基于人工神经网络模型的多个慢性病主要危险因素筛查研究  

Identification of Major Risk Factors for Multiple Chronic Diseases Based on Artificial Neural Network

作　　者：曹文君[1] 徐勇勇[2] 谭志军[2] 王庸晋[1] 

机构地区：[1]长治医学院心血管病研究所,山西省长治市046000 [2]第四军医大学军事预防医学院卫生统计学教研室

出　　处：《中国全科医学》2015年第25期3050-3053,3058,共5页Chinese General Practice

基　　金：国家自然科学基金资助项目(81302518)

摘　　要：目的探讨基于人工神经网络(ANN)模型的多个慢性病主要危险因素筛查。方法选取2008年1月—2010年12月参加北京某健康管理中心体检的年龄45岁及以上人群6 938例。采用逐步回归和遗传算法相结合的方法确定ANN输入变量,尝试构建高血压、糖尿病、冠心病及慢性病患者预测模型,并采用受试者工作特征(ROC)曲线评价预测模型的准确性。结果 6 938例体检人群中高血压患者1 665例(24.0%),糖尿病患者609例(8.8%),冠心病患者443例(6.4%)。年龄、体质指数、胸围、腰臀比、总胆固醇、高密度脂蛋白胆固醇、尿酸、性别、尿糖、高血压家族史、糖尿病家族史、心血管疾病家族史是慢性病患者的主要危险因素,其中以年龄对慢性病患病的影响最大,作用效应为25.3%。高血压、糖尿病、冠心病及慢性病ANN预测模型ROC曲线下面积分别为0.80、0.87、0.81、0.78,预测高血压、糖尿病、冠心病、患任一慢性病的准确性分别为75.1%、91.2%、93.7%、75.2%。结论利用ANN模型筛选出多个慢性病主要危险因素,可为慢性病的有效预防提供科学依据。Objective To discuss the identification of major risk factors for multiple chronic diseases based on artificial neural network（ANN）. Methods We enrolled 6 938 subjects aged 45 or older than 45 who received physical examination in a health management center in Beijing from January 2008 to December 2010. Stepwise regression combined with genetic algorithm was used to determine the input variables of artificial neural network（ ANN）. We tried to build the prediction models for hypertension,diabetes mellitus,coronary heart disease and chronic diseases and then evaluated the accuracy of these models by receiver operator characteristic（ ROC）curve. Results Among 6 938 subjects,1 665 （ 24. 0% ）had hypertension,609 （8. 8% ） had diabetes mellitus, and 443 （ 6. 4% ） had coronary heart disease. Age, body mass index （ BMI ）, chest circumference,waist - hip ratio,total cholesterol,HDL - C,uric acid,gender,urine sugar,family history of hypertension, family history of diabetes mellitus and family history of cardiovascular disease are major risk factors for chronic diseases,among which age had the greatest influence on chronic diseases with an effect rate of 25. 3%. Moreover,the areas under ROC curves of ANN prediction models for blood pressure,diabetes,coronary heart disease and chronic disease were 0. 80,0. 87,0. 81 and 0. 78 respectively. The accuracy rates in the prediction for hypertension,diabetes,coronary disease and chronic disease were 75. 1% ,91. 2% ,93. 7% and 75. 2%. Conclusion Main risk factors for multiple chronic diseases could be identified by ANN model,which could provide scientific references for effective prevention of chronic diseases.

关 键 词：人工神经网络 慢性病 危险因素 

分 类 号：R195.4[医药卫生—卫生统计学]