糖尿病视网膜病变的风险揭示与关键因素分析

目的：通过构建组合模型对糖尿病并发视网膜病变（DR）的患病风险进行预测，为DR的预防和诊断提供参考。方法：基于3000例糖尿病患者的生化检测数据，运用互信息作为评价标准筛选出与DR有关的特征因素，将其作为入模变量构建5种常见的模型，以准确率、精确率、召回率和AUC作为评价标准筛选出预测能力较优的3种模型，并运用Stacking方法构建组合模型。结果：通过互信息筛选出39个特征因素，发现随机森林模型、SVM模型以及Logistic回归模型这3种模型表现较优；构建的3种组合模型中，发现以SVM、Logistic为初级分类器，随机森林为次级分类器的组合模型预测能力最好，其AUC高达0.877。结论：组合模型相比单一模型具有更好的DR风险预测能力，更有助于DR的临床诊断。

Risk disclosure and key factors analysis of diabetic retinopathy

Abstract: Objective To construct combination models for the risk prediction of diabetic retinopathy (DR), thereby providing a reference for the prevention and diagnosis of DR. Methods Based on the biochemical test data of 3 000 diabetic patients, the characteristic factors related to DR which were screened out by taking mutual information as the evaluation criterion were input as the modeling variables to construct 5 common models. According to the evaluation criteria of accuracy rate, precision rate, recall rate and AUC, 3 models with the superior prediction ability were screened out for model combination by Stacking method. Results A total of 39 characteristic factors were screened out through mutual information. The prediction performances of random forest model, SVM model and Logistic regression model were better. Among 3 combination models constructed, it was found that the combination model with SVM and Logistic as the primary classifier and random forest as the secondary classifier had the optimal prediction ability, and its AUC was as high as 0.877. Conclusion The combination model is superior to the single model in predicting DR risk, and is helpful for the clinical diagnosis of DR.