Development of a multivariate prediction model for antidepressant resistant depression using reward-related...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning in Healthcare

Background:

• Depression non-responsive to serotonergic antidepressants often involves reward processing deficits and internalizing symptoms.
• A validated self-report method is lacking to predict antidepressant resistance based on these symptom dimensions.

Purpose of the Study:

• To leverage machine learning to identify differences in self-reported anhedonia and internalizing symptoms between antidepressant non-responders and responders.
• To establish a proof-of-concept for using these indicators to predict medication responsiveness in depression.

Main Methods:

• An online case-control study utilized random forest classifiers on 24 reward predictors from 393 participants.
• Feature selection refined models to identify key predictors distinguishing medication-resistant, non-resistant, and non-depressed individuals.
• Factor analysis and cluster analysis explored retained predictors and data structures.

Main Results:

• Machine learning models achieved accuracies ranging from .42 to .70, with feature-selected models showing high sensitivity (.82) for non-responders.
• Three to five key predictors were identified as significant.
• Non-responders exhibited three distinct symptom profiles related to anxiety, anhedonia, motivation, and cognitive function.

Conclusions:

• This study demonstrates the validity of using a limited anhedonia and internalizing self-report instrument to differentiate between antidepressant-resistant and responsive depression.
• The findings suggest potential for a more targeted approach to antidepressant selection.
• Replication in a prospective cohort is recommended to confirm predictive validity.