The research team, led by Joseph T. LeCheminant of Brigham Young University’s Department of Mechanical Engineering together with collaborators at BYU and the University of Rochester Medical Center, recruited 32 individuals with clinically confirmed insomnia and 32 good-sleeper controls, all between the ages of 18 and 49 and evenly split by gender. Insomnia diagnoses were verified using the Sleep Structured Clinical Interview for DSM-5 Sleep Disorders, and severity was gauged with the Insomnia Severity Index, with scores of 10 or above classifying a participant as having insomnia. Rather than bringing volunteers into a sterile sleep laboratory, the researchers conducted ambulatory polysomnography in participants’ own homes. Research assistants arrived roughly an hour and a half before habitual bedtime to set up electrodes, and participants then slept in their own beds, following their usual routines and waking spontaneously, an approach designed to preserve the natural sleep environment that laboratory studies often disrupt.

The technical core of the study lies in how the team quantified brain activity. EEG signals were recorded at 256 Hz from six electrode positions on the international 10-20 system: frontal sites F3 and F4, central sites C3 and C4, and occipital sites O1 and O2, referenced to the mastoid channels. Each 30-second epoch was manually scored according to the American Academy of Sleep Medicine manual, and the team computed relative spectral power in the Delta band, spanning 0 to 3.99 Hz, and the Beta band, above 13 Hz, using Welch’s periodogram with a Hamming window to minimize spectral leakage. Delta power reflects large, synchronized post-synaptic potentials characteristic of deep sleep, while Beta power reflects the fast, desynchronized activity of wakefulness and active cognition. In parallel, the researchers calculated sample entropy, a measure of a signal’s irregularity, and multiscale sample entropy, which extends the calculation across 30 different time scales to capture genuine complexity rather than mere randomness. The distinction matters: pure white noise yields high sample entropy but low multiscale entropy, while complex physiological signals maintain their irregularity across scales.

When the team correlated these metrics with physiologically scored sleep stages, the results were textbook-clean. Delta power and multiscale entropy rose from wakefulness through deep sleep, while Beta power and sample entropy fell, with occipital channels showing the strongest correlations. Sample entropy proved especially effective at tracking the sleep continuum, dropping steadily as neuronal firing became more synchronized from wake through stages N1, N2, and N3, then rebounding during REM sleep. Multiscale entropy followed a more idiosyncratic trajectory, peaking during light sleep and REM sleep, the stages associated with dynamic neural processing, dreaming, and the transition into sleep. Theta and alpha bands, by contrast, showed weaker and less consistent correlations and were excluded from further analysis.

The study’s most conceptually ambitious element was its treatment of sleep discrepancy, the gap between what the polysomnograph says and what the sleeper perceives. Using morning diary entries, the team assigned each sleep-onset epoch to one of four conditions: agreed wake, where both PSG and self-report indicated wakefulness; agreed sleep, where both indicated sleep; positive discrepancy, where participants reported sleep while PSG recorded wake; and negative discrepancy, where participants reported wake while PSG recorded sleep. The discrepancy window began at the earlier of the two onset times and extended 20 minutes past the later one. The findings here were striking. EEG metrics barely differed between conditions sharing the same physiological state, but they diverged clearly between conditions sharing the same self-reported state. In other words, the brain waves tracked the body’s objective sleep state far more faithfully than the sleeper’s conscious perception of it.

To test whether these measures could separate insomnia sufferers from healthy sleepers, the team turned to linear mixed-effects modeling, treating insomnia status, sleep stage, and their interaction as fixed effects while accounting for repeated measures within subjects. Because brain activity varies spatially, separate models were fit for each electrode rather than pooling across the scalp. The models generated 120 pairwise comparisons across the four metrics, five sleep stages, and six electrodes, of which 22, or 18 percent, reached statistical significance, including 11 at the stricter threshold of p less than or equal to 0.01. The team also considered practical significance, noting that every difference significant at the 0.01 level came with non-overlapping 95 percent confidence intervals, suggesting the effects were meaningful rather than mere statistical artifacts of the large dataset.

