A randomized trial had suggested that high doses of erythropoiesis-stimulating agents (ESAs) might increase the risk of cardiovascular outcomes in predialysis diabetic patients. To evaluate this risk in diabetic patients receiving dialysis, we used data from 35,593 elderly Medicare patients on hemodialysis in the US Renal Data System of whom 19,034 were diabetic. A pooled logistic model was used to estimate the monthly probability of mortality and a composite cardiovascular end point. Inverse probability weighting was used to adjust for measured time-dependent confounding by indication, estimated separately for diabetic and non-diabetic cohorts. The adjusted 9-month mortality risk, significantly different between an ESA dose of 45,000 and 15,000 U/week, was 13% among diabetics and 5% among non-diabetics. In diabetic patients, the hazard ratio (HR) for more than 40,000 U/week was 1.32 for all-cause mortality and 1.26 for a composite end point of death and cardiovascular events compared with patients receiving 20,000 to 30,000 U/week. The corresponding HRs in non-diabetic patients were 1.06 and 1.10, respectively. A smaller effect of dose was found in non-diabetic patients. Thus, higher ESA doses, which are often necessary to achieve high hemoglobin levels, are not beneficial, and possibly harmful, to diabetic patients receiving dialysis. Our findings support a Food and Drug Administration advisory recommending that the lowest possible ESA dose be used to treat hemodialysis patients.