PHENOTYPING SLEEP FROM SMART-RING TRAJECTORIES: UNSUPERVISED DISCOVERY OF FOUR LONGITUDINAL SLEEP PATTERNS IN A 456-USER INDIAN COHORT
DOI:
https://doi.org/10.65605/a-jmrhs.2026.v04.i02.pp2322-2326Keywords:
Digital Chronobiology, Smart Ring Wearable, Sleep Staging, Unsupervised Clustering, Time-Series Phenotyping, Sleep Regularity, Social Jetlag, India.Abstract
Traditional sleep statistics condense weeks' worth of data into one night’s value, flushing the organization of restorative across days. Unsupervised clustering was applied to longitudinal sleep-stage trajectories derived from a smart ring (Loop Ring, CarePlix) worn continuously by 456 adults in India over four months between Jan 2026 and Apr 2026. After cleaning, we retained 6,006 nights and limited discovery of phenotypes to the 75 subjects with ≥25 valid nights. Using K-means clustering on an 11-dimensional trajectory feature vector, four phenotypes were identified: Circadian Anchor (28%), Persistent Drifter (32%), Weekend Compensator (24%), and Restless Sleeper (16%). Only 25.2% achieved the adult recommendation of ≥7 h per night while 49.4% of nights were <6 hours. The classic textbook "rebound" sawtooth pattern failed to cluster discretely; rebound events were small and occasional (≈1% of nights). The phenotypic unit of importance for digital health is not the arithmetic mean but rather the geometry of a month of sleep.
