A Novel Wearables-Based Sleep Quality Index to Quantify Postoperative Sleep Quality.

OBJECTIVE: To develop a novel wearables-based sleep index that can quantify sleep disruption after lung resection and to apply machine learning analysis to these indices to identify patients exhibiting distinct postoperative sleep patterns. METHODS: This prospective observational cohort study analyzed patients undergoing lung resection who wore wearable devices preoperatively and at least 90-days postoperatively. To capture patient-level deviations in sleep-wake states postoperatively, we developed the Circadian Sleep Recovery Index (CSRI). We used K-means clustering to identify patient groups exhibiting distinct sleep patterns based on PCA-reduced CSRIs features from postoperative days 2-7. Associations between these groups and long-term sleep quality and patient-reported quality of life (QOL) were evaluated. RESULTS: Among 113 patients, the median CSRI dropped sharply on postoperative day 0 and remained lower during the first two postoperative weeks than observed preoperatively, indicating significant sleep disruptions. Unsupervised analysis of the CSRIs revealed two groups: one that experienced minimal sleep disruptions ("Fast Sleep Recovery Group") and one that experienced significant sleep disruptions ("Slow Sleep Recovery Group"). Compared to the Fast Sleep Recovery Group, the Slow Sleep Recovery Group experienced greater reductions in QOL at 30-days postoperatively and greater sleep disruptions during the 3-months following surgery. CONCLUSIONS: Machine learning analysis of a novel wearable-based sleep index identified patients experiencing significant early postoperative sleep disruptions; these sleep disruptions were associated with long-term sleep disruptions and greater reductions in QOL.