基于神经网络的上海市中老年人群胆囊结石风险预测模型研究  

Study on the risk prediction model of gallbladder stones in middle-aged and elderly populations in Shanghai based on neural network

作　　者：袁筱祺 朱乐兰 高玮 徐昕昕[3] YUAN Xiao-qi;ZHU Yue-lan;GAO Wei;XU Xin-xin(Clinical Medical School of Affiliated Shanghai the First Hospital,Nanjing Medical University,Shanghai 200080,China;Medical Department of Shanghai the First People's Hospital,Shanghai 20080,China;Graduate School of Heilongjiang University,Harbin Heilongjiang 150000,China)

机构地区：[1]南京医科大学附属上海一院临床医学院,上海200080 [2]上海市第一人民医院医务处,上海200080 [3]黑龙江大学研究生院,黑龙江哈尔滨150000

出　　处：《卫生软科学》2021年第12期28-33,共6页Soft Science of Health

基　　金：上海交通大学中国医院发展研究院医院管理建设项目(CHDI-2019-B-14)。

摘　　要：[目的]探讨上海市中老年胆囊结石的相关风险因素,提高风险预测的精度,为胆囊疾病的科学诊断、普适性提供更多的理论依据。[方法]选取上海市某三甲医院的中老年健康体检者3462条有效数据,根据彩色多普勒超声诊断结果分为是否患胆囊结石组,单因素分析筛选出6个重要风险因素,作为多层感知器神经网络及二元Logistic回归的分析变量,以构建胆囊结石预测模型。采用ROC曲线下面积(AUC)及模型敏感度、特异度及约登指数评估2种模型的预测效能。[结果]多层感知器神经网络风险预测模型的ROC曲线下面积(AUC)=0.937,约登指数最大值为0.755,灵敏度为94.91%,特异度为80.56%,均优于二元logistic回归预测模型(AUC=0.821,约登指数最大值为0.704,灵敏度为85.63%,特异度为84.78%)。[结论]对上海市中老年人胆囊结石的患病风险客观化预测中,基于多层感知器神经网络建立的风险预测模型较二元logistic回归模型具有更好的预测效能,为预防性治疗和干预提供理论依据。Objective This paper discusses the related risk factors of gallbladder stones in middle-aged and elderly people in Shanghai,improves the accuracy of risk prediction so as to provide more theoretical basis for scientific diagnosis and universality of gallbladder disease.Methods It selected 3462 valid data of middle-aged and elderly health check-ups in a top-three hospital in Shanghai,divided them into whether gallbladder stone group according to the color Doppler ultrasound diagnosis result,single-factor analysis screened out 6 important risk factors as multi-layer perceptron neural network and binary Logistic regression analysis variables,in order to build gallbladder stone prediction model.The predictive effectiveness of the two models was evaluated by ROC curve area(AUC)and model sensitivity,specificity and Jordon index.Results The ROC curve area(AUC)of the multi-layer perceptron neural network risk prediction model was 0.937,the maximum value of the Jordon index was 0.755,the sensitivity was 94.91%,and the specificity was 80.56%,both of them were better than the binary logistic regression prediction model(AUC=0.821,the maximum value of the Jordon index was 0.704,the sensitivity was 85.63%,the specificity was 84.78%).Conclusions In the objective prediction of disease risk of gallbladder stones in middle-aged and elderly people in Shanghai,the risk prediction model based on multi-layer perceptron neural network has better predictive effectiveness than the binary logistic regression model,which provides theoretical basis for preventive treatment and intervention.

关 键 词：神经网络 胆囊结石 风险预测模型 

分 类 号：R194[医药卫生—卫生事业管理]