Fig. 2: AUC-ROC curves for identification and prediction of CKD and the correlation between input parameters and GMF. | npj Digital Medicine

Fig. 2: AUC-ROC curves for identification and prediction of CKD and the correlation between input parameters and GMF.

From: A digital twin model incorporating generalized metabolic fluxes to identify and predict chronic kidney disease in type 2 diabetes mellitus

Fig. 2

a The identification of CKD in EVAS at baseline yielded an AUC of 0.80. b The identification of CKD in NHANES at baseline yielded an AUC of 0.82. c The prediction of CKD within 3 years in SDR with complete inputs yielded an AUC of 0.86. d The prediction of CKD in CDMD with incomplete inputs yielded an AUC of 0.75. The LR models for the AUC-ROC curve generations utilized the GMF values, age and gender as input parameters. The CI represents the 95% confidence interval of the AUC-ROC curve in the EVAS, NHANES and CDMD datasets. The SDR AUC-ROC curve is depicted for the Testing-2 subset of the SDR dataset, generated by the LR model trained on the SDR training set. e The correlation between biochemical and physiological input parameters with GMF in the CDMD dataset. The numbers in each box represent Kendall’s τ correlation value between two variables, the input parameters vs the GMF values. Stronger positive correlation is indicated by values closer to +1 and depicted in a deeper red hue. Conversely, stronger negative correlation is denoted by values closer to -1 and depicted in a deeper blue hue. These relationships are illustrated in the legend located on the right. Serum creatinine is positively correlated with a number of individual GMF, particularly the GMF related to the respiration-circulation pathways, reactive oxygen species and HbA1c production pathways. It is also negatively correlated to the individual GMF related to the albumin-ACR pathways. LDL, Cholesterol, BMI, glucose and HbA1c cluster together, whereby LDL, Cholesterol and BMI are strongly correlated to the individual GMF related to the lipid metabolism pathways. CKD chronic kidney disease, GMF generalized metabolic fluxes, AUC-ROC area under the curve receiver operating characteristic, LR logistic regression, CI confidence interval, SN Sensitivity, SP Specificity, HbA1c glycated hemoglobin, ACR albumin-creatinine ratio, LDL low density lipoprotein, BMI body mass index.

Back to article page