Automatic sleep/wake identification from wrist activity.
Level 3 - non-randomized controlled study
Diagnostic algorithm development and prospective split-sample validation against a gold-standard reference (polysomnography)
PubMed 1455130 · doi:10.1093/sleep/15.5.461
What was done
Forty-one participants (18 healthy controls and 23 individuals with sleep or psychiatric disorders) wore a wrist actigraph during overnight polysomnography. The dataset was split into an optimization group of 20 randomly selected subjects to train candidate sleep/wake scoring algorithms against polysomnography across various epoch lengths, and a validation group of the remaining 21 subjects to prospectively evaluate the best-performing algorithm.
What was found
The validated algorithm correctly classified sleep versus wakefulness with approximately 88% accuracy. Actigraph-derived sleep percentage correlated with polysomnography-derived sleep percentage at r = 0.82 (p < 0.0001), and sleep latency estimates correlated at r = 0.90 (p < 0.0001).
Why it matters
This study established early algorithmic foundations for automated actigraphy scoring, demonstrating that wrist-worn accelerometry can reliably estimate total sleep percentage and sleep latency against lab polysomnography.
Limits
The total sample size was small (n = 41 overall, n = 21 in the validation set). Recordings were restricted to a single overnight laboratory session alongside polysomnography, leaving multi-night outpatient stability and specificity during prolonged quiet wakefulness unmeasured in the abstract.
Cited by
- supports Wrist-based movement detection can determine whether someone is awake or asleep minute-to-minute with over 90% accuracy relative to EEG brain wave activity.