Towards Outcome-Driven Patient Subgroups: A Machine Learning Analysis Across Six Depression Treatment Studies

Area of Science:

• Neuroscience
• Computational Psychiatry
• Clinical Psychology

Background:

• Major Depressive Disorder (MDD) is complex with varied treatment responses.
• Predicting treatment outcomes in MDD remains challenging.
• Clinical interpretability of machine learning (ML) models for MDD is limited.

Purpose of the Study:

• To develop interpretable ML models for predicting treatment response in MDD.
• To derive patient profiles using clinical and demographic data.
• To enhance precision medicine approaches for MDD.

Main Methods:

• Analyzed data from 5438 participants across six depression treatment trials.
• Utilized the Differential Prototypes Neural Network (DPNN) ML model.
• Trained a model to predict remission probabilities for various treatments based on patient data.

Main Results:

• A 3-prototype ML model achieved an AUC of 0.66.
• Identified three distinct patient clusters with differential treatment responses.
• Cluster A: younger, severe symptoms, fatigue. Cluster B: older, less severe symptoms, high remission. Cluster C: severe symptoms, agitation, suicidal ideation, somatic symptoms.

Conclusions:

• Novel, interpretable patient profiles can be generated using ML.
• This approach can improve the interpretability of ML models in clinical settings.
• Enhanced ML interpretability holds potential for advancing precision medicine in MDD.