Leveraging clinical sleep data across multiple pediatric cohorts for insights into neurodevelopment: the...

Sleep disturbances are prominent across neurodevelopmental disorders (NDDs) and may reflect specific abnormalities in brain development and function. Overnight polysomnography (PSG) allows for detailed investigation of sleep architecture, offering a unique window into neurocircuit function. Analysis of existing pediatric PSGs from clinical studies could enhance the availability of sleep studies in pediatric patients with NDDs towards a better understanding of mechanisms underlying abnormal development in NDDs. Here, we introduce and characterize a retrospective collection of 1527 clinical pediatric overnight PSGs across five different sites. We first developed an automated stager trained on independent pediatric sleep data, which yielded better performance compared to a generic stager trained primarily on adults. Using consistent staging across cohorts, we derived a panel of electroencephalography (EEG) micro-architectural features. This unbiased approach replicated broad trajectories previously described in typically developing sleep architecture. Further, we found sleep architecture disruptions in children with Down's syndrome (DS) that were consistent across independent cohorts. Finally, we built and evaluated a model to predict age from sleep EEG metrics, which recapitulated our previous findings of younger predicted brain age in children with DS. Altogether, by creating a resource pooled from existing clinical data we expanded the available datasets and computational resources to study sleep in pediatric populations, specifically towards a better understanding of sleep in NDDs. This Retrospective Analysis of Sleep in Pediatric cohorts dataset, including staging annotation derived from our automated stager is deposited at https://sleepdata.org/datasets/rasp. Statement of Significance We introduce the Retrospective Analysis of Sleep (RASP) cohorts, a collection of 1527 clinical pediatric overnight polysomnographies that includes typically developing and neurodevelopmental disorder cases. As a first step towards addressing the analytic bottleneck inherent in manual sleep staging, we developed and validated a pediatric-specific sleep stager. Leveraging the retrospective RASP cohort's dataset, we redemonstrated known developmental trajectories in sleep architecture. To summarize changes in brain function reflected in sleep, we developed a model to predict brain age from sleep measures. We recapitulate younger predicted age in RASP Down's syndrome cases. This valuable resource underscores the utility of existing clinical polysomnography studies for studying sleep disturbances in pediatric neurodevelopmental disorders populations.