(주) 뷰노

(주) 뷰노

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

Enhancement of Prediction of In-Hospital Cardiac Arrest of Deep learning based Early Warning Score (DEWS) in a Single Tertiary Hospital

  • Oct. 2025
  • by Sun Jung Kwon et. al.

INTRODUCTION
  In-hospital cardiac arrest (IHCA) is an event that significantly impacts on patient safety. In previous studies, Deep learning based early warning scoring (DEWS) has been effective in predicting IHCA and to reduce false alarms.


OBJECTIVES
  We examined whether DEWS outperforms MEWS in IHCA prediction while reducing false alarms and improving with site-specific learning.

 

METHODS
  DEWS analyzes age, blood pressure, heart rate, respiratory rate, and body temperature (DeepCARSTM, VUNO Inc, Korea). A retrospective analysis was conducted on adult patients admitted to general wards. The implementation period was from January 2022 to January 2023 and the site-specific learning period was from December 2022 to November 2023. The model was trained using one year of data for each period. The improvement in IHCA prediction was compared between the two periods.

 

RESULTS
  During the implementation period, 56,155 admissions were recorded, while 23,649 admissions were analyzed during the learning period. During implementation and learning period, there were 2.22 and 2.36 cardiac arrests per 1,000 admissions, respectively. During the implementation period, the AUROC of DEWS was 0.832 (p<0>   During the learning period, the DEWS outperformed conventional triggering system in predicting IHCA and unplanned intensive care unit transfer (UIT)(AUROC; 0.8623 vs. 0.7158 NEWS vs.0.5790 SPTTS) and in predicting IHCA (AUROC; 0.8589 vs. 0.7141 NEWS vs.0.5723 simple parameter track-and-trigger system (SPTTS)) (Figure 1). The DEWS increased significantly sensitivity and specificity at implementation and learning period (20.2% and 99.2%  vs 17.6% and 98.6%) in same mean alarm counts per day (Table 1). False alarm was more reduced in DEWS than conventional triggering system. DEWS predicted 6 additional IHCA earlier than NEWS in the 18 hours pre-IHCA interval and earlier than NEWS (Figure 2).

 

Figure 1. The ROC analysis demonstrates that DCARS outperforms NEWS and SPTTS  in predicting all events (A) and cardiac arrest (B) at learning period.

Figure 2. Comparison of Cumulative Sensitivity for CPR Prediction within 6 to 24 Hours Before Event. Overall, DeepCARS (≥91) demonstrates superior early prediction capability compared to NEWS (≥5). In the 18-hours pre-CPR interval, DCARS predicted 6 additional CPR events earlier than NEWS. A similar level of early prediction was achieved 5.7 hours earlier with DCARS compared to NEWS.
 

 

CONCLUSIONS
  Our study showed that DEWS outperformed MEWS in predicting IHCA and that site-specific customization enhanced IHCA prediction performance.


REFERENCES
1. Lee YJ, Cho KJ, Kown O, et al. A multicentre validation study of the deep learning-based early warning score for predicting in-hospital cardia arrest in patients admitted to general wards. Resuscitation. 2021; 163: 78-85
2. Kwon SJ, Lee MH, Kang DE, et al. Deep learning based early warning score for predicting in-hospital cardiac arrest in a rapid response team with minimal manpower. Abstract. ICMx, 11(suppl 1), 72(2023).
3. Yang J, Soltan AA, Clifton DA. Machine learning generalizability across healthcare settings: insights from multi-site COVID-19 screen. Digital Medicine 2022; 69:

Tag in

#biosignal

#VUNO Med®-DeepCARS™

이전글
이전글이 없습니다.
다음글
Development and Validation of an AI-ECG model for Detecting Transthyretin Amyloid Cardiomyopathy in Multi-Hospital Data 2025. 10. 16
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.