【作者】 米生权；

【导师】 赵文华； 李剑虹； 胡楠；

【作者基本信息】 中国疾病预防控制中心 ， 流行病与卫生统计学， 2011， 博士

【摘要】 糖尿病已经成为影响我国人民健康的主要慢病之一,近几十年来,随着生活方式的改变,患病人数和糖尿病前期人数进一步增加,由糖尿病导致的并发症、致残、死亡以及随之而来的疾病负担将更加严重。因此,如何识别糖尿病高危人群,及时进行生活方式干预,将成为糖尿病防治的主要任务。糖尿病个体发病预测模型能有效识别糖尿病高危人群、是选择干预方案的必要工具,可为健康管理、疾病预防决策以及干预方案效果评价提供依据,已经被广泛认可。目前,针对不同人种,国外已经建立了不同人群的糖尿病个体发病预测模型,而我国,由于缺乏队列人群数据,尚未建立有效的糖尿病个体发病风险预测模型。因此建立适合中国成人糖尿病个体水平的发病预测模型,显得十分必要。论文系统回顾了中国人群糖尿病主要常见危险因素的单因素未调整OR或RR值,以Meta分析方法定量研究各种危险因素与糖尿病的关系；应用国际上认可的合成分析方法在2002年中国居民营养与健康状况调查数据上,消除各危险因素之间的共线性,建立中国成人糖尿病个体发病风险预测模型；以北京市某大型企业队列数据进行了验证,利用受试者工作曲线(ROC)分析方法,确定模型预测的最佳切点及其相应的灵敏度和特异度；模型应用于北京和浙江两地人群慢病干预项目,对人群进行风险评估,按照患病风险进行干预,探讨模型应用价值。经过Meta分析,共获得了14个DM危险因素的OR值和1个DM危险因素的RR值,最终确定10个危险因素,分别为年龄、体质指数(BMI)、DM家族史、高血压史、总胆固醇(TC)、甘油三酯(TG)、低密度脂蛋白(LDL-C)、血脂异常、腰围(WC)和高浓度空腹血糖(FBG)；确定2个保护因素,分别为教育程度和血清高密度脂蛋白胆固醇脂(HDL-C)含量；与糖尿病发病关系不密切的有吸烟、饮酒和性别。采用合成分析方法,利用2002年中国居民营养与健康状况调查数据,建立了20-70岁中国成人未来10年糖尿病发病风险的简单和复杂预测模型,其中,简单预测模型只需基本信息和人体测量数据就能自我完成预测,复杂模型需要空腹血糖和血脂指标。经队列人群数据验证结果显示,简单模型ROC曲线下面积(AUC)为0.68,最佳切点取13.4%时,灵敏度和特异度分别为57.9%和71.2%；复杂模型ROC曲线下面积达到了0.81,在最佳切点处,灵敏度和特异度分别74.7%和80.0%,预测效能台湾危险评分模型相同。如果简单模型和复杂模型联合使用,对简单模型预测风险≥4%的人群进行空腹血糖和血脂检测,再用复杂模型预测,AUC可达0.83,最佳切点处灵敏度和特异度分别为80.0%和77.3%,约登指数(YDI)达到了0.573,既优于台湾模型(YDI为0.518)又优于简单模型和台湾模型的联合使用效果(YDI为0.517)。基于模型开发的风险评估软件,应用于北京和浙江两地人群重要慢病干预项目7171名调查对象,为每一个研究对象计算了未来患病风险,并提供报告。考虑到很多对象在应用模型评估之前,并不知道自己已经患有糖尿病,还验证了模型对未诊断糖尿病人群的识别,结果显示,无论简单模型还是复杂模型都有一定的识别作用。评估对象在不做血液生化检测的情况下,根据自己年龄、BMI、腰围、糖尿病家族史和高血压史以简单模型完成预测,AUC为0.707,按照发病风险预测切点取4%时,灵敏度可达91.4%；在不进行糖耐量检测的情况下,只进行空腹血糖检测,以复杂模型预测获得的AUC可达0.953,最佳切点28.1%处,模型预测灵敏度和特异度分别93.09%和90.85%,YDI为0.839,具有很好的临床应用价值。对于非糖尿病人群,根据个体未来10年患糖尿病的风险高低,把人群分为高危人群和低危人群,对高危人群进行个体化干预,并结合评估对象自身特点,改变可控因素,制定了个体量化干预措施,获得了评估对象和基层疾控部门的认可。本研究建立了中国成人个体糖尿病发病预测模型,能够有效预测个体未来10年糖尿病患病风险,对未诊断糖尿病也具有很好的识别作用,可用于糖尿病个体化干预方案的制定和卫生资源配置等。这不仅是中国人群首次建立的糖尿病个体未来10年发病风险预测模型,也是合成分析方法首次应用于建立糖尿病模型,所建立的模型适合社区医院、公共卫生部门常规健康管理和疾病控制,有利于自我健康促进、整体卫生资源配置和疾病负担的预计等相关卫生政策的制定。更多还原

