Machine Learning for Predicting Hyperglycemic Cases Induced by PD-1/PD-L1 Inhibitors

Area of Science:

• Oncology
• Pharmacovigilance
• Computational Biology

Background:

• Immune checkpoint inhibitors (ICIs), such as programmed death-1/ligand-1 (PD-1/L1) therapies, are associated with immune-related adverse events.
• Hyperglycemia is a frequent and severe (grade 3 or higher) immune-related adverse event linked to PD-1/L1 inhibitor use.

Purpose of the Study:

• To develop a machine learning algorithm for rapid and efficient prediction of hyperglycemic adverse reactions in patients receiving PD-1/L1 inhibitors.
• To leverage post-marketing surveillance data for predictive modeling of drug-induced hyperglycemia.

Main Methods:

• Utilized the US Food and Drug Administration Adverse Event Reporting System (FAERS) database for patient data.
• Employed a multivariate pattern classification Support Vector Machine (SVM) model for predicting hyperglycemic adverse reactions.
• Optimized SVM parameters using 10-fold 3-time cross-validation in R language software.

Main Results:

• Developed an SVM prediction model with optimized parameters (nu and gamma) and kernel types ('rbf', 'nu-regression').
• Achieved significant improvements in model performance metrics, including accuracy, F1 score, kappa, sensitivity, and Area Under the Curve (AUC).
• Demonstrated high adjusted R-squared values in curve regressions for key parameters.

Conclusions:

• An effective machine learning model was constructed using specific kernels and computable parameters.
• The SVM model enables noninvasive and precise prediction of hyperglycemic adverse drug reactions (ADRs) in patients on PD-1/L1 inhibitors.
• This predictive tool can assist clinicians in identifying high-risk patients for proactive management and improve clinical decision-making for ADRs.