This study uses deep neural networks to estimate age and mortality risk from polysomnograms (PSGs). The models, trained on 2500 PSGs and tested on 10,699 PSGs, estimated age with a mean absolute error of 5.8 ± 1.6 years, significantly outperforming basic sleep scoring measures. Increased age estimation error (AEE) was associated with higher all-cause mortality, with a 10-year increment in AEE linked to an estimated 8.7-year decrease in life expectancy. Greater AEE correlated with increased sleep fragmentation, suggesting its importance as a biomarker.

Andreas Brink-Kjær, Eileen B. Leary, Haoqi Sun, M. Brandon Westover, Katie L. Stone, Paul E. Peppard, Nancy E. Lane, Peggy M. Cawthon, Susan Redline, Poul Jennum, Helge B. D. Sorensen, Emmanuel Mignot