New publication by Manuel Schabus: AI and wearables for sleep, health and wellbeing

How well do smartwatches, rings and similar devices actually measure our sleep? In a new paper, Manuel Schabus looks at what wearables combined with machine learning can do for sleep research and sleep medicine, and where they fall short.

Polysomnography (PSG) is still the gold standard. But it usually covers only one or two nights, while many sleep problems change from night to night or come and go. Validated long-term measurement at home could fill part of this gap.

The paper draws on three studies from our group. A neural network trained on 8,898 PSG nights classifies four sleep stages from heart rate data with about 84% accuracy (κ = 0.76). That is close to the agreement between human scorers, which is around 88%. In a home comparison against ambulatory PSG, six widely used consumer trackers reached between 56% and 73% and systematically underestimated time spent awake when sleep was disturbed. So values like sleep efficiency, sleep onset latency or wake after sleep onset from these devices should be read with caution.

The paper also covers the regulatory difference between certified medical devices and lifestyle wearables. It describes several use cases: objectively documented adherence in digital insomnia therapy, long-term data from special forces, and a 100-night recording in a patient with Kleine-Levin syndrome.

Schabus, M. KI und Wearables: Neue Wege für Schlaf, Gesundheit und Wellbeing. Möglichkeiten und Grenzen der kontinuierlichen Schlafmessung zu Hause als Ergänzung zur Polysomnographie. psychopraxis.neuropraxis (2026). https://doi.org/10.1007/s00739-026-01206-y

The paper is written in German. You can read the full text here.