Leveraging Machine Learning Approaches for Predicting Antidepressant Treatment Response Using Electroencephalography...

Area of Science:

• Neuroscience
• Psychiatry
• Computational Biology

Background:

• Major depressive disorder (MDD) treatment response varies significantly among individuals.
• Predicting antidepressant efficacy with objective biomarkers early in treatment remains a critical unmet need.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting antidepressant response in MDD patients.
• To identify key electroencephalographic (EEG) and clinical features that predict treatment outcomes.

Main Methods:

• Utilized a machine learning approach, focusing on Random Forests, to analyze EEG and clinical data from 51 MDD patients undergoing a 12-week pharmacotherapy trial.
• Collected electroencephalographic (EEG) data at baseline and 1 week post-treatment, alongside clinical assessments (MADRS scores).
• Employed feature selection and dimensionality reduction techniques (eLORETA, kernel PCA) to identify predictive biomarkers for treatment response.

Main Results:

• Machine learning models achieved high predictive utility, with 88% accuracy when all features were included.
• Identified specific EEG features (e.g., frontopolar theta, parietal alpha2) and clinical factors (e.g., week 1 concentration difficulty) as significant predictors.
• A final model with 12 key features demonstrated 78% accuracy in predicting antidepressant response.

Conclusions:

• Machine learning models integrating pre- and early-treatment EEG profiles and clinical data can effectively predict antidepressant response.
• These findings support the development of personalized, biomarker-guided treatment strategies for MDD.
• Further validation in larger, independent cohorts is necessary to establish clinical utility.