神经网络模型在儿童ALL患病风险预测中的应用  

Back Propagation Neural Network model applied in the predication of the childhood acute lymphoblastic leukemia

作　　者：张丹丹[1] 赵亮[1] 曾强[1] 张丽 符刚[1] 张磊[1] 王睿[1] 冯宝佳[1] 侯常春[1] ZHANG Dandan;ZHAO Liang;ZENG Qiang;ZHANG Li;FU Gang;ZHANG Lei;WANG Rui;FENG Baojia;HOU Changchun(Department of Environment and Health,Tianjin Centers for Disease Control and Prevention,Tianjin 300011,China)

机构地区：[1]天津市疾病预防控制中心环境与健康室,天津300011 [2]中国医学科学院血液病医院儿科

出　　处：《公共卫生与预防医学》2018年第4期76-79,共4页Journal of Public Health and Preventive Medicine

基　　金：天津市卫生计生委科技基金重点项目(2014KR12);天津市疾病预防控制中心科技基金项目(CDCKY1406)

摘　　要：目的构建儿童急性淋巴细胞白血病(ALL)患病风险预测模型,探索其危险因素,为儿童ALL的预防和诊断提供参考依据。方法利用Logistic回归模型进行单因素分析,筛选出的变量作为预测指标构建BP神经网络模型和Logistic回归模型,并对2种模型的预测结果进行比较。结果 2种模型均显示影响儿童ALL的主要环境危险因素为儿童不爱吃蔬菜和水果以及母亲孕期服用药物史。BP神经网络模型的灵敏度、特异度和AUC分别为83.09%、80.88%和0.820,均高于Logistic回归模型(灵敏度=74.26%,特异度=76.47%,AUC=0.754)。结论 BP神经网络模型可以快速、有效的从众多数据中挖掘出影响儿童ALL患病的主要因素,并预测其患病风险,在流行病学研究中具有较高的应用价值。Objective To conduct predictive models for childhood acute lymphoblastic leukemia and explore the risk factors to provide references for the prevention and diagnosis of the childhood acute lymphoblastic leukemia. Methods The logistic regression model was used for single factor analysis, the selected variables were used as predictors to construct BP neural network model and logistic regression model. The predictive results of the two models were also compared. Results Both models showed that the main environmental risk factors affecting childhood ALL were children who did not like vegetables and fruits and the history of maternal drug use during pregnancy. The sensitivity, specificity, and AUC of BP neural network model were 83.09%, 80. 88% and 0. 820 respectively, all higher than the results of logistic regression model： sensitivity, specificity, and AUC were 74. 26% , 76.47% and 0. 754 respectively）. Conclusion BP neural network model could excavate the main factors influencing the childhood ALL from the numerous data quickly and efficiently, and predict its risk, thus had higher application value in epidemiological studies.

关 键 词：儿童 急性淋巴细胞白血病 BP神经网络 LOGISTIC回归 

分 类 号：R745[医药卫生—神经病学与精神病学]