Analysis of sleep for the diagnosis of sleep disorders such as Type-1 Narcolepsy (T1N) currently requires visual inspection of polysomnography records by trained scoring technicians. Here, we used neural networks in approximately 3,000 normal and abnormal sleep recordings to automate sleep stage scoring, producing a hypnodensity graph—a probability distribution conveying more information than classical hypnograms. Accuracy of sleep stage scoring was validated in 70 subjects assessed by six scorers. The best model performed better than any individual scorer (87% versus consensus). It also reliably scores sleep down to 5 s instead of 30 s scoring epochs. A T1N marker based on unusual sleep stage overlaps achieved a specificity of 96% and a sensitivity of 91%, validated in independent datasets. Addition of HLA-DQB1*06:02 typing increased specificity to 99%. Our method can reduce time spent in sleep clinics and automates T1N diagnosis. It also opens the possibility of diagnosing T1N using home sleep studies.

Study Objectives Up to 5% of adults in Western countries have undiagnosed sleep-disordered breathing (SDB). Studies have shown that electrocardiogram (ECG)-based algorithms can identify SDB and may provide alternative screening. Most studies, however, have limited generalizability as they have been conducted using the apnea-ECG database, a small sample database that lacks complex SDB cases. Methods Here, we developed a fully automatic, data-driven algorithm that classifies apnea and hypopnea events based on the ECG using almost 10 000 polysomnographic sleep recordings from two large population-based samples, the Sleep Heart Health Study (SHHS) and the Multi-Ethnic Study of Atherosclerosis (MESA), which contain subjects with a broad range of sleep and cardiovascular diseases (CVDs) to ensure heterogeneity. Results Performances on average were sensitivity(Se)=68.7%, precision (Pr)=69.1%, score (F1)=66.6% per subject, and accuracy of correctly classifying apnea–hypopnea index (AHI) severity score was Acc=84.9%. Target AHI and predicted AHI were highly correlated (R2 = 0.828) across subjects, indicating validity in predicting SDB severity. Our algorithm proved to be statistically robust between databases, between different periodic leg movement index (PLMI) severity groups, and for subjects with previous CVD incidents. Further, our algorithm achieved the state-of-the-art performance of Se=87.8%, Sp=91.1%, Acc=89.9% using independent comparisons and Se=90.7%, Sp=95.7%, Acc=93.8% using a transfer learning comparison on the apnea-ECG database. Conclusions Our robust and automatic algorithm constitutes a minimally intrusive and inexpensive screening system for the detection of SDB events using the ECG to alleviate the current problems and costs associated with diagnosing SDB cases and to provide a system capable of identifying undiagnosed SDB cases.