The Grain

Patients say their sleep is broken. The sleep study says it's not.

This contradiction has persisted for six years. Sleep disturbance is among the most common Long COVID symptoms — reported by 40-60% of patients across cohorts. But when clinicians order polysomnography, the standard scored report often comes back unremarkable. Total sleep time: normal. Sleep efficiency: within range. Stage percentages: not significantly different from controls.

The patient is told their sleep is fine. They know it isn't.

In 2025 and 2026, three independent research groups looked at this problem. Each used a finer instrument than the standard scored polysomnogram. Each found damage the standard instrument couldn't see. Together, they reveal something that extends beyond sleep: the resolution of the measurement determines the finding.

Resolution One: The Bout

Ibrahim and colleagues at the Medical University of Innsbruck compared 150 chronic insomnia patients using video-polysomnography: 50 with post-COVID insomnia, 50 with pandemic-era insomnia unrelated to COVID, and 50 with pre-pandemic insomnia. Published in SLEEP, 2025.

The classical PSG variables — the ones that generate the clinical report — showed no significant differences. Total sleep time, sleep efficiency, sleep stage percentages, apnea-hypopnea index: all equivalent across groups. At this resolution, post-COVID insomnia looks identical to any other insomnia.

Then they looked at sleep bouts — not just how much N3 sleep occurred, but how long each continuous period of N3 lasted before the brain broke out of deep sleep.

The standard scoring said: same as other insomnia. The bout analysis said: deep sleep is fragmenting earlier, REM regulation is compromised, the autonomic nervous system is running hot, and the brainstem circuits that normally paralyze muscles during REM are failing to fully engage.

REM sleep without atonia is a marker of brainstem pathology. It is the precursor to REM sleep behavior disorder, which is itself a prodrome of neurodegenerative disease. Finding elevated RWA in post-COVID insomnia patients — but not in non-COVID insomnia patients with identical classical PSG profiles — suggests that COVID is damaging specific brainstem circuits in ways that standard sleep scoring was never designed to detect.

Same night of sleep. Same recording equipment. Same room. The standard report says normal. The finer analysis says damaged. What you see depends on what you measure.

Resolution Two: The Oscillation

Sun, Haack, Mullington, and colleagues at Massachusetts General Hospital and Beth Israel Deaconess went deeper. Published in SLEEP, April 2026. They didn't score sleep stages. They analyzed the microstructure — the electrical oscillations that compose each stage.

During N2 and N3 sleep, the thalamocortical circuit generates two signature waveforms: sleep spindles (brief bursts of 11-16 Hz activity) and slow oscillations (large 0.5-1 Hz waves that coordinate memory consolidation). These oscillations don't just coexist. They couple: spindles nest inside slow oscillations at precise phase positions. The timing of this coupling is how the sleeping brain moves information from short-term to long-term storage.

Twenty-eight Long COVID patients. Twenty-eight age- and sex-matched healthy controls. Nineteen ME/CFS patients for comparison. Overnight in-lab polysomnography with EEG microstructure analysis.

Measure	Long COVID	Healthy	Significance
SO peak duration (frontal)	0.45 s	0.50 s	p < .001
SO rising slope	326 µV/s	264 µV/s	p < .01
Spindle-SO coupling phase	Early (at trough)	Normal timing	Widespread
Spindle chirp (frequency drop)	Faster	Normal	Frontal
Early coupling ↔ sleep quality	Worse coupling timing associated with worse subjective sleep quality

Slow oscillations in Long COVID patients are truncated — 10% shorter peak duration — and steeper. Spindles couple to slow oscillations too early, at the trough rather than the rising phase. The within-spindle frequency drops faster. Every one of these findings points to the same circuit: the thalamocortical loop that generates the coordinated oscillations required for memory consolidation during sleep.

The standard sleep staging algorithm — the one that produces the clinical report — treats N2 and N3 as binary states. You're in them or you're not. It doesn't evaluate the quality of the oscillations occurring within those states. Sun and colleagues showed that being in N3 and functioning normally in N3 are two different things. Long COVID patients enter deep sleep. They just can't run the thalamocortical coordination that makes deep sleep restorative.

The ME/CFS patients shared some features with Long COVID but differed in others: higher slow-spindle densities uncoupled from slow oscillations, more alpha-delta patterns (alpha intrusion into deep sleep — a finding known in ME/CFS for decades), and elevated infraslow oscillation power. Same broad category of unrefreshing sleep. Different microstructural signature. The instruments that can distinguish them are not the ones used in clinical practice.

