Starting patients on dialysis early has been increasing in incidence in several countries. However, some studies have questioned its utility, finding a counter-intuitive effect of increased mortality when dialysis was started at a higher estimated glomerular filtration rate (eGFR). To examine this issue in more detail we measured mortality hazard ratios associated with Modification of Diet in Renal Disease eGFR at dialysis initiation for 11,685 patients from the French REIN Registry, with sequential adjustment for a number of covariates. The eGFR was analyzed both quantitatively by 5-ml/min per 1.73 m(2) increments and by demi-decile (i.e., 5 percentiles of the distribution); the 15th demi-decile, including values around 10 ml/min per 1.73 m(2), was our reference point. The patients more likely to begin dialysis at a higher eGFR were older male patients; had diabetes, cardiovascular diseases, or low body mass index and level of albuminemia; or were started with peritoneal dialysis. During a median follow-up of 21.9 months, 3945 patients died. The 2-year crude survival decreased from 79 to 46%, with increasing eGFR from less than 5 to over 20 ml/min per 1.73 m(2). Each 5-ml/min/1.73 m(2) increase in eGFR was associated with a 40% increase in crude mortality risk, which weakened to 9%, but remained statistically significant after adjusting for the above covariates. Analysis by demi-decile showed only the highest to be at significantly higher risk. Hence we found that age and patient condition strongly determine the decision to start dialysis and may explain most of the inverse association between eGFR and survival.