BP神经网络在结核疑似病患者就诊延迟影响因素分析中的应用探讨

BPNN Applied in the Assessment of Factors Associated with the Seeking-care Behavior of TB Suspects

目的在重庆市结核病疑似患者就医行为的影响因素调查的基础上,运用BP神经网络进行建模分析,并与传统统计学模型logistic模型的结果对比,来探讨BP神经网络在结核病防治领域应用前景,挖掘其在流行病疾病的相关因素分析工作中的应用潜力。方法通过MATLAB中的神经网络工具箱,建立BP神经网络,并计算出各个因素的平均影响值(meani mpact value,MI V),并按照其绝对值大小列出各因素对于应变量影响的相对重要性顺位,从而达到分析就诊影响因素的目的,与多因素logistic模型进行比较,最终得出影响结核病疑似患者就医行为的危险因素。结果影响患者就医行为的主要因素有:性别、婚姻状况、体力工作者、医疗保险、结核病认知得分、医疗卫生服务的可及性、患者既往病后采取自我医疗措施、患者既往病后看医生、本次病后采取自我医疗措施、本次病后不采取任何措施、首发症状、退休。结论BPNN同样能够用于结核病疑似患者就诊影响因素研究。BPNN模型分析对变量不作任何要求,能较好地处理因素间复杂的关系,具有其独特的优势。

Objective Based on the survey of seeking-care behavior of TB suspects in Chongqing, BPNN and logistic model are employed in the affecting factors analysis, and the two results are compared to explore the application prospective of BPNN in TB prevention and cure field and then try to dig out the potential ability of BPNN in the factors analysis in epidemiology. Methods Using the NNtools Box of MATLAB, to establish BPNN and calculate the Mean Impact Value （MIV） of each factor, and listed by its absolute MIV, thus the relative importance order for all the factors is clear and then compare with logistic Model, to get a more complete and accurate results. Results The main factors affecting the patients＇ behavior are gender, marital status, medical insurance and so on. Conclusion BPNN is competent for the affecting factor study in the seeking-care course of TB suspects. Compared with traditional statistical models, BPNN has its unique advantage. BPNN model has never asked independent variables to meet some conditions, and it does well in dealing with the complicated rela- tions between factors.