Baseline characteristics were compared according to the occurrence of outcomes using chi-square test and t-test methods. For non-normally distributed continuous variables, Mann–Whitney U-test was used. Incidence differences with respect to the diabetic condition and CKD stages were compared using Poisson regression analysis. To explore the effect of CKD stages on adverse outcomes, we developed cause-specific hazard models as the competing risk analysis with stage G1 as a reference18. Renal replacement therapy (RRT) prior to the outcome was considered a competing risk. Using variables that showed significant differences between CVD ( +) and CVD (−) groups and after removing variables that could act as mediators, adjusted models were constructed as follows: model 1, adjusted for age and sex; model 2, adjusted for the variables in model 1 plus diabetes and preexisting CVD; and model 3, further adjusted for diastolic blood pressure, total cholesterol, high-density lipoprotein cholesterol, and hsCRP. Sensitivity analysis was performed to determine the HR of outcomes in each CKD stage of G3a to G5 compared to combined stages G1–G2, to address the possibility that subjects with glomerular hyperfiltration might have been included in stage G1 in our CKD population. The results were presented as hazard ratios (HRs) and 95% confidence intervals (CIs). Subgroup analyses were performed according to age, sex, diabetes, urine protein-to-creatinine ratio of 1 g/g Cr, and cause of CKD. In subgroup analysis, HRs were derived from cause-specific hazard models adjusted for age and sex, with CKD stages G1–G2 as the reference. To compare the outcome incidence with the Korean general population, the NHIS—National Sample Cohort (2002–2013) was used 19. The sample cohort consists of a million subjects (2% of the total Korean population), and was selected as a representative sample using systematic stratified random sampling. After excluding patients with preexisting CVD events within 1 year, pregnancy, cancer, liver cirrhosis, organ transplantation, and RRT, a total of 710,362 subjects aged 20–79 were finally included in the comparative analysis (Supplemental Methods). Both unadjusted and age-and-sex adjusted Cox-proportional hazards models were conducted to calculate the HRs of CVD and the composite outcome for the KNOW-CKD cohort in comparison with the Korean general population. P < 0.05 was considered statistically significant in all statistical analyses. All statistical analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC) and R version 3.6.1 (Foundation for Statistical Computing, Vienna, Austria).