社区2型糖尿病视网膜病变风险预测模型的构建与验证

背景　糖尿病视网膜病变（DR）发病率高，危害性大，是导致成人失明的主要原因，但社区实施眼底筛查较为困难，而DR风险预测模型可帮助全科医生早期识别DR高危人群。目的　探讨社区门诊就诊的2型糖尿病患者发生DR的危险因素，构建并验证DR风险预测模型，进一步提高DR患者筛查效率。方法　收集2018年6月-2019年6月于方庄社区卫生服务中心门诊就诊的421例2型糖尿病患者，采用随机数字表法分为模型组336例，检验组85例。收集患者的一般资料，血压以及空腹血糖（FBG）、餐后2 h血糖、糖化血红蛋白（HbA1c）、血肌酐、尿素氮、总胆固醇（TC）、三酰甘油（TG）、低密度脂蛋白胆固醇（LDL-C）及尿蛋白，均行非散瞳眼底彩色照相（NMFCS）。模型组采用单因素Logistic回归分析计算出发生DR的相关因素，得出的相关因素用多因素Logistic回归分析进一步探讨，在此基础上构建DR风险预测模型，并由检验组评估DR风险预测模型的可行性。结果　在421例2型糖尿病患者中DR患者共87例，其中模型组69例，检验组18例。多因素Logistic回归分析结果显示，DR的危险因素有病程（β=0.196，OR=1.217，P<0.001），收缩压（SBP）（β=0.028，OR=1.028，P=0.038），FBG（β=0.409，OR=1.506，P=0.003），HbA1c（β=0.594，OR=1.811，P=0.001），LDL-C（β=0.360，OR=1.434，P=0.038）。据此，构建的DR风险预测模型为Y=1/〔1+e-（0.196X1+0.028X2+0.409X3+0.594X4+0.360X5-16.482）〕，其中，Y指DR发生概率，X1指病程，X2指SBP，X3指FBG，X4指HbA1c，X5指LDL-C。DR风险预测模型预测模型组发生DR的ROC曲线下面积是0.884，诊断临界值是0.192。DR风险预测模型预测检验组发生DR的ROC曲线下面积是0.803，灵敏度为72.2%，特异度为79.1%。结论　糖尿病患者的病程、SBP、FBG、HbA1c、LDL-C与DR显著相关，DR风险预测模型对DR有一定的预测价值。

Establishment and Verification of a Risk Prediction Model of Diabetic Retinopathy in Patients with Type 2 Diabetes in the Community

Background　Diabetic retinopathy（DR） has a high incidence and is a leading cause of blindness in adults.Although it is difficult to perform fundus screening in a wide range of population in communities，the risk prediction model of DR can help general practitioners identify high risk population of DR at an early stage.Therefore，carrying out the research on the risk prediction model for DR is particularly important.Objective　To explore the risk factors of DR in type 2 diabetic outpatients in community settings，and use them to establish and verify a risk prediction model for DR，further improving the screening efficiency of DR.Methods　From June 2018 to June 2019，421 type 2 diabetic outpatients from Fangzhuang Community Health Service Center were invited and were divided into model group（336 cases） and verification group（85 cases） by a table of random numbers.Demographic information，blood pressure，fasting blood glucose（FBG），2-h postprandial blood glucose，glycosylated hemoglobin（HbA1c），serum creatinine，urea nitrogen，total cholesterol（TC），triacylglycerol（TG），low-density lipoprotein cholesterol（LDL-C），urine protein，and results of non-mydriatic fundus color photography（NMFCS） of the patients were collected.Univariate Logistic regression analysis was used to identify DR-related factors in model group，and the related factors obtained were discussed with multivariate Logistic regression analysis to explore the risk factors of DR based on which the DR risk prediction model was constructed，and the feasibility of this prediction model was evaluated in the verification group.Results　Altogether，87 patients were found with DR，including 69 in the model group and 18 in the verification group.Multivariate Logistic regression analysis showed that years of having T2DM（β=0.196，OR=1.217，P<0.001），systolic blood pressure（β=0.028，OR=1.028，P=0.038），FBG（β=0.409，OR=1.506，P=0.003），HbA1c（β=0.594，OR=1.811，P=0.001） and LDL-C （β=0.360，OR=1.434，P=0.038） were risk factors of DR in type 2 diabates patients.Therefore，the DR risk prediction model was Y=1/〔1+e-（0.196X1+0.028X2+0.409X3+0.594X4+0.360 x5-16.482）〕，Y refers to the probability of DR，X1 to years of having T2DM，X2 to systolic blood pressure，X3 to FBG，X4 to HbA1c，and X5 to LDL-C.The area under the ROC curve of DR in the model group was 0.884，and the cut-off point was 0.192.The area under the ROC curve was 0.803，the sensitivity was 72.2%，and the specificity was 79.1% in type 2 diabates patients occurred DR in verification group.Conclusion　Years of having T2DM，systolic blood pressure，FBG，HbA1c and LDL-C were significantly correlated with DR.DR risk prediction model has certain prediction value for DR.