Predicting short-term survival after liver transplantation using machine learning

• Main Authors: Jiang Guo-Wei, 江國瑋
• Other Authors: Liu, Chien-Liang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/5rq9gb

碩士 === 國立交通大學 === 工業工程與管理系所 === 106 === Liver disease is one the leading causes of death worldwide owing to the change of lifestyle and environmental pollution, and liver transplantation is the only curative treatment for end-stage liver disease. However, the demand for livers is much higher than the number of available donor livers, explaining why patients on the waiting list for a liver transplant have to be prioritized. Currently, most medical staffs rely on the Model for End-stage Liver Disease (MELD) score to prioritize patients, but previous research studies have indicated that MELD score may fail to predict well for the postoperative patients. This work proposes to use data driven approach to devise a new scoring system to estimate postoperative survival rate based on patient’s preoperative physiological measurement values, in which we consider stability and prediction accuracy simultaneously. This work uses machine learning along with real data set to construct and validate the proposed model. We use random forest to select important features, including clinically used features and new features discovered from physiological measurement values. And proposed a new method to replace the missing value. Then, we use patients’ blood test data within 2-10 days before surgery to construct a predictive model to predict postoperative patients’ survival rates, in which we use random forest as the learning algorithm and compare with XGBoost, decision tree and logistic regression. The experimental results indicate that random forest outperforms the other alternatives, and it could achieve comparative results (AUC=0.817). As a whole, the proposed scoring system can effectively help medical staffs allocate donor livers to suitable candidates for maximizing the benefits.