Machine Learning Model to Predict Iodine Contrast Media-Related Acute Adverse Reaction in Patients Without a Similar...

Area of Science:

• Radiology
• Medical Informatics
• Clinical Prediction Modeling

Background:

• Iodine Contrast Media (ICM) administration is common in CT scans.
• Acute Adverse Reactions (AAR) to ICM can occur, necessitating risk prediction.
• Predicting ICM-related AAR in patients without prior reactions is crucial for patient safety.

Purpose of the Study:

• To develop and compare machine learning models for predicting ICM-related AAR risk.
• To identify key clinical variables associated with ICM-related AAR.
• To construct a predictive nomogram based on the best-performing model.

Main Methods:

• Retrospective enrollment of 546 patients without prior ICM-related AAR history.
• Feature selection using univariate analysis and LASSO.
• Development and evaluation of four ML models: LR, decision tree, k-NN, linear SVC.
• Construction of a nomogram using the superior model.

Main Results:

• Key predictors identified: ICM exposure history, other allergies, hypertension, ICM type/dose, metformin use, hyperglycemia, GFR.
• Logistic Regression (LR) model showed superior performance.
• LR model achieved AUCs of 0.894 (test) and 0.814 (external validation).

Conclusions:

• The LR-based model accurately predicts ICM-related AAR risk using accessible clinical data.
• The developed nomogram serves as a practical tool for pre-administration risk assessment.
• This facilitates timely implementation of preventive strategies for high-risk patients.