Interestingly, median 24-hour urine osmolality is greater than that of plasma in humans, suggesting continuous antidiuretic action, which has been associated with renal function decline. Consequentially, Wang et al. recently devised a quantitative method to determine the amount of water needed on a case-by-case basis to achieve a mean urine osmolality Ginsenoside-Ro equivalent to that of plasma. Relationships between urine osmolarity and GFR decline have been described in two studies with contrasting results. We were interested in studying urine volume and urine osmolarity in terms of harder endpoints in chronic kidney disease. Thus we set out to study these variables in terms of risk of initiating dialysis, with death as a competing event. To describe intra- versus inter-individual variance of urine osmolarity, we conducted a variance component analysis including all run-in urine osmolarity values that were available for each patient using a mixed model with patients as levels of a random factor. The outcome variable was time to dialysis, with death as the competing event. Patients who were alive without dialysis at the time of their last visit were censored. Absolute event rates were computed as the number of events divided by the total follow-up time for all patients. Observations with missing values were not used in the calculated models. We described the distribution of time to dialysis using cumulative incidence functions, and compared groups using Gray��s test. Due to the established relationship between baseline creatinine clearance and risk of initiating dialysis/ESRD, and the known progressive loss in urine concentration ability with decreasing renal function, it seemed important to introduce creatinine clearance 
as an Gentiopicrin adjustment factor in all further analyses. We fi ed two multivariate proportional sub-distribution hazards models for competing risk data according to Fine and Gray in order to assess the effect of urine osmolarity or volume on the risk for initiating dialysis. In these models, we considered osmolarity or urine volume and included those variables that either proved significant in a multivariate model or changed the log hazard ratio of osmolarity or urine volume by more than 15% when those variables were excluded from the analyses. We assumed that any variable not selected would have no relevant impact on our conclusions. All variables listed in Table 1 were considered as potential confounders. Results from multivariate competing risk regression were described by means of sub-distribution hazard ratios and 95% confidence intervals, and by computing and visualising estimated cumulative incidence curves at specific covariate values. As urine osmolarity and creatinine clearance were log-base-2 transformed, their SHR correspond to each doubling of these variables. We checked for significant pairwise interactions of variables and for time-dependent effects by including interactions with follow-up time. Non-linear effects were assessed by the method of fractional polynomials. For sensitivity analysis, we also estimated a cause-specific Cox regression model. The R-software, version 2.12, was used for statistical analysis. Other models were calculated for follow-up urine osmolarities and did stay significant after further adjustments.