糖尿病视网膜病变预测模型的构建与验证

Construction and validation of prediction model for diabetic retinopathy

目的:分析筛选糖尿病患者发生视网膜病变的影响因素,建立列线图预测模型并进行验证分析。方法:选取2013-01/2021-01国家人口健康科学数据中心数据仓储PHDA的《糖尿病并发症预警数据集》中的1252例患者为研究对象,通过随机化将研究对象划分为建模组(n=941)和验证组(n=311)。采用单因素分析、LASSO回归和Logistic回归分析筛选影响糖尿病患者并发视网膜病变的影响因素,构建列线图预测模型,采用ROC曲线、校正曲线和Hosmer-Lemeshow检验对模型进行验证评估,采用临床决策曲线(DCA)评估模型的临床收益情况。结果:年龄、高血压、肾病、收缩压(SBP)、糖化血红蛋白(HbA1c)、高密度脂蛋白胆固醇(HDL-C)、血尿素(BU)是糖尿病患者发生视网膜病变的影响因素。建模组AUC值为0.792(95%CI:0.763-0.821),验证组AUC值为0.769(95%CI:0.716-0.822);Hosmer-Lemeshow检验和校准曲线提示模型拟合度较高(建模组:χ^(2)=14.520,P=0.069;验证组:χ^(2)=14.400,P=0.072),DCA曲线结果显示,建模组和验证组的阈值概率分别为0.09-0.89和0.07-0.84,临床净收益较高。结论:本研究构建包含年龄、高血压、肾病、SBP、HbA1c、HDL-C、BU的风险预测模型,该模型有较高的区分度和一致性,可用于预测糖尿病患者发生视网膜病变的风险。

AIM:To analyze and screen influencing factors of diabetic patients complicated with retinopathy,and establish and validate prediction model of nomogram.METHODS:A total of 1252 patients from the Diabetes Complications Early Warning Dataset of the National Population Health Data Archive(PHDA)between January 2013 to January 2021 were selected and randomly divided into a modeling group(n=941)and a validation group(n=311).Univariate analysis,LASSO regression and Logistic regression analysis were used to screen out the influencing factors of diabetic retinopathy,and a nomogram prediction model was established.The receiver operating characteristic curve,Hosmer-Lemeshow test and calibration curve were used to evaluate the model.The clinical benefit was evaluated by the decision curve analysis(DCA).RESULTS:Age,hypertension,nephropathy,systolic blood pressure(SBP),glycated hemoglobin(HbA1c),high-density lipoprotein cholesterol(HDL-C),and blood urea(BU)were the influencing factors of diabetic retinopathy.The area under the curve(AUC)of the modeling group was 0.792(95%CI:0.763-0.821),and the AUC of the validation group was 0.769(95%CI:0.716-0.822).The Hosmer-Lemeshow goodness of fit test and calibration curve suggested that the theoretical value of the model was in good agreement(modeling group:χ^(2)=14.520,P=0.069;validation group:χ^(2)=14.400,P=0.072).The DCA results showed that the threshold probabilities range was 0.09-0.89 for modeling group and 0.07-0.84 for the validation group,which suggested the clinical net benefit was higher.CONCLUSION:This study constructed a risk prediction model including age,hypertension,nephropathy,SBP,HbA1c,HDL-C,and BU.The model has a high discrimination and consistency,and can be used to predict the risk of diabetic retinopathy in patients with diabetes.