The United States Renal Data System (USRDS) registry of end-stage renal disease has often been used to study the timing of dialysis initiation, measured by estimated glomerular filtration rate (eGFR) at dialysis initiation. We conducted an observational study and examined how well variables in the USRDS database explain the trends and variation in eGFR at dialysis initiation.We identified 971,481 patients who initiated dialysis between 1995 and 2012 in the USRDS registry.The mean eGFR at dialysis initiation monotonically rose from 7.7 in 1995 to 11.1 in 2009, and then leveled off to 10.9 mL/min/1.73 m in 2012. The trend of rising, then leveling off was similar across all subgroups studied. Substantial variation in eGFR at dialysis initiation was observed, with standard deviation of 4.38 (95% CI: 2.0-18.4). A total of 11.4% of the total variation occurred across physicians and 88.6% within physicians. Adjustment for measured factors only modestly decreased the total variation. Of the total variance, 10.7% was explained by measured patient-level variables and 1.2% by measured physician and other factors, while 9.2% of physician-level variation and 78.9% of patient-level variation remained unexplained. The extent of variation explained by measured variables was similar over the entire study period.The finding that the majority of variation in eGFR at dialysis initiation is unexplained by measured variables casts doubt on how well eGFR serves as a measure for "timing" of dialysis initiation, and it indicates the need to collect more focused data to gain understanding of factors that affect timing of dialysis initiation in the US.