Journal Article

Evaluating the impact of machine learning models on adult major depressive disorder using conventional treatment strategies: a systematic review approach

Gulati, Anamika, Yadav, Nishant, Gulati, Varun, Yadav, Prashant

2025

Overview

Background Major Depressive Disorder (MDD) is a leading cause of global disability often treated through a trial-and-error approach, yet treatment response to antidepressants remains highly variable, with remission rates below 50% after initial treatment. Predicting treatment outcomes through machine learning (ML) models offers promise, potentially enabling more personalized and effective interventions. However, methodology heterogeneity, varied sample sizes, and lack of external validation of these models limit their clinical use. Methods A comprehensive systematic review of 30 studies employing ML models for MDD treatment response prediction was conducted. The analysis included models such as Support Vector Machines (SVM), Random Forest (RF), Ensemble Models, Deep Learning, and Graph Neural Networks. Studies were selected based on predefined inclusion and exclusion criteria. Key factors evaluated included model performance, interpretability, dataset characteristics, and external validation. Results SVM models consistently demonstrated robust predictive performance across multiple studies (AUC 0.65–0.74) using clinical and symptom data, balancing accuracy and interpretability. EEG-based ML models achieved high accuracy (up to 88%) and are emerging as scalable, cost-effective tools for outpatient monitoring. Multi-omics and neuroimaging-based models showed promise in precision psychiatry but were limited by small sample sizes and generalizability challenges. Advanced models like Deep Learning and Graph Neural Networks provided valuable research insights but remain distant from clinical application. Conclusions ML models hold significant potential in enhancing the precision of antidepressant treatment selection in MDD. SVM and EEG-based ML models currently represent the most clinically viable approaches, while multi-omics, neuroimaging, and advanced deep learning models remain research-intensive. Future efforts should prioritise large-scale validation, model interpretability, and realistic implementation strategies to bridge the gap between research and clinical practice.

• Publisher: BioMed Central,Nature Publishing Group,Springer
• Subject: Accuracy / Algorithms / Antidepressant response prediction / Antidepressants / Clinical outcomes / Comorbidity / Datasets / Deep learning / EEG / Epidemiology / Graph neural networks / Health Sciences / Heterogeneity / Learning algorithms / Machine learning / Major depressive disorder (MDD) / Medical imaging / Medicine / Medicine & Public Health / Mental depression / Mental disorders / Neural networks / Neuroimaging / Performance evaluation / Precision psychiatry / Psychiatry / Psychotherapy / Public Health / Remission / Research Methodology / Review / Statistics for Life Sciences / Support vector machine (SVM) / Support vector machines / Systematic review