Clinical Scorecard: Creation and external validation of a diagnostic framework to distinguish major depressive disorder from bipolar disorder

At a Glance

Key Highlights

• BD affects ~2% globally, often misdiagnosed as MDD due to symptom overlap, especially during depressive episodes.
• Machine learning models using EMR data can aid in distinguishing BD from MDD, but require rigorous external validation.
• This study developed and externally validated ML models with SHAP interpretability to improve diagnostic accuracy.

Guideline-Based Recommendations

Diagnosis

• Use ICD-10 criteria confirmed by structured and semi-structured psychiatric interviews (MINI, SCID).
• Consider machine learning models incorporating sociodemographic, clinical, and biochemical variables for differential diagnosis.
• Recognize the high rate of initial misdiagnosis of BD as MDD and monitor for diagnostic revision.

Management

• Implement targeted interventions early to prevent irreversible damage and improve outcomes in BD.
• Avoid inappropriate treatment by improving diagnostic accuracy between BD and MDD.

Monitoring & Follow-up

• Regularly reassess diagnosis during hospitalization and follow-up to detect conversion from MDD to BD.
• Monitor clinical and biochemical markers as part of ongoing evaluation.

Risks

• Misdiagnosis of BD as MDD can lead to ineffective treatment and poor clinical outcomes.
• Lack of clear biomarkers necessitates reliance on comprehensive clinical and EMR data.

Patient & Prescribing Data

Inpatients with initial diagnosis of MDD or BD based on ICD-10

Correct diagnosis is critical to guide appropriate pharmacological and psychosocial interventions; misdiagnosis delays effective treatment.

Clinical Best Practices

• Employ structured diagnostic interviews alongside clinical judgment to confirm diagnosis.
• Utilize validated machine learning models with external validation to support differential diagnosis.
• Incorporate biochemical and physiological data from routine blood tests to enhance diagnostic accuracy.
• Apply SHAP analysis to interpret ML model predictions and understand feature contributions.
• Maintain vigilance for diagnostic changes over time, especially in patients initially diagnosed with MDD.

References

• Global prevalence and features of bipolar disorder
• Challenges in differential diagnosis of BD and MDD
• Machine learning applications in psychiatry
• Use of SHAP values for model interpretability
• External validation of ML models for BD vs MDD diagnosis