Data-driven cluster analysis of insomnia disorder with physiology-based qEEG variables

Abstract

We propose a novel, data-driven method to group people with insomnia disorder into different subtypes (phenotypes) using cluster analysis of physiology-based spectral quantitative electroencephalography (qEEG) variables, representing different brain activity. Motivated by previous work on insomnia disorder, we consider one group of qEEG variables extracted from non rapid-eye movement sleep stage 2 and 3. We apply the infinite feature selection and principal component analysis algorithms to select a small set of informative features from the two sleep stages, reducing the initial set of 54 variables to 2 linear combinations of 6 features. These features are then used to cluster the data with the variational Bayes Gaussian mixture model clustering algorithm. We found that people with insomnia can be successfully partitioned into three meaningful clusters: insomnia with low beta frequency of peak power, insomnia with high delta peak power and insomnia with low delta peak power. We analyse the results and discuss the distinct characteristics of each cluster. In addition, we found that the most informative features identified by our feature selection method were all from sleep stage 3. These features were the peak power in the delta band for the O1, F3 and C3 channels, and then the peak frequency in the beta band in the O1, F3 and C3 channels, which is consistent with previous insomnia studies. Our results are useful for improving the understanding of insomnia disorder, developing objective diagnostic measures and determining more personalised treatments.