·糖尿病临床流行病学·

　　【摘要】目的 采用北京有代表性的社区自然人群的简单易得的健康信息，建立简便易行的针对未诊断糖尿病和糖尿病前期的预警模型和筛查工具。方法 选取北京市顺义区南法信镇35岁以上既往没有糖尿病病史的常住居民，对空腹血糖大于等于5.6mmol/l者行口服葡萄糖耐量试验(OGTT)。220例2型糖尿病(DM)、173例糖尿病前期(PDM)和1868名血糖正常人群参加此项研究。采用决策树方法分别建立DM+PDM及单纯DM两个预测模型。结果 预测模型均包括3个变量，分别是年龄、体重指数(BMI)和是否患高血压。DM预测模型的敏感性、特异性、阳性预测值、阴性预测值和ROC曲线下面积(AUC)分别为0.759 (95%CI：0.697-0.814)、0.657(95%CI：0.600-0.710)、0.618(95%CI：0.558-0.677)、0.788(95%CI：0.732-0.837)、0.708(95%CI：0.663-0.753)。DM+PDM预测模型的上述指标分别为0.756(95%CI：0.710-0.797)、0.603(95%CI：0.553-0.651)、0.651(95%CI：0.606-0.695)、0.715(95%CI：0.664-0.763)、0.679(95%CI：0.642-0.717)。结论 本研究根据普通人最容易得到的健康资料通过决策树模型建立并验证了北京社区糖尿病前期及糖尿病的风险评估和预警模型，具有简单易懂、方便可行的优点，推荐作为糖尿病筛查的一线方法。

　　【关键词】2型糖尿病;糖尿病前期;预测模型;筛查;决策树

The study of a simple prediction model for detecting diabetes and pre-diabetes in the community natural population   Xin Zhong, Hua Lin, Yuan Jing, Ma Ya-Hong, Zhao Lei, Yang Jin-Kui. Department of Endocrinology, Beijing Tongren Hospital, Capital Medical University, Beijing 100730, China.

Corresponding author：Yang Jin-Kui，Email：jinkui.yang@gmail.com

　　【Abstract】objective To develop and evaluate a simple prediction model, using data that are collected in Beijing representative community natural population. Methods The study subjects were selected from the individuals aged 35 years or older and without previously known diabetes in the Nanfaxin, a satellite town of Beijing. All subjects were invited to attend a health questionnaire, health examination and laboratory examination, and the subjects with FPG ≧ 5.6 mmol/L had a 75-g oral glucose tolerance. A total of 220 Type 2 diabetes (DM)and 173 Prediabetes (PDM)and 1,868 normal blood glucose subjects took part in this study. Decision tree analysis were used to build two models ：DM and DM+PDM. Results There are three risk factors in the models: age, BMI and whether suffering from hypertension. The sensitivity, specificity, positive predictive value, negative predictive value and receiver operating characteristic area(AUC)for detecting DM were 0.759 (95%CI：0.697-0.814), 0.657(95%CI：0.600-0.710), 0.618(95%CI：0.558-0.677), 0.788(95%CI：0.732-0.837)and 0.708(95%CI：0.663-0.753), respectively. For DM+PDM, the results were 0.756(95%CI：0.710-0.797), 0.603(95%CI：0.553-0.651), 0.651(95%CI：0.606-0.695), 0.715(95%CI：0.664-0.763), 0.679(95%CI：0.642-0.717), respectively. Conclusion The classification tree model is a simple and accurate tool to identify those at high risk of DM and PDM. We recommend it as a first-line method of screening diabetes patients.

　　【Keywords】 Type 2 diabetes; Prediabetes; prediction model; Screening; Decision tree analysis