The Insomnia EEG Score: a new tool for the classification of people with poor sleep

Quantitative features could help objectively identify and grade insomnia severity, though there is currently no pathophysiological biomarker of insomnia. In this study, we used the largest cohort of individuals with and without insomnia to date to train a model capable of distinguishing people with insomnia from those without insomnia. We identified 720 spectral, 606 spindle, and 16 macro-architecture features from EEG channels of the PSG and examined their ability to classify people with insomnia and insomnia subtypes (subjective, maintenance, onset, and combined maintenance and onset), compared to individuals without insomnia. Consistent with prior work, these features poorly classified individuals when assessed independently. However, a linear combination of these features (the "Insomnia EEG Score," IES) was able to distinguish individuals with (n = 2123) and without (n = 930) insomnia, with spectral features providing the greatest discrimination (outperforming models based on hypnogram macroquantities or spindle features). These results demonstrate that insomnia is likely a quantifiable sleep disorder, albeit one requiring multiple objective measures (such as IES) to gauge disease subtype and, potentially, response to therapeutic interventions.