Predicting Response to the Antidepressant Bupropion using Pretreatment fMRI

Area of Science:

• Neuroimaging
• Computational psychiatry
• Machine learning in medicine

Background:

• Major depressive disorder (MDD) is a leading cause of adult disability globally, with current antidepressant treatments often requiring extensive trial-and-error.
• Patient-specific responses to antidepressants like bupropion are unpredictable, leading to prolonged suffering and treatment delays for many.
• Existing treatment selection protocols can take over a year, negatively impacting employment, relationships, and mental health, including suicidal ideation.

Purpose of the Study:

• To develop and validate a predictive model for identifying individual patients likely to respond favorably to bupropion, a common antidepressant.
• To utilize pretreatment functional magnetic resonance imaging (fMRI) data for personalized treatment prediction, specifically for bupropion.
• To be the first model to predict individual patient response to bupropion using neuroimaging data.

Main Methods:

• A deep learning predictor was trained using task-based fMRI data from a randomized controlled trial to estimate changes in the Hamilton Rating Scale for Depression (HAMD) score.
• An extensive neural architecture search was performed across 800 model and brain parcellation combinations to optimize prediction accuracy.
• The model's performance was evaluated based on its ability to predict treatment response and remission rates.

Main Results:

• The developed deep learning model successfully identified patients who would achieve remission with bupropion, demonstrating a low Number Needed to Treat (NNT) of 3.2.
• The model achieved a significant neuroimaging study effect size, explaining 26% of the variance in treatment response (R² = 0.26).
• The predictor accurately estimated the post-treatment change in the 52-point HAMD score with a Root Mean Square Error (RMSE) of 4.71.

Conclusions:

• Pretreatment fMRI data, analyzed with deep learning, can effectively predict individual patient response to bupropion for major depressive disorder.
• This predictive model has the potential to significantly expedite the antidepressant selection process, improving clinical outcomes and reducing patient burden.
• The findings support further research and development of fMRI and deep learning-based predictive tools for various depression treatments.