Learning optimal biomarker-guided treatment policy for chronic disorders

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Informatics

Background:

• Antidepressant treatment response rates for major depressive disorder are often low.
• Electroencephalogram (EEG) signals, particularly in alpha and theta bands, show potential as biomarkers for predicting antidepressant response.
• Personalized medicine approaches are needed to optimize treatment selection and improve outcomes.

Purpose of the Study:

• To develop an integrated pipeline for personalized antidepressant treatment selection in major depressive disorder.
• To utilize resting-state pre-treatment EEG recordings and other modifiers to guide treatment policy.
• To improve the overall response rate to antidepressant therapies.

Main Methods:

• An automatic, site-specific EEG preprocessing pipeline was designed to extract robust features.
• Causal forests were used to estimate conditional average treatment effects (CATE).
• A doubly robust technique enhanced the efficiency of average treatment effect estimation.
• An efficient policy learning algorithm developed an optimal depth-2 treatment assignment decision tree.

Main Results:

• Evidence of significant heterogeneity in treatment effects was found, modulated by EEG features.
• A significant average treatment effect was identified, surpassing conventional methods.
• The developed treatment assignment decision tree demonstrated competitive performance against Q-Learning and outcome-weighted learning in simulations and clinical trial application.

Conclusions:

• The proposed pipeline effectively uses pre-treatment EEG features to personalize antidepressant treatment for major depressive disorder.
• This approach offers a data-driven strategy to improve treatment efficacy and patient outcomes.
• The findings support the integration of neuroimaging biomarkers and advanced machine learning for optimizing psychiatric care.