Fig. 2

Models for predicting diabetic kidney disease and model evaluation of performance and validation. (A) Receiver operating characteristic curves for evaluating the discrimination ability of the model. SuperLearner had the highest area under the curve compared with the other models (p < 0.05*). *roc.test() was used for pairwise comparison of receiver operating characteristic curves and the results are presented in the Table S3. AUC: area under curve; MLR: multivariate logistic regression; RF: random forest. (B) Comparison of survival curves (end event: diabetic kidney disease) in different risk groups using SuperLearner (P < 0.01). *Regrettably, two individuals from the ‘predicted high-risk’ group and one individual from the ‘predicted low-risk’ group lacked the necessary time stamp for their final follow-up, necessitating their omission from the survival curve analysis