叢集式類神經網路應用於到院前心肺功能停止病患之生存率輸入變數靈敏度分析

Ensembled Neural Networks Applied to Sensitivity Analysis of Survival Rate Index for the Patients with Out of Hospital Cardiac Arrest

江元彰(Yuan-Chang Chiang)

指導教授 ： 謝建興

元智大學/工程學院/機械工程學系/碩士(2012年)

https://doi.org/10.6838/YZU.2012.00148

摘要

本研究針對到院前心肺功能停止之大規模資料分析，將2010年1月至2011年6月的OHCA病患資料進行生存率與葛式昏迷指數的預測，以了解影響生存率與葛式昏迷指數的重要變因，資料清理後共有4095筆完整資料來供叢集式類神經網路建立模型，我們將4095筆資料分為3636筆內科病患與459筆外科病患資料來探討11個變因(反應時間、現場時間、運送時間、年齡、CPR時間、目擊者是否CPR、是否使用除顫器、性別、是否現場CPR、呼吸道處置、是否用藥)的重要性，分別以5個輸出(2小時存活、24小時存活、24小時葛式昏迷指數、48小時葛式昏迷指數、72小時葛式昏迷指數)做測試，結果為72小時葛式昏迷指數所建立之模型較為準確，再做靈敏度分析後內外科相同者且排名前五名的重要預測因素分別為是否使用除顫器、目擊者是否CPR、呼吸道處置和CPR時間，此結果與傳統統計分析中羅吉斯迴歸分析檢定結果相似，僅年齡與羅吉斯迴歸分析檢定結果不同。此結果可知緊急救護團隊能掌握的病患存活關鍵在於是否使用除顫器、呼吸道處置過程。

關鍵字

緊急醫療救護 ； 敏感度分析 ； 類神經網路

並列摘要

This research has analyzed large database of out of hospital cardiac arrest from January 2010 to June 2011. After data cleaning, we only have 4095 patients left to use for further analysis. In order to separate trauma type, we divided 4095 patients to 3636 patients for internal medicine type and 459 patients for surgery type. Then, we use these two types to build five ensemble neural network for different outputs (2-hour survival, 24-hour survival, 24-hour GCS, 48-hour GCS, 72- hour GCS). We separate trauma type and invistigate which 11 variables (response time, on-scene time, patient transfer time, age, time to CPR, eyewitness on the scene, using AED, gender, CPR on the scene, using drugs, using airway) are important. Because the accuracy of 72- hour GCS model is best, we use sensitivity analysis to find out the important variables in this model. After we compare these two types of internal medicine and surgery, the same results of four input variables in the top five ranking are using AED, eyewitness on the scene, using airway and time to CPR. Moreover, these variables have also been shown significant differences when we use traditional one variable statistics analysis except only age. So, emergency medical system should focus on using AED and using airway to improve the survival rate.

並列關鍵字

Emergency Medical System ； sensitivity analysis ； neural network

參考文獻

[1]. D.L. Atkins, S.E. Stewart, G.K. Sears, M. Daya, M.H. Osmond, C.R. Warden, R.A. Berg: "Epidemiology and Outcomes From Out-of-Hospital Cardiac Arrest in Children (The Resuscitation Outcomes Consortium Epistry–Cardiac Arrest) " Circulation, pp.1484-1491, 2009.

[2]. American Heart Association (AHA): "Highlights of the 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care" Volume 16 Number 4 Winter, 2005-2006.

[3]. J. Hollenberga, J. Lindqvist, M. Ringha, J. Engdahl, K. Bohm, M. Rosenqvist, L. Svensson: "An evaluation of post-resuscitation care as a possible explanation of a difference in survival after out-of-hospital cardiac arrest" Resuscitation, pp. 242-252, 2007.

[4]. American Heart Association (AHA): "Highlights of the 2010 American Heart Association Guidelines for CPR and ECC", Circulation, 2010.

[5]. Y. Youichi, S. Toshihisa: "Analysis of Prehospital Care For Cardiac Arrest In An Urban Setting In Japan" The Journal of Emergency Medicine, Vol. 38, No. 3, pp.340–345, 2010.