Exploratory Analysis for Development Predictive Models of Immune Checkpoint Inhibitor-Induced Myocarditis Using a...

Area of Science:

• Oncology
• Immunology
• Data Science

Background:

• Immune checkpoint inhibitors (ICIs) are vital cancer treatments but can cause severe immune-related adverse events (irAEs).
• ICI-induced myocarditis (ICIM) is a rare but potentially fatal irAE with unclear pathogenesis and risk factors.
• Early prediction of ICIM is crucial for patient management and improving outcomes.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the onset of ICIM within three months of initiating ICI therapy.
• To identify key clinical variables associated with ICIM risk using predictive modeling.

Main Methods:

• Utilized a large health insurance database for an exploratory study.
• Developed predictive models using Light Gradient Boosting Machine (LightGBM) and Random Forest algorithms.
• Employed undersampling and bagging techniques to address dataset imbalance and analyzed feature importance using SHapley Additive exPlanations (SHAP).

Main Results:

• The Random Forest model outperformed the LightGBM model in predicting ICIM.
• SHAP analysis identified concurrent ICI use as the most significant predictor of ICIM.
• The developed models achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of approximately 0.63, indicating limited but feasible predictive performance.

Conclusions:

• Data-driven approaches using machine learning show promise for developing risk prediction models for ICIM.
• Concurrent ICI use is a critical factor in ICIM prediction.
• Future studies require larger datasets and integration of laboratory data to enhance predictive accuracy and clinical utility.