Three findings stood out. First, individuals with insomnia showed elevated Delta power across all six electrodes during wakefulness, which the authors interpret as increased sleep drive, possibly reflecting localized sleep deprivation even while awake. Second, sample entropy was significantly higher in the insomnia group during REM sleep, particularly in frontal, central, and occipital regions, hinting at heightened cortical arousal or conscious awareness during the dreaming stage. Third, and most intriguingly, the most consistent difference during discrepancy conditions was reduced Beta power during positive discrepancy, the paradoxical state in which people with insomnia perceived themselves as asleep while their brains were objectively awake. Rather than the heightened high-frequency arousal that the classic hyperarousal model of insomnia would predict, this pattern suggests diminished arousal and increased sleep drive precisely when subjective and objective sleep states collide.

The study also uncovered a curious regional compensation during wakefulness: sample entropy was significantly lower in the right frontal electrode F4 but higher in the right occipital electrode O2 among insomnia sufferers, a pattern the authors suggest may reflect compensatory activity, where reduced irregularity in one brain region corresponds with increased irregularity in another. Meanwhile, multiscale entropy was significantly lower in some frontal and central regions during wake and light sleep in the insomnia group, and drifted lower in five of six channels during REM sleep. These spatially specific effects underscore why the authors fit separate models per electrode: a single global model would have washed out the very regional signatures that distinguish the disorder.

Yet the study’s overarching conclusion is a caution against overselling EEG. The metrics correlated strongly with physiological sleep stages but could not reliably distinguish self-reported sleep perception within the same polysomnographic stage, and they did not consistently separate insomnia sufferers from good sleepers across all conditions. The authors frame their results through a heuristic model of sleep-wake states built on three dimensions: sleep drive, indexed by Delta power; arousal, indexed by Beta power; and conscious awareness, indexed by entropy. Insomnia, in this framework, represents an imbalanced combination, such as high conscious awareness coexisting with high sleep drive. The data partially support this model but reveal that standard EEG measures capture these dimensions only under specific conditions.

The authors acknowledge several limitations. EEG signal processing beyond AASM stage-scoring and filtering is not fully standardized, requiring substantial in-house programming for automated artifact rejection using a rolling interquartile range method. The distribution of insomnia severity scores clustered near the classification threshold, and self-reported measures such as bedtime and sleep latency may lack accuracy. The discrepancy analysis was limited to sleep onset because self-reported awakenings after sleep onset were not timestamped, potentially overlooking misperceptions later in the night. Still, the study’s contribution is clear: by pairing spectral power with both single-scale and multiscale entropy across sleep stages and discrepancy conditions, it demonstrates that no single EEG metric provides a consistent biomarker of insomnia, but that meaningful signatures of sleep drive, arousal, and conscious awareness emerge under specific conditions. Future work integrating physiological measures with self-report data, the authors conclude, may finally yield an enhanced characterization of the neurophysiological changes that make insomnia such a pervasive and misunderstood disorder.

Subject of Research: Neuroscience

Article Title: EEG correlates of sleep stages, sleep discrepancy, and insomnia: a comparative analysis of people with insomnia and healthy sleepers using spectral power and sample entropy

Article References: LeCheminant, J. T., Kay, D. B., Feland, B., Yang, C., & Blotter, J. (2026). EEG correlates of sleep stages, sleep discrepancy, and insomnia: a comparative analysis of people with insomnia and healthy sleepers using spectral power and sample entropy. Frontiers in Sleep, 5, Article 1882890. https://doi.org/10.3389/frsle.2026.1882890

Image Credits: AI Generated

DOI: 10.3389/frsle.2026.1882890

Keywords: insomnia, EEG, sleep stages, sample entropy, sleep discrepancy, polysomnography, correlates, sleep, stages, discrepancy, comparative, analysis