Anemia is a common complication of nondialysis-dependent chronic kidney disease (CKD) and confers a greater morbidity and mortality (1–7) and lower quality of life (8). Consensus-based guidelines endorse the use of erythropoiesis-stimulating agents (ESAs) for treatment of CKD-related anemia (9). The efficacy of these agents in raising hemoglobin (Hb) (10), reducing the need for red blood cell (RBC) transfusion (9), and improving symptoms has been established (11). Observational studies have also demonstrated that CKD patients receiving ESA treatment have improved survival after the onset of ESRD (12,13). However, recent randomized controlled trials (RCTs) of ESAs in CKD comparing a high versus lower Hb target have raised questions regarding the risks and benefits of treatment and the appropriate treatment targets (14). The recently reported Trial to Reduce Cardiovascular Endpoints with Aranesp Therapy (TREAT) study in anemic diabetic CKD patients did not find a mortality benefit among those treated with darbepoetin compared with placebo and reported an excess risk of stroke. Risk of RBC transfusion was reduced, and a modest improvement in patient-reported fatigue was seen (15).
Current prescribing information for ESAs recommend against complete correction of anemia in CKD. Nevertheless, an important unresolved question in anemia management in CKD is the severity of anemia at which ESA therapy should be initiated and the relative effect of timing of ESA initiation on preventing subsequent blood transfusions and other morbid events. Although RCTs are considered the gold standard for assessing the efficacy of pharmacotherapy approaches such as Hb thresholds for ESA initiation, observational studies can provide complementary evidence, particularly because they can estimate the effectiveness of a therapy in a “real-world” clinical setting. However, these studies are susceptible to threats to validity, including selection bias and confounding by indication, which need to be addressed in the design and statistical analysis.
The objective of this retrospective observational study is to determine the association of timing of ESA initiation with clinical outcomes, including mortality, hospitalization, and blood transfusion, among nondialysis CKD patients, accounting for potential confounding with propensity matching. We hypothesized a priori that ESA initiation when anemia is moderate (Hb 10 to 11 g/dl) compared with more advanced (Hb 8.0 to 9.9 g/dl) would be associated with a lower risk of each clinical outcome.
Materials and Methods
Data were abstracted from national Veterans Health Administration (VHA) data sets, which collect administrative data across multiple domains of medical care. The specific data files used for this analysis included the Patient Treatment File; the Outpatient Care File; the Decision Support System laboratory result files, which contain results for selected laboratory tests; the Decision Support System laboratory ordering files; the Vital Status File; and the Pharmacy Benefits Management data set (16,17).
The study population included patients with incident CKD during one of three time intervals: April 1, 2000 to March 31, 2001; April 1, 2002 to March 31, 2003; or April 1, 2004 to March 31, 2005. These periods were selected because they occurred before, during, and after the release of major guidelines for CKD management (18). Active outpatients with an initial estimated GFR (eGFR) < 60 ml/min/1.73m2 (based on outpatient creatinine) within one of the three time periods were identified; subjects were further required to have eGFR < 60 ml/min/1.73m2 at the time of ESA initiation. Patients with subsequent outpatient ESA use on or after their date of initial low eGFR and with outpatient Hb 8.0 to 11.0 g/dl preceding treatment were eligible for inclusion. Other inclusion criteria included evidence of provider recognition of anemia, defined as one or more of the following within 6 months before initiation of ESA: (1) diagnosis code of anemia; (2) laboratory testing relating to anemia; and (3) an outpatient prescription for iron, folic acid, or vitamin B12.
Exclusion criteria were designed to identify patients who were likely to have other causes for anemia. These included ESA use before incident CKD, hematologic malignancy, chemotherapy recipient, or hospitalized gastrointestinal bleeding on the basis of inpatient and outpatient International Classification of Diseases, 9th Revision, Clinical Modification diagnosis codes. Furthermore, we excluded patients with ESRD at ESA initiation (receiving maintenance dialysis at the Veterans Administration (VA) or eGFR < 10 ml/min/1.73m2).
