Induced by decreasing light, people affected by seasonal mood fluctuations may suffer from low energy, have low interest in activities, experience changes in weight, insomnia, difficulties in concentration, depression, and suicidal thoughts. Few studies have been conducted in search for biological predictors of seasonal mood fluctuations in the brain, such as EEG oscillations. A sample of 64 participants was examined with questionnaires and electroencephalography in summer. In winter, a follow-up survey was recorded and participants were grouped into those with at least mild (N = 18) and at least moderate (N = 11) mood decline and those without self-reported depressive symptoms both in summer and in winter (N = 46). A support vector machine was trained to predict mood decline by either EEG biomarkers alone, questionnaire data from baseline alone, or a combination of the two. Leave-one-out-cross validation with lasso regularization was used with logistic regression to fit a model. The accuracy for classification for at least mild/moderate mood decline was 77/82% for questionnaire data, 72/82% for EEG alone, and 81/86% for EEG combined with questionnaire data. Self-report data was more conclusive than EEG biomarkers recorded in summer for prediction of worsening of depressive symptoms in winter but it is advantageous to combine EEG with psychological assessment to boost predictive performance.
Keywords: EEG biomarkers; cognitive vulnerabilities; machine learning; prediction; seasonal affective disorder winter depression; seasonal mood fluctuations.
Copyright © 2022 Höller, Urbschat, Kristófersson and Ólafsson.