Predicting the longitudinal efficacy of medication for depression using electroencephalography and machine learning.

Area of Science:

• Neuroscience
• Computational psychiatry
• Biomedical engineering

Background:

• Major depressive disorder (MDD) treatment often involves trial-and-error.
• Predicting antidepressant response is crucial for effective MDD management.

Purpose of the Study:

• To assess machine learning (ML) for predicting short- and long-term antidepressant response.
• To utilize electroencephalographic (EEG) recordings and functional connectivity for prediction.

Main Methods:

• Applied ML algorithms to predict treatment outcomes at 4, 6, and 8 weeks.
• Used clinical features, EEG recordings, and functional connectivity metrics from 77 patients.
• Employed generalized estimation equation (GEE) for longitudinal data integration.

Main Results:

• ML models achieved prediction accuracies of 83.1% (Week 4), 73.3% (Week 6), and 80.0% (Week 8).
• Functional connectivity analysis and phase synchronization were key predictors.
• EEG-based predictions show potential for early identification of responders.

Conclusions:

• Non-invasive EEG and data-driven methods can optimize MDD treatment decisions.
• Functional connectivity metrics may help clinicians identify early responders, reducing trial-and-error.
• Future research should explore larger datasets and multimodal biomarkers for personalized treatment.