, user-interface structure, imagery and messaging wireframes were co-created with 20 participants using Yardley’s Person Based Approach. Second, data-extraction interfaces and protocols were built and tested with 10 participants: a daily-digital diary, Dreem (EEG headband), Oura (tracker), LYS (light) and Sleep Angel (a sleep-environment monitor built to privately and securely measure and transmit sound, temperature, and light levels. Third, the full digital interface and then the full system with supporting protocols were built and refined with 5 participants over 3 months.

A 10-participant pilot tested the digital system and protocols. Baseline data was collected over 2 months and, if there was an opportunity to improve sleep, participants implemented their chosen interventions over 28 days.

Results: 8 participants provided sufficient quality data to proceed to baseline; 6 chose to complete an intervention.

- TST increased by 23.3 minutes (+6%) on average over baseline [95% CI: 11.0-35.7]

- SWS increased 7.5 minutes [-3.4 to +18.4], +11%

We will share our full results, the Sleep Angel device and digital-intervention interface.

Conclusions: The impact of 16 sleep-hygiene practices on TST and SWS can be quantified objectively, using smart devices and closed-loop ML. Impact of each practice is notably different for each participant.

Personalised-advice packages, incorporating BCTs and the interventions individuals are most sensitive to, can increase objectively measured TST and SWS. Compliance is markedly different for each intervention.

Acknowledgements: My participants and Bath ART-AI and Oxford Sleep Medicine collaborators for support and challenge.