應用基因演算法暨機器學習對於急診創傷病人作未來相關預後預測之研究

A Genetic algorithm with Machine Learning for prediction of progression to Emergency Trauma Patients for Prediction of Prognosis

莊家焮(Chia-Hsin Chuang)

指導教授 ： 趙坤茂

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201902067

摘要

在台灣，急診每天都要處理許多創傷病患，以新北市的區域醫院台北慈濟醫院為例，這些傷患中，外科病人約占三分之一的來診量。其中重大創傷是這些外科病人最常見的死亡原因，然而，有些傷患的生命徵象在剛入院時相對狀況穩定，但病情卻在數小時內病情急速惡化。目前醫師沒有一個固定的評分標準，而是以各自不同的評估指數，如: 休克指數(SI)、改良式創傷嚴重程度指標(RTS)、外傷嚴重度分數(ISS)、創傷嚴重度分數(TRISS)、新外傷嚴重度分數(NISS)以及qSOFA為依據，再根據各評分將病人作不同處置。倘若急診當時若發生大量傷患事件，多位病人情況緊急且多個評分在較差的情況下，哪些病人需要多加注意或者以社區醫院來說可能會因為設備不足的問題需要轉診大型醫院，這些問題仍是以醫師的經驗作決定，所以在能讓醫師快速判斷一位創傷病人的後續處理是一件很重要的事情。
本研究主旨為利用機器學習的方式將2008年1月1日至2019年5月16日的慈濟醫院急診創傷病人總共13144筆資料160個特徵值做分析預測。欲得知除了以上的評分方式，是否還有其他的特徵值可作為判斷病人後續的依據，經過資料前處理過後，資料量為11656個病人資料及27個特徵值以C4.5演算法做資料離散化為153個，最後將處理後的資料做監督式機器學習，探討兩種類型，其一為輸出值為2種: 康復、死亡。將資料特徵利用特徵工程創造新欄位，利用Genetic algorithm (GA)及Logistic Regression (LR) 找特異性與敏感性之和越高的特徵，並創造找到新的判斷創傷病人的評分方式，讓醫師遇到病患時能夠快速判斷其是否高機率死亡。

關鍵字

Logistic Regression(LR) ； C4.5 ； Genetic algorithm (GA) ； 急診 ； 創傷 ； 預後分析

並列摘要

According to Taipei Tzu Chi Hospital, Traumatic injuries in emergency department are common and constitute one third of emergency patients in the hospital. However, vital signs of some patients are in stable condition, but dramatically worsened. Currently, doctors use some score like shock index, revised trauma score (RTS), injury severity score (ISS), trauma injury severity score (TRISS), new injury severity score (NISS) and quick sepsis related organ failure assessment score (qSOFA) to estimate condition of patients, but the effective and efficient were necessary to be investigated in traumatic population.
In this case, we totally included 13144 patients with 161 fields in emergency department of Taipei Tzu Chi Hospital from January 1st 2009 to May 16th 2019. The data of January 2009 to December 2017 were set as training model and the data after 2018 were set as testing model. After data processing, the quantity of data become 11656 patients with 27 features. Then the features are discretized by the C4.5 algorithm and then become 153 features. Finally, the processed data is used for supervised machine learning. Two types are discussed: One is the output value: live and death. We used a genetic algorithm (GA) to find the highest sum of specificity and sensitivity for suitable prognosis factor and logistic regression for classification and prediction. The aim of our research is to find new prognosis grading score that make physicians make clinical decision faster.

並列關鍵字

Logistic Regression (LR) ； C4.5 ； Genetic algorithm (GA) ； Emergency ； Trauma ； Prediction of Prognosis

參考文獻

[1] 衛生福利部中央健康保險署. 全民健康保險急診品質提升方案. 2016 03.01

[2] 衛生福利部中央健保署, 潘. 107年醫院總額部門執行成果報告. 2018

[3] Chalfin, D.B., et al., Impact of delayed transfer of critically ill patients from the emergency department to the intensive care unit. Crit Care Med, 2007. 35(6): p. 1477-83.

[4] Richardson, D.B., Increase in patient mortality at 10 days associated with emergency department overcrowding. Med J Aust, 2006. 184(5): p. 213-6.

[5] Sprivulis, P.C., et al., The association between hospital overcrowding and mortality among patients admitted via Western Australian emergency departments. Med J Aust, 2006. 184(5): p. 208-12.