Nomogram for prediction of diabetic retinopathy in patients with type 2 diabetes mellitus: A retrospective study

To develop a nomogram for the risk of diabetic retinopathy (DR) among type 2 diabetes mellitus (T2DM). Questionnaires, physical examinations and biochemical tests were performed on 5900 T2DM patients in the Second Hospital of Shijiazhuang. The least absolute shrinkage and selection operator regression was used to optimize feature selection, and the importance of selected features was analyzed by random forest. Logistic regression was performed with selected features, and the nomogram was established based on the results. The Harrell's C-statistic, bootstrap-corrected C-statistic, area under curve (AUC), calibration curve, decision curve analysis (DCA) and clinical impact curve (CIC) were used to validate the discrimination, calibration and clinical usefulness of the nomogram, and further assessment was running by external validation. Predictors included duration of diabetes, diabetic neuropathy, diabetic kidney disease, diabetic foot, hyperlipidemia, hypoglycemic drugs, glycated albumin, Lactate dehydrogenase. The model displayed medium predictive power with a Harrell's C-statistic of 0.820, bootstrap-corrected C-statistic of 0.813 and AUC of 0.820 in the training set, and which was respectively 0.842, 0.835 and 0.842 in the validation set. The calibration curve displayed good agreement (P > 0.05). The DCA and CIC showed that the nomogram could be applied clinically if the risk threshold is between 2 % and 75 % and 2 %–88 % in validation set. This nomogram incorporating 8 features is useful to predict the risk of DR in T2DM patients. •The nomogram was developted based on the general information and biochemical indexes of patients.•The nomogram was constructed by LASSO regression and logistic regression.•LASSO regression was used for screening the related factors to improve the statistical efficiency.•The nomogram can visually display the contribution of each risk factor to diabetic retinopathy.