Can Machine Learning help us in dealing with treatment resistant depression? A review

Area of Science:

• Neuroscience and Psychiatry
• Computational Biology and Machine Learning

Background:

• Approximately one-third of patients treated with antidepressants do not achieve sufficient symptom relief.
• Up to 15% of patients remain symptomatic after multiple treatment trials, a condition known as treatment-resistant depression (TRD).
• The precise definition and underlying mechanisms of TRD remain unclear, hindering effective management of major depressive disorder.

Purpose of the Study:

• To review and synthesize existing Machine Learning (ML) classification studies related to treatment-responsive depression and TRD.
• To evaluate the potential of ML in predicting treatment response in patients with established TRD.
• To assess the feasibility of applying ML models in clinical practice for diagnosing and managing TRD.

Main Methods:

• Bibliographic search conducted on PubMed, Google Scholar, and Medline.
• Inclusion of studies focusing on clinical, imaging, genetic, and EEG ML classification for depression and TRD.
• Analysis of eleven selected studies, with seven focusing on TRD predictors and four on predicting treatment response in TRD.

Main Results:

• ML models demonstrated good accuracy in classifying between major depressive disorder (MDD) responders and TRD using clinical variables.
• Electroencephalogram (EEG) measures show potential for predicting treatment response to various therapies in established TRD.
• Significant variability exists in TRD definitions, variable selection, ML algorithms, and pipelines across studies.

Conclusions:

• ML represents a viable approach to enhance the understanding and classification of TRD.
• Standardization of ML methodologies is crucial for clinical implementation in assessing major depressive disorders.
• Further research is needed to refine ML models for accurate TRD stratification and personalized treatment strategies.