Korea University Anam Hospital announced Monday that a joint research team led by Professor Lee Jae-myung of critical care and trauma surgery and Professor Baek Seung-min of critical care surgery at Ewha Womans University Mokdong Hospital has developed a model that uses artificial intelligence (AI) machine learning technology to predict the early death risk of trauma patients with high accuracy.
Trauma refers to bodily injury caused by external factors such as traffic accidents, falls, and assault. It is one of the leading causes of death worldwide and, in particular, a cause of early death among younger age groups. Because the condition of trauma patients can deteriorate rapidly in a short time, it is important to accurately identify patients at high risk of death in the early stages. Through risk assessment, medical staff can provide rapid treatment and appropriately allocate necessary medical resources such as intensive care unit and operating room assignments and blood transfusions. However, predicting the prognosis of trauma patients has been difficult because factors such as the circumstances of the accident, the location and severity of injuries, the patient's age and physical condition, pre-hospital emergency treatment, and the course of treatment after arrival at the hospital all interact in complex ways.
The research team obtained community-based trauma survey data published on the National Injury Data portal of the Korea Disease Control and Prevention Agency from 2016 to 2020. After selecting 207,012 analyzable cases out of a total of 237,616, the team applied six machine learning algorithms — logistic regression, k-NN, decision tree, random forest, MLP, and XGB — to develop AI models and comparatively validated their performance in predicting early death in trauma patients.
As a result, the XGB model showed the best performance, recording an AUROC (area under the ROC curve) of 0.985 and an AUPRC (area under the precision-recall curve) of 0.957. AUROC and AUPRC are indicators that evaluate how well an AI model distinguishes actual high-risk patients; the closer to 1, the better the performance. The XGB model maintained stable performance even with 2020 data from the COVID-19 pandemic period, recording an AUROC of 0.984. This suggests the possibility that the model works relatively robustly even in situations where the emergency medical system is heavily affected. In addition, the random forest model also showed high predictive performance with an AUROC of 0.984 and an AUPRC of 0.956.
Through additional analysis, the research team confirmed that factors such as pre-hospital cardiac arrest, Injury Severity Score (ISS), age, and time to first transfusion are major factors that significantly influence the prediction of death risk. ISS is an indicator that expresses in scores how severe a patient's injury is. The study is evaluated as a meaningful achievement in that the model was developed based on nationwide public data rather than a single medical institution, while considering predictive performance, interpretability, scalability, and generalizability together.
Professor Lee Jae-myung said, "In the future, it could be used to quickly screen the risk levels of critically ill patients in the emergency medical system and at trauma treatment sites."
Professor Baek Seung-min said, "We confirmed the possibility of screening early death risk using nationwide trauma registry data." He added, "Through further calibration and prospective validation going forward, we will develop this into research that integrates AI-based risk screening."
The findings were published in the World Journal of Emergency Surgery, a prominent academic journal in the field of emergency medicine and surgery.
The research team obtained community-based trauma survey data published on the National Injury Data portal of the Korea Disease Control and Prevention Agency from 2016 to 2020. After selecting 207,012 analyzable cases out of a total of 237,616, the team applied six machine learning algorithms — logistic regression, k-NN, decision tree, random forest, MLP, and XGB — to develop AI models and comparatively validated their performance in predicting early death in trauma patients.
As a result, the XGB model showed the best performance, recording an AUROC (area under the ROC curve) of 0.985 and an AUPRC (area under the precision-recall curve) of 0.957. AUROC and AUPRC are indicators that evaluate how well an AI model distinguishes actual high-risk patients; the closer to 1, the better the performance. The XGB model maintained stable performance even with 2020 data from the COVID-19 pandemic period, recording an AUROC of 0.984. This suggests the possibility that the model works relatively robustly even in situations where the emergency medical system is heavily affected. In addition, the random forest model also showed high predictive performance with an AUROC of 0.984 and an AUPRC of 0.956.
Through additional analysis, the research team confirmed that factors such as pre-hospital cardiac arrest, Injury Severity Score (ISS), age, and time to first transfusion are major factors that significantly influence the prediction of death risk. ISS is an indicator that expresses in scores how severe a patient's injury is. The study is evaluated as a meaningful achievement in that the model was developed based on nationwide public data rather than a single medical institution, while considering predictive performance, interpretability, scalability, and generalizability together.
Professor Lee Jae-myung said, "In the future, it could be used to quickly screen the risk levels of critically ill patients in the emergency medical system and at trauma treatment sites."
Professor Baek Seung-min said, "We confirmed the possibility of screening early death risk using nationwide trauma registry data." He added, "Through further calibration and prospective validation going forward, we will develop this into research that integrates AI-based risk screening."
The findings were published in the World Journal of Emergency Surgery, a prominent academic journal in the field of emergency medicine and surgery.
