(주) 뷰노

(주) 뷰노

문의사항 남기기
KR EN
VUNO MED INC.
Publications
BACK TO LIST
Medinfo 2025

Enhancing Cardiac Arrest Prediction in Critically Ill Patients: A Sequence-Based Embedding Approach with Mamba

  • Aug. 2025

Abstract. We developed a novel model for cardiac arrest prediction using ICU data from SNUH and externally validated it on PNUYH. Our sequence-based embedding approach with a Mamba encoder outperformed NEWS and LightGBM, achieving AUROC 0.957 (internal) and 0.889 (external). This study highlights the potential of the proposed model as a robust and generalizable tool for early ICU intervention.


Keywords. Intensive care unit, cardiac arrest, predictive model, validation study

 

1. Introduction
Cardiac arrest in the intensive care unit (ICU) is a critical event with high mortality, necessitating early prediction to enable timely interventions and improve patient
outcomes. We developed and validated a novel predictive model using ICU data from Seoul National University Hospital (SNUH) and externally validated it on data from Pusan National University Yangsan Hospital (PNUYH). This dual-validation approach underscores the model’s robustness and generalizability across different clinical settings.

 

2. Methods
Data were collected from ICU patients admitted between 2007–2021 at SNUH and 2010–2021 at PNUYH. SNUH dataset was divided into a development set (2007–2018) and a validation set (2019–2021). The model predicts the risk of cardiac arrest within a 12-hour window in real time, utilizing nine vital signs, including the Glasgow Coma Scale, and 21 routine blood tests. This study was approved by the institutional review board of SNUH and PNUYH (IRB No. 2111-140-1275, IRB No. 05-2021-291). The requirement for individual consent was waived due to the retrospective nature of the study.
The model combines a Mamba encoder with a multi-layer perceptron, using sequence-based embedding to capture temporal variations and heterogeneous feature characteristics. Inputs with varying recording intervals, such as vital signs and laboratory results, are encoded separately. This allows the model to maximize data utilization before and at the prediction point and adapt to the unique patterns of each modality.


3. Results
Our model achieved superior performance with AUROC 0.957 (internal) and 0.889 (external), significantly outperforming existing scoring systems and predictive models.

 

4. Conclusion
This study demonstrates that our model consistently surpasses traditional scoring systems in cardiac arrest prediction, highlighting its potential as a reliable tool for proactive ICU management.

 

 

Link

https://pubmed.ncbi.nlm.nih.gov/40776222/

Author

Ukdong Gim¹, Yunseob Shin¹, Dongjoon Yoo¹ ², Kyungjae Cho¹, Hyung-Chul Lee³, Leerang Lim³, Woo Hyun Cho⁴

이전글
이전글이 없습니다.
다음글
Revolutionizing Radiology Reporting With Artificial Intelligence-Based Voice Recognition: A Pilot Study on Lumbar Spine Magnetic Resonance Imaging 2025. 08. 04
List

이 웹사이트는 원활한 서비스 제공을 위하여 당사 개인정보 처리방침에 명시된 쿠키를 사용합니다.

만일 당사의 쿠키 사용 정책에 대해 더 알고 싶으신 경우에는 개인정보 처리방침 을 참고하여 주시기 바랍니다.