【Abstract】 The most updated research showed that prevalence of diabetes among Chinese adults is 9.7% hitting the top one across the world. The incidence of pre-diabetes is 15.5 per 100 person-years. It is estimated the number of people living with diabetes will keep increasing if there is no appropriate interventions. The epidemic will be more serious as diabetes result in complex complications, high morbidity and mortality rate, as well as heave disease burdens. Therefore, it is significant to identify high risk population more likely to develop diabetes and then to conduct life-style-related interventions at early stage. It is believed that incidence prediction model of diabetes individuals is effective to identify high risk population and then to propose appropriate interventive plans. The prediction model has been widely used to provide key information for disease prevention, policy making, and to evaluate the intervention outcomes in many countries. However, due to the lackage of effective population-based cohort study, China has not employ this model so far. Given the worsen epidemic of diabetes, it is of great importance to establish a prediction model which can fit into Chinese context at present.We systematically reviewed studies with unadjusted OR or RR values to address major high risk factors associated to diabetes and applied Meta analysis to measure the association. Synthesis analysis was applied to eliminate the collinearity of each risk factor and establish the individual level prediction model based on the data collected from National Nutrition and Health Status Survey conducted in 2002. To verify the prediction models, we used it to analyze a cohort study of a large enterprise in Beijing. We applied ROC approach of the interviewees to define the best cut point of the model with its sensitivity and specificity. We also applied the model on diabetes survey at Beijing and Zhejiang province.Through Meta analysis,10 risk factors, such as age, BMI, family history of diabetes, history of hypertension, TC, TG, LDL-C, hyperlipidemia, WC, and high concentration of fasting blood glucose and 2 protective factors, such as education level and concentration of serum HDL-C, were identified from 14 factors with OR values and 1 factor with RR value....We also found smoking, alcohol consumption and gender were weakly associated with diabetes.2 prediction models (Concise & Complex) were established based on the data of Nutrition and Health Status Survey (2002) to predict the incidence of diabetes in next 10 years among Chinese adults at the age of 20-70 years old. The concise model can be operated by the target population themselves without any professional assistant because only basic information and anthropometric data are needed in the model. Whereas, to apply the complex model, data of fasting blood glucose and lipid concentration are required. We verified the 2 models by applying them to analyzing data collected from a cohort study. It was showed the area under the curve of ROC was 0.68 with the sensitivity 57.9% and specificity 71.2% at the optimal cut-point 13.4% in the concise model. In the complex model, it was found the AUC was 0.81 and sensitivity and specificity was 74.7% and 80.0% respectively at the optimal cut-point. The prediction efficiency of complex model was similar to that of Taiwan’s prediction model.If we used consice prediction model first and test fasting blood glucose and lipid of the individual whose diabetes risk in 10 years above 4%, than evaluated again with complex one, we will get a better ROC, which AUC would be 0.83 and the sensitivity and specificity was 80.0% and 77.3% respectively at the optimal cut-off (Youden index was 0.573). The prediction efficiency was better than that of the Taiwan’s model (0.518) and also better than that of the combination of concise model and Taiwan’s model (0.517).We calculated the chronic disease risk for each individual in Zhejiang and Beijing based on the software derived from the models. Considering many participants did not know they have diabetes before the appliaction of the model, we assessed the identification of the model to undiagnosed diabetes population and the result showed that the concise prediction model have the effect of prediction. The concise model could be applied based on individual’s age, BMI, WC, family history of diabetes, and history of hypertension without biochemical testing. The area under the curve of ROC was 0.707. The sensitivity was 91.4% when the cutpoint was 4%. For population without diabetes, we could divide them into high risk and low risk group based on the individual risk of diabetes in 10 years and administer health management for those at high risk. Individual intervention measures can be made based on individual’s own features and changed modifiable factors through the risk assessment, which was widely recognized and accepted by the assessment individuals and county level departments of disease control and prevention.A diabetes incidence prediction model for Chinese adults is established in this study. The model could be used to predict the risk of diabetes incidence in the next 10 years, to identify undiagnosed diabetes, to apply in diabetes health management, to develop individual intervention plans and to adjust health resources. It is the first model to predict the risk of diabetes of the next 10 years in China and it is the first time for the synthesis approach to be used in this field. The model is appropriate to be applied for health management and disease control in community hospitals and public health departments, for self health promotion, allocation of health resources, and for decision making on burden of disease related health policies.更多还原