Can we predict sleep health based on brain features? A large-scale machine learning study using the UK Biobank

Area of Science:

• Neuroscience
• Sleep Science
• Computational Psychiatry

Background:

• Correlational studies show links between sleep health (SH) and brain organization.
• Individual differences are crucial, necessitating person-specific analyses.

Purpose of the Study:

• To investigate if brain imaging features can predict individual SH traits using machine learning (ML).
• To assess the predictive power of neuroimaging markers for self-reported sleep characteristics.

Main Methods:

• Utilized UK Biobank data from 28,088 participants.
• Extracted 4677 structural and functional neuroimaging markers.
• Trained linear and nonlinear ML models to predict seven SH traits (insomnia, duration, waking ease, chronotype, napping, sleepiness, snoring).

Main Results:

• Predictive performance of brain features was consistently low (balanced accuracy 0.50-0.59).
• Highest accuracy (0.59) was for predicting ease of waking using a linear model.
• Demographic variables (age, sex) achieved comparable predictive performance.

Conclusions:

• Multi-modal brain imaging offers limited individual predictive value for SH traits.
• The relationship between self-reported sleep and brain structure/function is complex.
• Other unmeasured factors likely play a significant role in sleep-related behaviors.