基于广义回归神经网络的老年龋病高危人群预警模型研究

Senile dental caries is a common chronic disease, which seriously affects the general health of the aging population. As the country which has the most aging population in the world, China’s burden of the senile dental caries is prominent and its prevention work tends to be imminent, with relative lack of dental institutions and lower universality of health insurance service. With the implementation of the new medical reform in China, the management of chronic diseases, like senile dental caries, is mainly located in basic medical institutions,such as community services. However, lack of professional experiences is still regarded as a main issue of the basic-level medical prevention. As a result, we conduct the study of high risk factors for senile dental caries, which use mathematical modeling of biological early warning model to build its high-risk groups for screening, early diagnosis and treatment of senile dental caries. Currently, the use of generalized regression neural network (GRNN) to establish early warning model of high risk population of senile dental caries has not been involved in home and aboard. Based on the previous studies, The study designs to be clarified the high risk factors of senile dental caries by oral examination, face-to-face questionnaires and caries activity test (CAT), such as UWSFR, SWSFR and D-SM, which makes advantage of inter-disciplinary. Meanwhile, taking advantage of the powerful predictive functions of GRNN, the study designs to build an early warning model of senile dental caries regarding risk factors as independent variables, in order to provide a scientific decisional evidence for the formulation preventive measures of senile dental caries.

老年龋病是严重影响老年人全身健康的常见慢性病。我国是世界老龄人口最多的国家，因口腔机构相对缺乏，医疗保险服务未普及，使老年龋的疾病负担更为突出，其防治工作迫在眉睫。随着新医改的实施，龋病等慢性病的管理工作主要定位于社区服务中心等基层医疗机构，缺乏专业经验是困扰基层防治工作的主要问题。因此，研究老年龋危险因素，利用生物数学建模构建预警模型对于其高危人群的筛查、疾病的早期诊断及治疗都有着积极作用。目前，依据老年龋危险因素，利用广义回归神经网络(GRNN)建立其高危人群的预警模型研究在国内外尚无人涉及。本研究充分发挥学科交叉优势，在前期工作基础上，通过口腔检查、UWSFR、Dentocult SM等龋活跃性试验、问卷调查等方法，明确老年人患龋的危险因素；同时利用GRNN强大的预测功能，研究开发以这些危险因素为自变量的基于GRNN的老年龋高危人群预警模型，从而为老年龋的预测及防治措施提供依据。

老年龋病是严重影响老年人全身健康的常见慢性病。我国是世界老龄人口最多的国家，因口腔机构相对缺乏，医疗保险服务未普及，使老年龋的疾病负担更为突出，其防治工作迫在眉睫。随着新医改的实施，龋病等慢性病的管理工作主要定位于社区服务中心等基层医疗机构，缺乏专业经验是困扰基层防治工作的主要问题。因此，研究老年龋危险因素，利用生物数学建模构建预警模型对于其高危人群的筛查、疾病的早期诊断及治疗都有着积极作用。.本课题依据老年龋危险因素，利用广义回归神经网络(GRNN)建立其高危人群的预警模型研究。发挥学科交叉优势，在前期工作基础上，通过口腔检查、 检测宿主UWSFR、SWSFR等龋活跃性试验、问卷调查等一系列方法，并通过卡方检验、Ridit检验等方法进一步明确老年人患龋的危险因素，建立了以危险因素为自变量的Logisitic回归模型。采用Matlab软件编程建立GRNN广义回归神经网络预测模型，最终确立最优光滑因子为0.7。非条件多因素Logistic回归模型的ROC曲线下面积为0.590，95%可信区间为（0.508，0.672），P值为0.028；GRNN广义回归神经网络模型的ROC曲线下面积为0.626，95%可信区间为（0.544，0.708），P值为0.002。同时利用GRNN强大的预测功能，开发出以这些危险因素为自变量的基于GRNN的老年龋高危人群预警模型。将两种模型对比分析，发现GRNN预警模型具有更高的一致率、特异度和敏感度，具有一定的优势及实用性，为指导老年龋高危人群的早期筛查和防治工作提供技术支持。本研究通过三年的研究工作，在研究目标完成的基础上，将研究结果进行了一部分的外推研究，将龋风险评估初步应用于0-6岁婴幼儿及学龄前儿童的龋病风险管理研究工作中，目前研究工作进展良好。