基于随机森林法的2型糖尿病合并非酒精性脂肪肝预测模型

A Lightweight Model to Predict Nonalcoholic Fatty Liver Disease in Type 2 Diabetes via Random Forest Algorithm

目的:非酒精性脂肪性肝病(Nonalcoholic fatty liver disease,NAFLD)增加了心血管疾病死亡的风险,并可导致肝细胞癌。研究拟通过挖掘多个常规指标(特征)对NAFLD的影响,预测2型糖尿病患者罹患NAFLD的概率。方法:从上海市第六人民医院临床信息系统采集共26个维度的数据,数据挖掘方法采用集成学习算法中的随机森林法。同时,基于医生的经验知识,逐步缩小特征数量至9个维度。结果:在不同维度的模型中,最小预测精度为81.5%,而最大精度为83.6%,均优于其他类似模型。当有9个特征时,预测精度为83.2%。结论:随机森林法在2型糖尿病中预测并存NAFLD具有实际应用价值,甘油三酯、低密度脂蛋白、空腹胰岛素是最重要的预测指标。

Objective: Nonalcoholic fatty liver disease(NAFLD) increases the mortality of liver and cardiovascular disease, and is very likely to cause hepatocellular carcinoma. The aim of this study is to mine the impacts of multiple routine indicators(features) on NAFLD in type 2 diabetes mellitus(T2DM), and predict the occurrence rate of NAFLD in T2DM patients without complicated tests. Methods: A total data of 26 dimensions were obtained from the Hospital Information System. We apply an ensemble learning algorithm called Random Forest. Furthermore, based on physician’s empirical knowledge, we try to shrink the features attribute set, aiming to use less features while maintaining the prediction accuracy. Results: We successfully predict the test set’s classification in the different dimension models, the minimum prediction accuracy at 81.5%, and the maximum precision at 83.6%. It is superior to other similar models. When there were 9 features, the prediction result was 83.2%. Conclusion: Random forest algorithm has a practical use in predicting NAFLD in type 2 diabetes mellitus, and triglyceride, low density lipoprotein and fasting insulin are the most important predictors.