'24-Hour Cardiac Arrest Prediction' Published in SCI-Level Journal
Overcoming AI Algorithm Bias to Improve Prediction Accuracy
Integrated into Real-Time Inpatient Monitoring System
AI wearable medical device company Seeas Technology announced on the 19th that it has published research results in an SCI-level academic journal that improve the accuracy of predicting the risk of cardiac arrest in hospitalized patients.
The research results are a follow-up study to the paper titled "Explainable AI Warning Model Using an Ensemble Approach for In-Hospital Cardiac Arrest Prediction: Retrospective Cohort Study," published in the international journal JMIR (Journal of Medical Internet Research) in December 2023. The study was conducted to develop an AI cardiac arrest prediction algorithm optimized for clinical settings.
Seeas reduced the input cycle of vital signs in the AI model used in the previous study from 24 hours to 12 hours and enhanced the prediction time for cardiac arrest occurrence from within 6 hours to within 24 hours. Instead of using the commonly employed representative event collection method for validating prediction results, the study implemented an actual clinical environment to verify the real-time cardiac arrest prediction performance of hospitalized patients according to various inpatient environments.
The research results showed a performance indicator (AUROC) of 0.8 for prediction accuracy, which was superior to existing methods. This means that, regardless of different inpatient environments and patient characteristics, the model can provide cardiac arrest prediction alerts with up to 80% accuracy in real clinical settings. Under the same conditions, it improved accuracy by up to 26 percentage points (P) compared to existing cardiac arrest prediction models. The false alarm rate of predictions decreased by more than 20 percentage points compared to previous studies. Compared to existing methods, it not only predicts the likelihood of cardiac arrest within 24 hours with 80% accuracy but also increased the reliability of high-risk alarms by 20%.
A company representative stated, "The proposed AI model focuses on minimizing algorithmic potential bias by analyzing the statistical information of vital signs over time and the imbalance of vital sign data." They added, "Compared to existing cardiac arrest prediction models, it showed consistent cardiac arrest prediction results and accuracy across various inpatient environments."
Seeas also provides key vital sign information along with the early risk of cardiac arrest occurrence. The company explained that this means the AI prediction model can be used as a reliable clinical decision support system that warns medical staff early of cardiac arrest risk and helps identify its causes.
Seeas plans to integrate this AI model into its real-time inpatient monitoring system, thynC™. ThynC™ is a smart ward solution that automatically detects abnormal signs by analyzing patients' vital data measured by wireless wearable medical devices in real time based on AI, helping medical staff respond efficiently and immediately.
While existing cardiac arrest prediction models operated based on patients' vital signs and EMR data measured intermittently 3 to 4 times a day, thynC™ is based on real-time measured data, allowing immediate response to changes in patient condition, thus offering higher prediction performance and utility.
Lee Young-shin, CEO of Seeas, said, "Beyond diagnosis and monitoring services using wearable AI technology, we are now focusing our research capabilities on resolving blind spots in patient management due to increasing demand for medical care and hospitalization through disease prediction." He added, "We will continue to commercialize AI models related to predicting the worsening of hospitalized patients, such as cardiac arrest prediction, emergency arrhythmia prediction, and sepsis prediction, and apply them in clinical settings."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
