Towards causal inference-based antidepressant selection with brain and blood biomarkers

Area of Science:

• Neuroscience
• Psychiatry
• Biomarker Research

Background:

• Depression treatment selection is challenging, often relying on trial-and-error.
• Predictive biomarkers could personalize antidepressant therapy and improve remission rates.

Purpose of the Study:

• To integrate multi-modal biomarkers (brain and blood) for predicting antidepressant treatment remission.
• To utilize causal inference for selecting the most effective antidepressant treatment based on individual biosignatures.

Main Methods:

• Utilized data from the EMBARC study (N=197) including electroencephalogram, resting-state fMRI, and blood biomarkers.
• Employed multi-stage antidepressant treatment (sertraline, placebo, bupropion) with propensity score-matching for causal inference.
• Developed an unsupervised model using clinical and multi-modal features to predict remission probabilities.

Main Results:

• A combination of clinical and multi-modal features predicted Stage 2 remission with AUCs ranging from 0.71 to 0.74 across treatment arms.
• Observed outcomes across treatment arms had 82% accuracy.
• Predicted counterfactual outcomes showed promise but require further validation in prospective studies.

Conclusions:

• Multi-modal biomarkers show sensitivity in predicting antidepressant treatment response.
• Biomarker-based prediction of counterfactual outcomes supports personalized treatment selection.
• Future research should prospectively validate these findings for clinical application.