Sleep duration misses how rested you feel: a multi-scale spiking model runs five sleep disorders through one 72-hour photoperiod, and paradoxical insomnia sleeps normally yet recovers worst

The model links three layers of sleep regulation that are usually studied apart into one continuous multi-day simulation, so different sleep disorders can be compared under identical environmental conditions. Earlier sleep models existed in isolation: phenomenological equations for macroscopic circadian and homeostatic sleep propensity, lumped-rate models of hypothalamic switching, or spiking networks covering only seconds to minutes of local cortical oscillation. Five cohorts run in parallel over the same 72-hour photoperiod: healthy control, delayed-phase circadian rhythm sleep-wake disorder, chronic insomnia, paradoxical insomnia, and narcolepsy type 1.

A restoration index separates perceived recovery from total sleep time, so paradoxical insomnia shows the same sleep duration as healthy controls alongside the lowest restorative score. Clinical polysomnography has long scored sleep by macroscale measures such as total sleep time and sleep efficiency, which often fail to match patients' reported daytime impairment. The paradoxical insomnia cohort matches healthy controls on total sleep time, sleep onset latency, wake after sleep onset and cycle count, yet its restoration index falls from the optimal band to the non-restorative band, the lowest in the study.

The model attributes restoration to whether slow-wave sleep is preserved against cortical high-frequency intrusion rather than to sleep quantity. Cortical inhibition is scaled down in paradoxical insomnia, so fast recurrent synaptic currents that should be suppressed during slow-wave sleep are not. That cohort sustains waking-level high-frequency beta activity throughout stage N3 slow-wave sleep, and the model's internal subjective wakefulness meter stays elevated during objectively scored slow-wave sleep.

The framework targets mechanistic sleep research and virtual neuropharmacology, simulating five cohorts in parallel under one 72-hour photoperiod and producing synthetic local field potentials with derived sleep staging and a restoration index. The cortical network is currently a lumped single-channel C3-A2 representation; the authors propose adding multi-area spatial connectivity across frontal, central and occipital regions to capture traveling slow waves and topographic spindle gradients, and refining REM micro-architecture with extended brainstem networks.

The spectral convergence is measured against the PhysioNet Sleep-EDF database, with validation drawn from a single subject's C3-A2 channel. Drug effects are currently represented as static shifts in model variables, and the authors propose adding compartmental pharmacokinetic models to simulate absorption, distribution, clearance and half-life. Patient-specific digital twins are described as future work, requiring individual physiological parameters to be fitted to consumer wearables, actigraphy and longitudinal home EEG headbands.