A machine learning approach to predicting clinical trajectories in bipolar disorder

Area of Science:

• Psychiatry
• Computational Psychiatry
• Machine Learning in Healthcare

Background:

• Bipolar disorder (BD) presents with diverse clinical features, impacting diagnosis and treatment.
• Precision psychiatry aims to stratify patients for tailored interventions and improved outcomes.
• Understanding BD heterogeneity is crucial for early risk identification and management.

Purpose of the Study:

• To develop machine learning models for predicting bipolar disorder onset timing and illness severity.
• To identify distinct patient subtypes within the bipolar disorder cohort.
• To investigate factors contributing to diagnostic delays in bipolar disorder.

Main Methods:

• Retrospective analysis of 203 clinical datasets from bipolar disorder patients.
• Development of machine learning models including Random Forest, Weighted Logistic Regression, and XGBoost.
• Application of survival analysis and clustering techniques to identify patient phenotypes and diagnostic patterns.

Main Results:

• Weighted Logistic Regression (AUC=0.733) and XGBoost (AUC=0.729) demonstrated high accuracy in predicting onset timing.
• Cluster analysis revealed three distinct bipolar disorder phenotypes: low complexity, late-onset familial-severe, and early-onset comorbid.
• Diagnostic delay was not significantly influenced by clinical risk profiles, though severe cases showed longer latency.

Conclusions:

• Machine learning effectively predicts bipolar disorder clinical trajectories using routine data from an African cohort.
• Identified patient subtypes offer potential for personalized treatment approaches.
• Significant diagnostic delays in bipolar disorder remain unexplained by current clinical variables, necessitating further research.