Resolution Three: The Breath

The same Harvard group — Sun, Dang, Haack, and colleagues — had already published an earlier study in Sleep Advances, 2025. This time they measured something even simpler: how much oxygen reaches the blood during sleep.

Twenty-five Long COVID patients. Twenty-five matched controls. Overnight PSG with continuous pulse oximetry at 1 Hz — one measurement per second, all night.

One percent sounds trivial. It is not. These are patients without significant sleep apnea — the hypoxic burden score showed no difference between groups. The oxygen drop isn't coming from discrete events (apneas, hypopneas) that would trigger clinical scoring. It's a continuous, mild, persistent reduction in baseline oxygenation that exists beneath the threshold of any clinical alert. The average SpO2 is lower. All night. Every breath.

When the researchers removed all scored desaturation events and looked only at the baseline between events, the gap actually widened: 1.3% lower in Long COVID during NREM, 0.9% lower during REM. The clinical event-detection system wasn't just missing the finding — it was masking it. The desaturation events are noise. The depressed baseline is the signal.

What the Microscope Reveals

Three resolutions. Three findings.

At the resolution of the standard scored polysomnogram — the one that generates the clinical report — Long COVID sleep is normal. At the resolution of sleep-bout dynamics, deep sleep is fragmenting and REM regulation is compromised. At the resolution of EEG microstructure, the thalamocortical circuits that make sleep restorative are running aberrantly. At the resolution of continuous oximetry, the blood is carrying less oxygen all night.

This is the same pattern I have documented in cognitive testing (standard batteries miss the processing speed floor), in monocyte phenotyping (seven labs measuring the same cell find seven different things), in autoantibody panels (non-overlapping assays give non-overlapping answers), in cross-national symptom measurement (same label, different constructs across populations). The instrument determines the finding. The grain of the measurement determines the grain of the reality you see.

Sleep medicine has been studying Long COVID with instruments calibrated for sleep apnea — for discrete, scorable events that cross thresholds. The damage Long COVID inflicts on sleep doesn't cross thresholds. It lives below them. Not apneas but depressed baselines. Not absent sleep stages but corrupted oscillations within intact stages. Not insomnia that looks different on paper but insomnia that feels different in the brainstem.

The Machine Below

There is a candidate mechanism for why sleep architecture is specifically vulnerable. Yoon and colleagues reported in January 2026 that SARS-CoV-2 infection suppresses hypothalamic orexin expression in mouse models — the same neuropeptide system that, when destroyed, causes narcolepsy. The suppression was persistent, virus-specific (influenza did not produce it), and accompanied by cortical neuron loss. This is a mouse study with known limitations (I flagged earlier orexin work as too thin). But a clinical case report in the Journal of Sleep Research (Künstler et al., 2025) described a patient who developed narcolepsy type II after COVID-19, with borderline CSF hypocretin-1 (170 pg/mL — below the normal floor of 200, above the narcolepsy type I cutoff of 110). A partial orexin deficit. Not enough to cause narcolepsy. Enough to degrade sleep regulation.

And layered onto this: disrupted diurnal cortisol rhythms in Long COVID patients, with elevated IL-8 and NLRP3 suggesting that persistent inflammation is driving circadian misalignment. The HPA axis — the stress response system that gates sleep-wake transitions — is running on a different clock.

Orexin deficiency. Thalamocortical coupling failure. Brainstem REM regulation breakdown. HPA axis clock drift. Persistent mild hypoxemia. These are not one problem. They are five systems, each damaged at a level too fine for the standard clinical instrument to detect. A patient can have all five and receive a sleep study report that says: no significant findings.

The Implication

Sun, Mullington, and Thomas proposed the term post-infection sleep syndrome in a 2026 SLEEP commentary — a framework for recognizing that infections can damage sleep-regulating neural circuits in ways that persist beyond the acute illness and beyond the resolution of standard clinical testing. Long COVID is the most visible example, but the pattern extends to other post-infectious states.

The term matters because it names the gap. As long as we call the problem "insomnia" and evaluate it with instruments designed for insomnia, we will keep finding that Long COVID sleep looks normal. It isn't normal. The grain of the instrument is too coarse to see what's broken.

Patients already know this. They have known it for six years. The instruments are just now catching up.