Baseline estimated glomerular filtration rate and subsequent incident chronic kidney disease
First, the authors used a Validation cohort (n = 5718) and presented hazard ratios (HRs) of baseline eGFR with 60–69 and with 70–79 for CKD, which were extremely high with wide 95% confidence intervals. Although the total number of CKD events is enough for calculating HRs, I am afraid of unstable estimates by their statistical models. Caution should be paid to the applicability of Cox regression models by checking proportionality.2
Second, the authors used Derivation cohort (n = 11 435) and applied receiver operating characteristic (ROC) curve analysis for the prediction of incident CKD. By using this statistical method, the authors concluded that lowered baseline eGFR alone (Model 2) was sufficient for predicting incident CKD without using other factors. I speculate that incident CKD was determined mainly by eGFR and baseline eGFR would contribute to subsequent incident CKD. But there is a problem for excluding other significant variables only by ROC curve analysis. In addition, there is a fluctuation of eGFR for predicting actual GFR.3 Taken together, further study is needed to confirm the usefulness of eGFR without other contributors for incident CKD.
Finally, the authors did not use BMI in their statistical models 1 and 3 in the Derivation cohort. As there is a report that CKD risk increases with the growth of BMI in metabolically healthy individuals,4 BMI should also be considered for the prediction of incident CKD.