Repository logo
  • English
  • 中文
  • Log In
    New user? Click here to register.Have you forgotten your password?
Repository logo
    Communities & Collections
    Research Outputs
    Fundings & Projects
    People
    Organizations
    Statistics
  • English
  • 中文
  • Log In
    New user? Click here to register.Have you forgotten your password?
  1. Home
  2. College of Health Care and Management/健康管理學院
  3. Department of Medical Informatics/醫學資訊學系
  4. 運用集成學習提升預測急診病人發生院內不預期心跳停止風險因子之研究
 
  • Details
Options

運用集成學習提升預測急診病人發生院內不預期心跳停止風險因子之研究

Date Issued
2026
Author(s)
陳吟祺
CHEN,YIN-CHI
Advisor
許文馨
陳碩聰
HSU, WEN-SHIN  
CHEN, SHUO-TSUNG
URI
https://handle.ncl.edu.tw/11296/6s3s9c
https://scholars.csmu.edu.tw/handle/123456789/21216
Abstract
研究背景與目的:院內不預期心跳停止(In-Hospital Cardiac Arrest, IHCA)屬於一種緊急醫療狀況,只有四分之一的患者能事前被發現。IHCA是有機會避免的,在心跳驟停的前幾小時,可以偵測到病人生理指標惡化的情形。儘管現行有MEWS等生理評分工具,但在處理急診動態且非線性的病理變化上仍有局限。本研究旨在運用集成學習演算法,辨別急診病患發生IHCA之關鍵相關因素,並建立一套早期預測模型。研究方法:本研究採回溯性研究設計,分析北部某聯合醫院急診病患資料,收集共有228筆急診病患之臨床資料,涵蓋人口統計學、生命徵象及臨床診斷等變數。研究採用多種機器學習模型進行訓練與比較,包括隨機森林(Random Forest)、XGBoost及CatBoost。針對臨床數據缺失值,採刪除法策略以維持樣本完整性,並透過Python 3.9.21進行數據分析,以及超參數調校。研究結果:實驗分析顯示,機器學習模型在預測 IHCA 發生上展現出優於傳統評分工具之潛力。三種預測模型之準確率(Accuracy)均優於0.85。各模型皆具有一定程度的預測能力,沒有明顯特別劣勢的模型。綜合各項指標來看,以 CatBoost(準確率 0.89)與隨機森林(準確率 0.87)在整體的方面表現較為優異一些。在特徵重要性分析中,收縮壓、體溫、肺炎病史以及是否曾進行CPR與是否ROSC等變數對於模型預測具有顯著相關,反映出初始生理儲備能力與嚴重發炎反應對IHCA的關鍵影響。結論與建議:本研究初步證實了機器學習於急診IHCA預測之臨床價值。模型所識別之關鍵特徵,如:CPR狀態、血壓波動和感染史,都與臨床觀點相符,具備發展為即時分層預警系統之潛力。未來建議結合SHAP值提升模型可解釋性,或是擴大樣本量,加入分層預警機制,如中風險增加監測、高風險啟動醫療介入,以降低臨床警報疲乏,以進一步提升臨床決策的透明度與信任感。
Background and Purpose: In-Hospital Cardiac Arrest (IHCA) is a critical emergency, with only 25% of cases detected beforehand. Although IHCA is preventable through early physiological signals, conventional tools like MEWS fail to capture the dynamic, non-linear pathological changes in emergency settings. This study utilizes ensemble learning algorithms to identify key IHCA risk factors in emergency department (ED) patients and build an early predictive model.Methods: This retrospective study analyzed clinical data from 228 ED patients at a joint hospital in northern Taiwan, covering demographics, vital signs, and diagnoses. Random Forest, XGBoost, and CatBoost models were trained and compared. Listwise deletion was used for missing data, and analysis was performed using Python 3.9.21 with hyperparameter tuning.Results: Machine learning models outperformed traditional scoring tools, with all three models achieving accuracies above 0.85. CatBoost (Accuracy: 0.89) and Random Forest (Accuracy: 0.87) demonstrated the best overall performance. Feature importance analysis showed that sblood pressure, body temperature, history of pneumonia, and resuscitation status (prior CPR/ROSC) were significantly correlated with IHCA, reflecting the critical impact of baseline physiological reserve and severe infection.Conclusion: This study validates the clinical value of machine learning for ED IHCA prediction. The identified key features—CPR status, blood pressure fluctuations, and infection history—align with clinical insights, showing strong potential for real-time stratified early warning systems. Future work should integrate SHAP values to enhance explainability or expand the sample size to implement tiered interventions, reducing alert fatigue and improving clinical decision trust.
Subjects
院內不預期心跳停止
集成學習
急診醫學
預測模型
In-hospital cardiac arrest
Ensemble learning
Emergency medical
Predictive model
Type
thesis
File(s)
No Thumbnail Available
Name

114CSMU0585002-003.pdf

Size

2.09 MB

Format

Adobe PDF

Checksum

(MD5):5909519bf4db57931169b6637a04145c

Make it your own

DSpace-CRIS can be extensively configured to meet your needs. Decide which information need to be collected and available with fine-grained security. Start updating the theme to match your Institution's web identity.

Built with DSpace-CRIS software - Extension maintained and optimized by 4Science

  • Cookie settings
  • Privacy policy
  • End User Agreement
  • Send Feedback