Abstract
Resting-state scalp electroencephalography (EEG) is a promising method for predicting patient outcomes of antidepressant treatments. Machine-learning-based EEG analysis of averaged power features (APF) have predicted antidepressant responders in standalone samples but have not yet significantly impacted clinical care. Here, we applied new approaches for analyzing transient spectral event features (SEF) - single, short-lived increases in non-averaged power - in efforts to improve prediction of antidepressant response and yield novel mechanistic biomarkers of readiness to respond. We analyzed resting-state EEG data from the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care (EMBARC) trial, fitting linear elastic net models predicting depression score change post-treatment from pre-treatment SEF from frontal channels. We found that a model containing SEF only significantly predicted depression score changes rather than categorical response outcomes, and that its performance was similar to existing published models with APF alone. Additionally, a model including both SEF and APF outperformed both the others, emphasizing the utility of including SEF in models using EEG features for predicting antidepressant response to sertraline. We further investigated the most predictive SEF from the SEF only model, with the objective of revealing channel-level biomarkers to guide mechanistic insight into antidepressant response. We found that pre-treatment frontopolar beta duration was significantly correlated with depression score change, with greater degree of symptom response linked to shorter beta events at baseline. This finding replicates prior work on frontopolar beta duration as a possible biomarker of antidepressant response in rTMS and raises the possibility that beta events may be a cross-modal therapeutic biomarker in antidepressant treatment.