Web-Based Explainable Machine Learning-Based Drug Surveillance for Predicting Sunitinib- and Sorafenib-Associated...

Area of Science:

• Oncology
• Pharmacovigilance
• Data Science

Background:

• Traditional methods for identifying high-risk patients are limited.
• Machine learning models utilizing time-series data offer predictive capabilities for adverse events in cancer patients.
• Timely management of cancer treatment side effects is crucial.

Purpose of the Study:

• To develop and validate machine learning models for predicting thyroid dysfunction associated with sunitinib and sorafenib.
• To employ a time-series data collection approach for enhanced predictive accuracy.
• To identify key predictors of drug-induced thyroid dysfunction.

Main Methods:

• Collected time-series data from patients treated with sunitinib or sorafenib.
• Developed predictive models using logistic regression, random forest, adaptive Boosting, Light Gradient-Boosting Machine, and Gradient Boosting Decision Tree.
• Evaluated model performance using accuracy, precision, recall, F1-score, AUC-ROC, and AUC-PR.
• Utilized SHapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• The Gradient Boosting Decision Tree model demonstrated superior performance.
• The best model achieved an AUC-PR of 0.600 and AUC-ROC of 0.876.
• Key predictors identified include elevated cholesterol, prolonged medication duration, and clear cell adenocarcinoma histology.
• The model was integrated into a web-based application for practical use.

Conclusions:

• Developed an explainable adverse drug reaction surveillance system.
• The model effectively predicts sunitinib- and sorafenib-associated thyroid dysfunction.
• This tool supports proactive patient management and enhances drug safety monitoring.