Interpretable machine learning model predicting immune checkpoint inhibitor-induced hypothyroidism: A retrospective...

Area of Science:

• Oncology
• Endocrinology
• Artificial Intelligence

Background:

• Immune checkpoint inhibitors (ICIs) are crucial in cancer treatment but can cause hypothyroidism.
• Early prediction of this adverse event is vital for patient management.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for predicting hypothyroidism in patients receiving ICIs.
• To identify key predictors of ICI-induced hypothyroidism.

Main Methods:

• Retrospective cohort study of 458 patients treated with ICIs.
• Applied logistic regression, random forest, SVM, and XGBoost models.
• Utilized SHAP for model interpretability and AUC for evaluation.

Main Results:

• Hypothyroidism developed in 12.88% of patients.
• XGBoost achieved the highest predictive performance with an AUC of 0.833.
• Thyroid-stimulating hormone (TSH) was identified as the most significant predictor.

Conclusions:

• An interpretable ML model, particularly XGBoost, can effectively predict ICI-induced hypothyroidism.
• This approach supports personalized risk management and treatment strategies for cancer patients.
• ML offers a promising tool for proactive monitoring of endocrine adverse events during ICI therapy.