Reproducibility of electroencephalography biomarkers for diagnosis of major depressive disorder

Major depressive disorder (MDD) and other psychiatric diseases can greatly benefit from objective decision support in diagnosis and therapy. Machine learning approaches based on biomarkers extracted from electroencephalography (EEG) have the potential to serve as low-cost decision support systems. Although this approach has shown promise, inconsistent findings regarding the diagnostic value of those biomarkers impede their clinical translation. Therefore, the replicability and robustness of these biomarkers need to be established first. We employ a multiverse analysis to systematically investigate the robustness against variations in six data processing steps, which may be one source of contradictory findings. These steps are artifact removal, normalization, time-series segment length, biomarker from the alpha band, aggregation, and classification algorithm. To ensure replicability of our findings, we analyze two publicly available EEG datasets with eyes-closed resting-state data containing 25/18 MDD patients and 23/14 healthy control subjects. The accuracies of diagnostic classifiers range from 81% to chance level, dependent on dataset and combination of processing steps. We observe that the selection and combination of processing steps significantly influences the results. Overall, the replicability of our findings across the two datasets was inconsistent. This study is a showcase for the advantages of employing a multiverse approach in EEG data analysis and advocates for larger, well-curated datasets to further neuroscience research that can be translated to clinical practice.