Predictive Machine Learning-based Model for Iodinated Contrast Media Hypersensitivity Reaction

Abstract

Background
For present iodinated contrast media hypersensitivity reaction screening guidline in the radiology department, if there are any risk factors, the officers will consult each radiologist to make a decision about either premedication or changing the diagnostic procedure. These will depend on discretion and limitations in after-hours.
Objective
To develop a predictive machine learning-based model for iodinated contrast media hypersensitivity reaction with 24-hour service.
Methods
Descriptive research with retrospective data collection from HOSxP systems collected patients with iodinated contrast media computed tomography at the radiology department of Samutsakhon Hospital from January 2011- December 2021 about 32,489 patients with 1,353 risk factors. There are about 315 patients with contrast media hypersensitivity. Then develop a predictive machine learning-based model with supervised learning type and test efficiency.
Results
From the data of iodinated contrast media hypersensitivity reaction, the predictive machine learning-based iodinated contrast media hypersensitivity reaction risk factors are the history of iodinated contrast media hypersensitivity, hypertension, ischemic heart disease, acquired immune deficiency syndrome, and drug usages such as theophylline and chlorpheniramine. The results of efficiency are accuracy, precision, recall, and f1-score as 0.99, 0.98, 0.98,0.99; respectively. The model reveals in Line Chatbot named " Consult Raidiologist " on the application Line.
Conclusion
Predictive machine learning-based model for iodinated contrast media hypersensitivity reaction with 24-hour service and shows good efficiency.