OBJECTIVES: To identify predictors of off-label utilization of Erythropoietin and Darbepoetin across hospitals in the United States. METHODS: A retrospective database (Solucient®) review was performed on 169,288 discharged patients who received erythropoietin and darbepoetin across 187 hospitals. Based upon an evidence-based medicine framework, utilization of the two drugs was categorized as ‘on-label' (approved by the FDA), ‘off-label-supported' (not FDA-approved but with strong evidence supporting off-label use), and ‘off-label-unsupported' (minimal literature support for off-label indications). A multinomial logistic regression model clustered by hospitals was used. Model covariates were patient demographics, clinical outcomes, physician specialty, hospital size, teaching status, region, drug dose, and number of administrations. RESULTS: Relative to on-label, physician specialty, patient age group, race, and drug coverage were significant (at the 0.05 level) predictors of off-label use (supported and unsupported). Surgeons were twice as likely to prescribe off-label-unsupported (OFUS) than generalists and four times more likely than specialists. Infants (0-1 years), {RRR-164; 95%CI, 84-319}, children (1-17 years), {RRR-2.30; 95%CI, 1.45-5.0}, and young-adults (18-24 years) {RRR-2.30, 95%CI, 2.07-3.19} were more likely to receive OFUS compared to middle-aged adults (40-59 years), while OFUS prescribing for individuals over 75 years was weakly predictive (RRR-1.28; 95%CI, 1.03-1.6). African-Americans and Native-Americans were twice as likely to receive drugs for off-label-supported (OFS) but half as likely for OFUS use relative to whites. Moreover, Title-V, Worker's compensation, and self-pay patients were more likely to receive OFUS. CONCLUSIONS: Variations in off-label prescribing among physician specialties may reflect a lack of consensus on practice guidelines. The common use of OFUS prescribing in pediatrics may be explained by the limited clinical trial data on children. Racial differences in OFUS may indicate differing disease prevalence in populations. Knowing causes of off-label prescribing can help decision makers understand the degree to which it is appropriate.