Machine Learning Model Accurately Predicts Cardiac Arrest in ICU Patients Using ECG Data

Cardiac arrest within hospital settings, particularly in Intensive Care Units (ICUs), remains a significant challenge, occurring in 0.5–7.8% of patients upon hospital admission. Despite advancements in critical care, the unpredictable nature and diverse causes of these incidents make prevention difficult. Quick identification and immediate response are crucial for enhancing patient survival rates. Therefore, there's a pressing need for a system that can accurately and continuously predict in-hospital cardiac arrests, allowing for swift actions like early defibrillation and cardiopulmonary resuscitation (CPR).

To address this need, a team of researchers at Seoul National University Hospital (SNUH, Seoul, South Korea) has developed an innovative machine learning (ML) model. This model uniquely utilizes heart rate variability (HRV) measures from ICU patients to predict in-hospital cardiac arrests. Unlike traditional models that depend on comprehensive electronic medical records (EMR) data, this new approach simplifies prediction by relying solely on HRV measures, enabling real-time and continuous patient monitoring.

Image: Real-time machine learning model predicts in-hospital cardiac arrest using heart rate variability in ICU (Photo courtesy of 123RF)

The study showcased the effectiveness of the light gradient boosting machine (LGBM) model, which excelled in early detection and rapid prediction of in-hospital cardiac arrests. This improvement in prediction accuracy could significantly enhance patient outcomes in clinical settings. The model's strengths include its exclusive use of ECG data for risk prediction, the integration of various HRV measures, and its transparency in explaining risk through these measures.

The exclusive use of ECG data makes this model particularly practical and adaptable to various healthcare environments, as continuous ECG monitoring is a routine procedure in ICUs. This approach contrasts with previous models that required multiple data types, including demographic information, vital signs, and laboratory results. The SNUH team's model, by focusing only on ECG data, presents a more straightforward, feasible solution for predicting cardiac arrests in critical care settings.

Related Links:
SNUH