Development of a Glycated Albumin-Based Algorithm to Evaluate Diabetic Retinopathy in Adults with Type 2 Diabetes: A Cross-Sectional Study at a Hospital-Affiliated Physical Examination Center

Jing Bai,1 Yu Liu,1 Fan Wang,2 Shan Yu,1 Yungang Pu,1 Baobin Luo,1 Qingchen Meng,1 Mingze Jin,1 Dongning Chen,2 Xiangyi Liu1 1Department of Clinical Laboratory, Beijing Tongren Hospital, Capital Medical University, Beijing, 100730, People’s Republic of China; 2Beijing Tongren Hospital Health Management Center, Capital Medical University, Beijing, 100730, People’s Republic of ChinaCorrespondence: : Xiangyi Liu, Department of Clinical Laboratory, Beijing Tongren Hospital, Capital Medical University, No. 1, Dongjiaominxiang Street, Dongcheng District, Beijing, 100730, People’s Republic of China, Tel +8601058268695, Email [email protected] Dongning Chen, Beijing Tongren Hospital Health Management Center, Capital Medical University, No. 1, Dongjiaominxiang Street, Dongcheng District, Beijing, 100730, People’s Republic of China, Tel +86010585363574, Email [email protected]: Early diagnosis and management of diabetic retinopathy are essential to avoid vision impairment. We explored different glycated albumin levels in patients with or without diabetic retinopathy, and further constructed a glycated albumin based-model to predict the presence of diabetic retinopathy.Methods: This cross-sectional study, which was conducted at a physical examination center from June 2022 to June 2024, collected clinical information, laboratory test results, and information on diabetic retinopathy from type 2 diabetes adults. Least absolute shrinkage and selection operator regression was applied to select the variables associated with diabetic retinopathy, with optimal threshold determined by receiver operating characteristic curve analysis. Eight machine learning algorithms, including eXtreme Gradient Boosting, logistic regression, Light Gradient-Boosting Machine, random forest, adaptive boosting, K-nearest neighbors, support vector machine, and Gaussian naïve Bayes, were compared to select the model with the best performance in predicting the risk of diabetic retinopathy.Results: Of the 809 eligible patients, 85 (10.5%) and 724 (89.5%) had or had no diabetic retinopathy. Glycated albumin and glycated hemoglobin levels were higher in patients with diabetic retinopathy than in those without diabetic retinopathy. Glycated albumin had satisfactory performance for predicting diabetic retinopathy (area under the curve 0.657 at a threshold of 18.0%). Multivariate regression logistic analysis revealed that glycated albumin was independently correlated with diabetic retinopathy. Machine learning algorithm analysis illustrated that the random forest model had the best performance in predicting the presence of diabetic retinopathy (area under the curve 0.648 and 0.725 in validation and final test sets, respectively). Both calibration curve and decision curve analyses suggested high clinical predication value and applicability for this model.Conclusion: High glycated albumin levels were associated with diabetic retinopathy in patients with type 2 diabetes. A novel prediction model based on glycated albumin could be used to predict the risk of diabetic retinopathy.Keywords: diabetic retinopathy, glycated albumin, machine learning algorithm