Diagnostic Framework to Differentiate Major Depressive Disorder from Bipolar Disorder

Overview

This study developed and externally validated machine learning models using electronic medical record data to distinguish major depressive disorder (MDD) from bipolar disorder (BD). The models demonstrated good performance and generalizability across two independent medical centers, with interpretability enhanced by SHAP analysis.

Background

Bipolar disorder affects approximately 2% of the global population and is characterized by mood fluctuations between manic and depressive episodes. Differentiating BD from MDD is challenging due to overlapping depressive symptoms, leading to frequent misdiagnosis and inappropriate treatment. There is a lack of clear biomarkers and universal diagnostic tools for BD, highlighting the need for reliable screening methods. Machine learning offers potential for improving diagnostic accuracy but requires rigorous external validation to ensure clinical applicability.

Data Highlights

Key Findings

• Machine learning models were constructed using 34 EMR variables including sociodemographic, clinical, and biochemical data.
• Feature selection was performed using LASSO regression and Boruta algorithm to identify relevant predictors.
• Six different ML models were developed and internally validated on a large single-center dataset.
• External validation on an independent dataset demonstrated good generalization ability of the models.
• SHAP analysis provided transparent explanations of feature contributions to model predictions.
• The study addresses limitations of previous research by including rigorous external validation and interpretability.

Clinical Implications

The validated machine learning framework can assist clinicians in differentiating MDD from BD using routinely collected EMR data, potentially reducing misdiagnosis and improving treatment decisions. The interpretability of the model via SHAP values supports clinical trust and understanding of key diagnostic features. This approach may facilitate earlier and more accurate identification of BD in patients presenting with depressive episodes.

Conclusion

This study presents a rigorously validated, interpretable machine learning diagnostic tool that effectively distinguishes bipolar disorder from major depressive disorder using electronic medical record data. The framework holds promise for enhancing clinical diagnostic accuracy and guiding appropriate treatment strategies.

References

1. Zhu et al. 2023 -- Diagnostic model to distinguish BD from MDD using XGBoost
2. Huang et al. 2023 -- Logistic Regression model for BD vs MDD diagnosis
3. International Statistical Classification of Mental Disorders (ICD-10)
4. SHAP methodology references (Lundberg & Lee, 2017)