Medical Evidence Report Form CMS-2728 data is frequently used to study US dialysis patients, but the validity of these data have been called into question. We compared predialysis erythropoietin use as recorded on Form CMS-2728 with claims data as part of an assessment of quality of care among hemodialysis patients. Medicare claims were linked to Form CMS-2728 data for 18,870 patients. Dialysis patients, 67 years old or older, who started dialysis from 1 June 2005 to 31 May 2007 were eligible. Logistic and multivariate regressions were used to compare the use of either Form CMS-2728 or the corresponding claims data to predict mortality and the probability of meeting target hemoglobin levels. The sensitivity, specificity, and kappa coefficient for the predialysis erythropoietin indicator were 58.0%, 78.4%, and 0.36, respectively. Patients with a predialysis erythropoietin claim were less likely to die compared with patients without a claim (odds ratio = 0.80 and 95% confidence interval = 0.74-0.87), but there was no relationship observed between predialysis care and death using only Form CMS-2728 predictors. At the facility level, a predialysis erythropoietin claim was associated with a 0.085 increase in the rate of meeting target hemoglobin levels compared with patients without a claim (p = 0.041), but no statistically significant relationship was observed when using the Form CMS-2728 indicators. The agreement between Form CMS-2728 and claims data is poor and discordant results are observed when comparing the use of these data sources to predict health outcomes. Facilities with higher agreement between the two data sources may provide greater quality of care.