透過您的圖書館登入
IP:18.118.227.69
  • 期刊

運用機器學習於加護病房肝硬化重症病人之死亡預測

Machine Learning to Predict the Death of Patient in the Intensive Care Unit with Cirrhosis

摘要


機器學習已被應用於多個領域且在醫療預測方面亦已取得成效,目前臨床上尚無針對肝硬化病人在重症環境中的評估量表,因此本研究期望能利用機器學習方法建構肝硬化重症病人之死亡預測模型。研究採電子病歷回溯性調查,以東部某區域醫院2013年8月1日至2015年12月31日入住加護病房之肝硬化287位個案進行分析,透過R語言使用五種學習方法:Bagging、GLM、AdaboostNeyman-Pearson、Stacking分別建構五個模型,並且和臨床傳統常用的肝病病人嚴重度評估量表CTP score、MELD score以及常用於重症加護病房疾病的評估量表APACHE II進行比較。本研究建構的五個機器學習模型準確率達八成以上,其中Neyman-Pearson預測效果最佳準確率86.2%(AUC=0.871),且皆優於以傳統臨床評估量表APACHE II(AUC=0.806)、MELD score(AUC=0.755)、CTP score(AUC=0.747)建構的死亡預測模型,其預測結果期望能夠提供醫療人員決策參考。

並列摘要


Machine learning (ML) is used in a wide variety of applications and achieve certain results in medical science. At present, there is no clinical assessment scale to available patients with critical cirrhosis patients. Therefore, this study used ML methods to construct a death prediction model for critical cirrhosis patients. A retrospective investigation is designed with electronic medical records to make analysis on 287 patients with cirrhosis in the intensive care unit of a regional hospital from August 1, 2013 to December 31, 2015. Using R language to achieve five ML methods: Bagging, GLM, Adaboost, Neyman-Pearson, and Stacking to construct five models. The study compare the discriminative ability of traditional clinical scoring systems: CTP score and MELD score, and APACHE II. The accuracy of the five ML models in this research was above 80%. Top-performance is Neyman-Pearson accuracy rate was 86.2% (AUC=0.871), is get ahead of traditional scoring system: APACHE II (AUC=0.806), MELD score (AUC=0.755), CTP score (AUC=0.747), and its prediction results are expected to provide reference for medical decision making.

並列關鍵字

無資料

延伸閱讀