Predicting clozapine-induced adverse drug reaction biomarkers using machine learning

Area of Science:

• Pharmacovigilance
• Machine Learning in Healthcare
• Schizophrenia Treatment

Background:

• Clozapine is a critical antipsychotic for treatment-resistant schizophrenia.
• Severe neutropenia (agranulocytosis) is a dangerous adverse drug reaction (ADR) associated with clozapine.
• Current methods cannot reliably predict clozapine-induced agranulocytosis, limiting its use.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting agranulocytosis in patients taking clozapine.
• To improve patient safety by enabling early identification of individuals at risk for this severe ADR.

Main Methods:

• Utilized the Canada Vigilance Adverse Reaction Online Database (9395 reports).
• Addressed class imbalance using resampling techniques and appropriate metrics for rare event prediction.
• Evaluated five ML algorithms, with Gradient Boosting and Synthetic Minority Over-sampling technique (GB-SMOTE) performing best.

Main Results:

• The GB-SMOTE model achieved a recall of 0.85, AUC-PR of 0.77, and a Matthews Correlation Coefficient of 0.56.
• SHAP analysis identified blood/lymphatic disorders, leukocytosis, and neutropenia as key predictors.
• The model demonstrates significant potential for predicting clozapine-induced agranulocytosis.

Conclusions:

• Machine learning offers a promising approach for predicting clozapine-induced agranulocytosis.
• This predictive framework can enhance pharmacovigilance and patient safety in schizophrenia management.
• Early prediction of ADRs can help optimize clozapine therapy for patients with chronic schizophrenia.