本文根據台灣某醫療院所的實際病例,藉由統計分析的方式,找出造成心臟冠狀動脈繞道手術死亡與併發症的重要因素,並以「羅吉斯迴歸」與「決策樹」建立死亡與併發症的預測模型。我們發現NYHA心臟功能分級以及手術是否處於緊急狀態這兩個變數會同時影響死亡及併發症。另外,術前使用冠狀動脈擴張藥物以及六個月內的急性心肌梗塞亦會使死亡率上升。糖尿病與高血壓則在併發症上扮演重要的角色。各模型的預測正確率皆可達九成以上,ROC index也都高於0.9。本研究亦探討最常用的國外預測模型-EuroSCORE,是否可應用在台灣的案例上,並與本研究所建立之模型相互比較。結果顯示EuroSCORE在台灣也可以得到不錯的預測正確率。不過,我們所建立的預測模型所需要的變數比較少,而且都是手術前的變數,評估容易,預測正確率也比EuroSCORE更高。
Based on the empirical data of a medical center in Taiwan, we identify important risk factors affecting the mortality and morbidity of coronary artery bypass grafting surgery. Forecasting models of mortality and morbidity are established using logistic regression and decision tree. We find that New York Heart Association functional classification and emergent operation are two key factors to predict mortality and morbidity. In addition, renal function index contributes to postoperative mortality in the logistic regression model. The decision tree model identifies that preoperative dependence on nitroglycerine and history of myocardial infarction in recent 6 months increase mortality. With regard to morbidity, diabetes, hypertension, and combined operation are determining factors. The hit ratios of our models are higher than 90% with ROC index being above 0.9. We further investigate the performance of the most popular international forecasting model, EuroSCORE, on Taiwan's data. The results show that EuroSCORE produce decent hit ratios. However, our logistic and decision tree models perform better than EuroSCORE. Our models have better goodness of fit, employ fewer variables, and rely only on preoperative variables.