Association of quarterly average achieved hematocrit with mortality in dialysis patients: a time-dependent comorbidity-adjusted model.

Background: Recent publications suggest that increased mortality is associated with high hematocrit targets in erythropoietin-stimulating agent-treated patients with chronic kidney disease. We aim to further inform the debate about optimal hematocrit targets, advancing the hypothesis that the current hematocrit target may not optimize the survival of patients with end-stage renal disease.
Study Design: Cross-sectional observational study.
Setting & Participants: Medicare dialysis patients from 2002 to 2004 (n = 393,967).
Factors: Quarterly average hematocrit and erythropoietin alfa (EPO) dose.
Outcomes: Mortality hazard ratios from time-dependent Cox proportional hazard models, adjusting for comorbidities.
Results: N = 2,712,197 patient-facility quarters. During the study, 100,086 deaths were identified. Percentages of patient quarters within each hematocrit category: hematocrit less than 27% (2.0%), 27% to 28.49% (1.7%), 28.5% to 29.9% (2.9%), 30% to 31.49% (5.2%), 31.5% to 32.99% (9.0%), 33% to 34.49% (14.9%), 34.5% to 35.99% (19.2%), 36% to 37.49% (18.0%), 37.5% to 38.99% (12.0%), 39% to 40.49% (6.4%), 40.5% to 41.99% (3.0%), and 42% or greater (3.1%). Mortality hazard ratios from the fully adjusted model: hematocrit less than 27% (3.11), 27% to 28.49% (2.60), 28.5% to 29.9% (2.14), 30% to 31.49% (1.80), 31.5% to 32.99% (1.44), 33% to 34.49% (1.17), 34.5% to 35.99% (reference), 36% to 37.49% (0.98), 37.5% to 38.99% (1.01), 39% to 40.49% (1.13), 40.5% to 41.99% (1.32), and 42% or greater (1.57).
Limitations: First, potential confounding by indication related to associations between underlying illness and mortality, anemia, and EPO responsiveness. Second, Medicare claims data reflect a range of conditions and degrees of severity not easily translated into the clinical context. Third, for Medicare claims, EPO reporting is not required if EPO is not billed. Greater than 95% of "missing hematocrit" quarters are "EPO = 0" patient quarters. Interpretation of results for the missing hematocrit and EPO = 0 use categories is complicated by data source limitations.
Conclusions: We show an association between mortality and low hematocrit in dialysis patients, in part reflecting the presence of comorbidities. We also show an association between increased mortality and high hematocrit. Additional interventional trials should be undertaken to better define the optimal target for anemia management in patients with end-stage renal disease, with careful prospective identification of underlying comorbidities and clinical factors contributing to high erythropoietin-stimulating agent requirement.