In this study, researchers from the Severance Hospital Human Research Protection Center in South Korea, aimed to develop a machine learning (ML)-based clinical decision support system (CDSS) for emergency department (ED) practice. The study drew data from a Level 1 ED of a tertiary teaching hospital in South Korea, between June 2015 and December 2019. During the study, approximately 8000 adult patients visited the ED per month. The ED was divided into four different treatment areas to suit different levels of illness severity.
The researchers included any patient who was aged 18 or above and visited the ED. Data were collected anonymously from the electronic medical records on both physician- and nurse-recorded variables, such as vital signs, illness severity, and laboratory test results. They were combined and split into training and test sets in a 2:3 and 1:3 ratio, respectively. Randomized undersampling was applied to the training set owing to the dataset’s imbalanced characteristics. Subsequently, XGBoosting was used to develop 25 predictive models using 10 important predictors and 69 subfactors.
The outcomes measured for this study included ‘in-hospital cardiac arrest’ (IHCA), ‘inotropic use’, ‘intubation’, and ‘intensve care unit admission’ (ICU). All four variables were identified with the highest consideration as they are deemed to be the most critical outcomes in a medical board setting. The performance of the model was evaluated using sensitivity, specificity, precision, F1 scores, area under the receiver operating characteristic curve (AUROC), and area under the precision-recall curve (AUPRC).
Severance Hospital Human Research Protection Center has been operating since 1885, and is currently a very successful tertiary teaching hospital in South Korea. It is consistently working on providing better, safer healthcare to the South Korean Community. Over the years, it has developed and grown with the times, taking in and advancing the use of medical AI and automation to provide better and more efficient healthcare.
The lead researcher involved in this study was Dr. Jong Min Park. Dr Park is a professor and the director of the Center for Interprofessional Medical Informatics at Division of Medical Informatics, Severance Hospital, Yonsei University College of Medicine in South Korea. With deep knowledge and expertise on medical informatics, he has worked extensively in the development of AI-powered medical workflows and clinical decision support systems. His research in this field is internationally recognized as some of the foremost in this field, with the present study being a prime example.
