ML Model for Predicting Contrast-Induced Acute Reactions Achieves AUROC 0.66 in External Validation

A CatBoost model predicted acute adverse reactions to iodine contrast with AUROC 0.66 (95% CI 0.62-0.69) in external validation (111,334 patients), a limited improvement over chance. Top predictors: age and injection rate.

• Retrospective cohort of 332,090 CECT patients; CatBoost selected among five models, trained on 132,102, tested on 56,616, and externally validated on 111,334.
• SHAP identified age, injection rate, contrast type, injection dose, and examination site as top five predictors.
• Single-center retrospective design and modest discriminative ability (AUROC 0.66) limit clinical applicability.

RadPigeon summaries are original and for information only. They are not clinical advice.