VIDEO: Machine learning models linked to accurate prediction of type 1 narcolepsy

INDIANAPOLIS — Machine learning models predict the likelihood of type 1 narcolepsy with a high degree of accuracy, opening the door to increased access to testing and care, according to a speaker at SLEEP 2023.

“Over time, biomarkers for diagnosing, detecting and characterizing different types of narcolepsy have been studied,” Chris R. Fernandez, MS, co-founder, executive chairman and chief recruitment officer at EnsoData, said in this Healio video from the conference. “Those included the speed of onset of REM sleep during naps.

Fernandez and colleagues sought to evaluate effectiveness of machine learning in detecting narcolepsy in a cohort of those with type 1 narcolepsy (n = 225) and 455 T1N-negative controls within the confines of multiple models including a supervised PSG-EEG deep learning model revolving around sleep microarchitecture and a random forest (Sleep-RF) model.

Researchers observed excellent accuracy and performance from the models, with the best performance coming from PSG-EEG model which recorded 91% sensitivity and 92% specificity and good results from the PSG-DL model of 84% and 94% in sensitivity and specificity, respectively.