Novel Multi-Modal Screening Method for Classification of Narcolepsy Type 1 is Highly Specific and Sensitive

Study results presented at the 2025 Annual Meeting of the Associated Professional Sleep Societies (APSS) demonstrated the specificity and sensitivity of a large-scale screening model for the classification of narcolepsy type 1 (NT1) which integrates polygenic risk scores (PRS), human leukocyte antigen (HLA)-typing, and nocturnal polysomnography (PSG) data.  

The study included data from individuals with diagnosed narcolepsy across multiple research centers. Researchers developed a custom U-Sleep model for sleep staging using electroencephalography, electrooculography, and electromyography data from 19,381 PSG recordings from the National Sleep Research Resource and the Stanford Sleep Clinic. The model utilized multiscale transition matrix (MTM) features to identify NT1-specific sleep transitions. A Gaussian processes classifier was trained on features from 21,846 PSGs (NT1: n=327; controls: n=21,519) and tested on 634 additional PSGs. An ensemble model combined predictions across different time resolutions. The final multi-information model integrated PRS and HLA DQB1*06:02-typing, and PSG data were applied when genetic markers were inconclusive.

Key findings included:

• Adding PRS to HLA-typing increased specificity from 81.9% to 100.0% and achieved a sensitivity of 25.9%.
• Combining PSG with HLA achieved 99.4% specificity and 94.6% sensitivity.
• Incorporating PRS rescued HLA+ cases missed by PSG, and increased sensitivity to 95.9% while maintaining 99.4% specificity.
• MTM features improved classification by approximately 1% over existing state-of-the-art screening.

This study was supported by Takeda.