Options
以機器學習模型訓練臨床資料預測臺灣成年人肝癌與過重或肥胖之關聯性
Other Title
Machine Learning Model for Predicting the Risk of Liver Cancer Associated with Overweight or Obesity Among Taiwanese Adults Using Clinical Data
Type
thesis
Date Issued
2022-06-07
Author(s)
卡拉姆
Advisor
SHABBIR, SYED ABDUL
Subjects
系所名稱：醫學資訊研究所碩士班
Publisher
醫學資訊研究所碩士班
Description
口試委員：張資昊 CHANG, TZU-HAO;SHABBIR, SYED ABDUL SHABBIR, SYED ABDUL;吳育瑋 WU, YU-WEI
網際網路,開放日期為2022-07-01
網際網路,開放日期為2022-07-01
Abstract
Background: Liver cancer has become one of Taiwan’s increasingly vital public health problems due to an ageing population and environmental changes in lifestyle behaviour. The increase in the "overweight and obesity epidemic" may provide means for understanding why liver cancer is one of the few malignancies with rising incidence in developed countries over the last decades. With the increased liver cancer incidence and mortality in developed countries, the prevalence of overweight and obesity has also grown markedly over the past two decades. It has been estimated that approximately 1.9 billion adults are either overweight or obese
Research Aim: The main aim of this study is to develop and validate a machine-learning algorithm (XGBoost) to establish a 1-year liver cancer risk prediction model among Taiwanese adults with overweight or obesity using clinical data
Materials and Methods: We conducted a retrospective study on 4,149 patients from the MJ Taipei database between 2008 and 2017. We used stepwise logistic regression with both forward selection and backward elimination to select features. Akaike Information Criterion (AIC) was used as the selection method. The performance of the models was assessed using accuracy, sensitivity, specificity, AUROC, and the F1 score.
Results: The stepwise regression identified 13 predictors for liver cancer. There were 133 patients who developed liver cancer within one year of follow-up and 4,016 with non-liver cancer. Of those with liver cancer, 80% were men. The XGBoost had an AUROC of 90%, RF of 89%, and LR of 67%. In multivariate analysis, age, gender, BMI, total bilirubin, creatinine, body fat, lactate dehydrogenase, sGOT and waist circumference acted as independent predictors of liver cancer.
Conclusions: The XGBoost model had the optimal performance in predicting liver cancer compared to RF and LR. These findings represent an important step in the accurate prediction of liver cancer using ML techniques. If validated in prospective studies, the XGBoost model could serve as a useful approach for the prediction of liver cancer associated with overweight or obesity and may be potentially beneficial in implementing early preventive and therapeutic measures.
Research Aim: The main aim of this study is to develop and validate a machine-learning algorithm (XGBoost) to establish a 1-year liver cancer risk prediction model among Taiwanese adults with overweight or obesity using clinical data
Materials and Methods: We conducted a retrospective study on 4,149 patients from the MJ Taipei database between 2008 and 2017. We used stepwise logistic regression with both forward selection and backward elimination to select features. Akaike Information Criterion (AIC) was used as the selection method. The performance of the models was assessed using accuracy, sensitivity, specificity, AUROC, and the F1 score.
Results: The stepwise regression identified 13 predictors for liver cancer. There were 133 patients who developed liver cancer within one year of follow-up and 4,016 with non-liver cancer. Of those with liver cancer, 80% were men. The XGBoost had an AUROC of 90%, RF of 89%, and LR of 67%. In multivariate analysis, age, gender, BMI, total bilirubin, creatinine, body fat, lactate dehydrogenase, sGOT and waist circumference acted as independent predictors of liver cancer.
Conclusions: The XGBoost model had the optimal performance in predicting liver cancer compared to RF and LR. These findings represent an important step in the accurate prediction of liver cancer using ML techniques. If validated in prospective studies, the XGBoost model could serve as a useful approach for the prediction of liver cancer associated with overweight or obesity and may be potentially beneficial in implementing early preventive and therapeutic measures.