A Prediction of in-Hospital Cardiac Arrest Risk Scoring Based on Machine Learning

Le résumé fourni par la source

According to the Korea Disease Control and Prevention Agency (KCDC), 591 out of 33,402 cardiac arrests in 2021 occurred in hospitals. A recent study shows that the golden time to detect a cardiac arrest is less than three minutes. It means early detection of cardiac arrest is important. However, early warning systems predict cardiac arrest with low precision and recall. We research data from ICU patients aged 19 and older who were hospitalized at the Korea University Anam Hospital from 2021 to 2022. We grouped patients with similar characteristics based on clustering the selection, such as in prospective studies. We clustered the training data by window sliding age, SBP, DBP, BT, RR, BP, and BT over 8 hours. We applied a long short-term memory (LSTM) model, a recurrent gated model (GRU) model, and a self-attention-based LSTM model. Instead of linear regression, we used multiple classifications to predict values from 0 to 100. We assign weight to each score. We proposed a cardiac arrest risk score and developed a prediction model for cardiac arrest risk score using ICU patients from the Korea University Anam Hospital. We used the cardiac arrest risk score to predict cardiac arrest within 8 hours, 24 hours, and 72 hours. We evaluated the predicted cardiac arrest risk score as 0 below the threshold and 1 above the threshold. Our proposed GRU model shows 0.11% precision and 94.34% recall.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.