Related Experiment Videos
Characterizing sleep disturbances in developmental and epileptic encephalopathies through natural language
Tommaso Lo Barco1, Maxime Douillet2, Nicole Chemaly1,3
1Reference Center for Rare Epilepsies, Department of Pediatric Neurology, Necker Enfants Malades University Hospital, Assistance publique hôpitaux de Paris (AP-HP), member of European Reference Network EpiCARE, Université Paris Cité, Paris, France.
Objective:
Sleep disorders are frequent but undercharacterized comorbidities in developmental and epileptic encephalopathies (DEEs). We aimed to estimate the frequency, spectrum, and burden of sleep disturbances across genetic DEE subtypes using real-world data extracted from electronic health records (EHRs) through natural language processing (NLP).
Methods:
We conducted a retrospective cohort study using Dr. Warehouse, an NLP-enabled clinical data warehouse, at a tertiary pediatric center. Data were extracted from January to April 2025. Patients with a group of defined DEEs were identified from narrative clinical notes and matched 1:2 with pediatric neurology patients without history of epilepsy by age, follow-up duration, and document volume. Predefined sleep-related terms were extracted from free-text reports. The proportion of patients with ≥1 documented mention during follow-up was used to estimate occurrence, alongside multivariable logistic regression analyses and a cumulative sleep burden score.
Results:
The final cohort included 884 patients with DEE and 1280 matched controls. At least one sleep-related disturbance was documented in 63.7% of DEE patients, significantly higher than in controls. The most frequent terms were "sleep disorder," "nocturnal seizures," and "night awakenings." Marked heterogeneity emerged across DEE subtypes; SYNGAP1-DEE showed the highest frequency of sleep disorders, Lennox-Gastaut syndrome was strongly associated with nocturnal seizures, and KCNT1- and PCDH19-DEE were more frequently associated with sleep apnea. DEE patients also exhibited a significantly higher cumulative burden of sleep disturbances compared with controls, with Lennox-Gastaut syndrome showing the greatest burden.
Significance:
NLP-based mining of unstructured EHR data enables large-scale characterization of sleep disorders in DEEs, revealing DEE subtype-specific sleep phenotypes. Routine and etiology-informed assessment of sleep disturbances should be integrated into the clinical management of DEEs.