A Multidataset Investigation of Resting-State and Event-Related EEG Correlates of Major Depressive Disorder

Abstract

Introduction: Major Depressive Disorder (MDD) is a prevalent psychiatric condition which lacks reliable biomarkers for diagnosis and treatment. Electroencephalography (EEG) offers a non-invasive, cost-effective method to identify potential biomarkers of MDD. However, findings vary widely across studies due to differences in patient populations, analytic methods, and brain states. To investigate these issues, this study evaluates the consistency and reproducibility of EEG biomarkers of MDD across two datasets with both resting-state and task-based data.

Methods: We analyzed EEG data from two publicly available datasets of healthy controls and MDD patients. The MODMA dataset provided resting-state EEG data; the Cavanagh dataset included both resting-state and task-based data. Spectral power was extracted for canonical frequency bands, and task-evoked components were computed from feedback-locked event-related potentials (ERP). Group-level comparisons and brain-behaviour correlations were assessed using t-tests, Spearman correlations, and FDR correction.

Results: In the MODMA dataset, MDD patients showed increased beta power and reduced alpha power in frontal and temporal regions, with beta power having the strongest positive correlation with depression severity. In the Cavanagh dataset, spectral and ERP amplitudes did not differ significantly between MDD and control groups. Reward-Positivity (RewP) amplitudes were reduced in MDD patients, and time-frequency analyses showed altered theta and delta responses. However, none of the findings were significant after FDR correction. Cross-dataset comparisons revealed significant variability in spectral profiles, particularly in the beta band.

Discussion: Our findings highlight the challenges in identifying consistent EEG biomarkers of MDD across datasets and brain states. While trends in beta power and RewP amplitudes aligned with prior literature, they failed to replicate across datasets or remain significant after FDR correction. These results suggest that commonly reported neural markers may not generalize well across populations. This also highlights the need for greater standardization and cross-dataset validation in EEG biomarker research for MDD.

Conclusion: This study found limited consistency in EEG biomarkers for MDD across two datasets and brain states. Future research should investigate why spectral features like beta power show significance in some datasets but not others and assess how factors such as EEG hardware and recording conditions contribute to this variability.