"AI Predicts Cardiac Arrest in ICU Patients in Real Time"

Seoul National University Hospital Anesthesiology and Pain Medicine Research Team
Development of Real-Time Cardiac Arrest Prediction AI-Based Machine Learning Model

A domestic research team has developed an artificial intelligence (AI)-based machine learning model that can predict cardiac arrest in intensive care unit (ICU) patients in real time.

[Photo by Pixabay]

The Korea Health Industry Development Institute announced on the 15th that a research team from the Department of Anesthesiology and Pain Medicine at Seoul National University Hospital succeeded in developing an AI model that accurately predicts the risk of cardiac arrest within 24 hours using heart rate variability (HRV) extracted from electrocardiogram (ECG) data. HRV is an indicator that measures how much the time intervals between consecutive heartbeats vary, reflecting the heart's health status and autonomic nervous system activity.

Acute cardiac arrest in the ICU is known to occur globally at a rate of approximately 0.5% to 7.8%. Early prediction and rapid response to acute cardiac arrest play a crucial role in increasing patient survival rates and reducing complications.

Using patients' vital signs is important for early prediction of cardiac arrest in the ICU. In particular, ECG is the most commonly used vital sign in the ICU, and AI algorithms utilizing it have high versatility and applicability in the ICU environment.

The research team developed a machine learning model for real-time cardiac arrest prediction using HRV extracted from the ECGs of 5,679 patients admitted to the ICU at Seoul National University Hospital. The model development utilized 33 HRV indicators extracted from a single 5-minute channel ECG. According to the research team, this showed excellent performance in predicting cardiac arrest within 24 hours.

The AI prediction performance of the model was evaluated with an AUROC (Area Under the Receiver Operating Characteristic) value of 0.881. This is higher compared to the AUROC value of 0.735 for existing cardiac arrest prediction models based on vital signs. AUROC is an indicator that reflects both sensitivity, the ability to correctly predict true positives, and specificity, the ability to correctly predict true negatives. The closer the AUROC value is to 1, the better the performance.

Professor Hyunsoon Lee of Seoul National University Hospital said, "The model developed in this study represents a technological breakthrough in that it developed a new algorithm using only a single-channel ECG without additional clinical information," adding, "It will be easily applicable to clinical decision support systems (CDSS) using AI in ICU clinical settings in the future."

Professor Hyungchul Lee of Seoul National University Hospital stated, "The developed AI model will be utilized to develop real-time cardiac arrest risk prediction alarms in the ICU," and added, "By screening patients at high risk of cardiac arrest early, it is expected to reduce complications caused by it and also reduce medical costs."

This study was conducted with the support of the 'Ministry of Health and Welfare's ICU-specialized big data construction and AI-based CDSS development project' and was published on the 23rd of last month in the scientific journal Nature Digital Medicine (npj Digital Medicine).

Professors Hyun-Hoon Lee (left) and Hyung-Cheol Lee, Seoul National University Hospital. [Photo by Korea Health Industry Development Institute]

This content was produced with the assistance of AI translation services.