Objective: To develop a prediction model for the risk of diabetic retinopathy (DR) in patients with newly diagnosed type 2 diabetes mellitus (T2DM). Methods: Patients with new diagnosis of T2DM recorded in Yinzhou Regional Health Information Platform between January 1, 2015 and December 31, 2022 were included in the study. The predictor variables were selected by using Lasso-Cox proportional hazards regression model. Cox proportional hazards regression models were used to establish the prediction model for the risk of DR. Bootstrap method (500 resamples) was used for internal validation, and the performance of the model was assessed by C-index, the receiver operating characteristic curve and area under the curve (AUC), and calibration curve. Results: The predictor variables included in the final model were age of T2DM onset, education level, fasting plasma glucose, glycated hemoglobin A1c, urinary albumin, estimated glomerular filtration rate, and history of lipid-lowering agent and angiotensin converting enzyme inhibitor uses. The C-index of the final model was 0.622, and the mean corrected C-index was 0.623 (95%CI: 0.607-0.634). The AUC values for predicting the risk of DR after 3, 5, and 7 years were 0.631, 0.620, and 0.624, respectively, with a high degree of overlap of the calibration curves with the ideal curves. Conclusion: In this study, a simple and practical risk prediction model for DR risk prediction was developed, which could be used as a reference for individualized DR screening and intervention in newly diagnosed T2DM patients.
目的： 开发新诊断2型糖尿病（T2DM）患者糖尿病视网膜病变（DR）发病风险的预测模型。 方法： 选取2015年1月1日至2022年12月31日宁波市鄞州区域健康信息平台中新诊断T2DM患者为研究对象。使用Lasso-Cox比例风险回归模型筛选预测变量，采用Cox比例风险回归模型构建DR发病风险预测模型。采取Bootstrap 500次重抽样的方法进行内部验证，并使用C指数、受试者工作特征曲线、曲线下面积（AUC）和校准曲线评估模型的性能。 结果： 最终模型纳入的预测变量包括T2DM发病年龄、文化程度、FPG、糖化血红蛋白、尿蛋白、估算肾小球滤过率、脂质调节剂和血管紧张素转化酶抑制剂用药史。最终模型C指数为0.622，校正后C指数均值为0.623（95%CI：0.607~0.634），预测DR 3、5、7年内发病风险的AUC值分别为0.631、0.620、0.624，校准曲线与理想曲线重合度较高。 结论： 本研究构建了简洁且实用的DR发病风险预测模型，为新诊断T2DM患者制定个体化DR筛查和干预方案提供参考。.