On the basis of outpatient Hb levels measured before ESA initiation, patients were characterized as having “early” ESA initiation (Hb 10.0 to 11.0 g/dl) or “delayed” ESA initiation (Hb 8.0 to 9.9 g/dl). Patients with pre-ESA Hb values <8.0 g/dl or >11.0 g/dl were excluded.
Patient characteristics that might relate to ESA timing and/or morbidity/mortality were abstracted and considered as potential matching covariates. These characteristics included demographics (age, gender, race), comorbidity defined by inpatient and/or outpatient diagnosis codes any time before the ESA use [diabetes, cardiovascular diseases, chronic lung disease, viral hepatitis, nonhematologic cancer (except nonmelanomatous skin cancer)], outpatient medications within 6 months before ESA initiation (iron preparations, B12 and folic acid, and angiotensin converting enzyme inhibitors/angiotensin receptor blockers), healthcare utilization within the 6 months before ESA use, and socioeconomic factors (insurance, service-connectedness to VA, homelessness, income, and distance to the nearest VHA facility). Outpatient laboratory data included creatinine and Hb within the 6 months before ESA initiation; GFR was estimated using the four-variable Modifications of Diet in Renal Disease equation (19). For each patient, we identified the highest outpatient Hb within these 6 months and computed the linear slope of Hb change preceding treatment initiation.
The primary outcomes were all-cause mortality, all-cause hospitalization, and receipt of a RBC transfusion. Mortality was determined from the Vital Status file, which captures mortality and date of death from several VHA sources and from the Social Security Administration (20). Acute-care hospitalization for any cause was ascertained from the inpatient care data files, and RBC transfusion was ascertained through inpatient and outpatient procedure codes.
Patient characteristics were compared between those with early versus delayed ESA initiation using χ2 and t tests, as appropriate. Those patients with early versus delayed ESA initiation were expected to differ with regards to important prognostic factors that might confound the outcome analyses. Thus, we used propensity matching to balance the two groups with respect to known characteristics (21). In this method, the propensity or likelihood of receiving early versus delayed ESA treatment given their covariate profile was estimated for each patient; each patient who actually received early ESA initiation was matched to a patient who received delayed initiation but had a similar propensity score. Nonparsimonious multiple logistic regression was used to estimate the propensity for early ESA initiation, including as potential covariates those factors that differed (P < 0.10) between the early and delayed groups and significant (P < 0.10) interaction terms. A final propensity-score model was developed containing nine main covariates (race, use of angiotensin converting enzyme inhibitors or angiotensin receptor blockers, prior nephrology care, diabetes, comorbidity index, elapsed time from prior Hb to ESA initiation, prior RBC transfusion, the highest outpatient Hb within prior 6 months, and the linear slope of outpatient Hb change) and one interaction term (RBC transfusion × days from prior Hb to ESA initiation). The c-statistic for this model was 0.69; predictive accuracy did not improve after inclusion of additional covariates or nonlinear continuous covariates. One-to-one nearest-neighbor matching (caliper: 0.02 probability) was used to match individuals in the early and delayed ESA groups, and those participants without a corresponding match were excluded.
The differences in survival and time to first hospitalization and blood transfusion were compared using Kaplan–Meier methods and the log-rank test. At-risk time was defined from the date of ESA initiation until the outcome of interest (death, hospitalization, or transfusion), with censoring on the first of loss to follow-up, development of ESRD, death, or May 31, 2007 (last date of available data). The rationale for censoring on ESRD is that outcome data for VHA patients after the initiation of renal replacement therapy is incomplete. Cox proportional hazards models were used to estimate the hazard ratio (HR) of mortality, hospitalization, and blood transfusion between the matched groups. Negative binomial regression was used to compare rates of all hospitalizations and RBC transfusions (initial and recurrent). For Cox and negative binomial models, robust variance estimators were used to account for within-matched pair correlation. Interaction with prior hospitalization and prior RBC transfusion was examined by testing appropriate interaction terms.
A total of 8034 patients were prescribed outpatient ESAs on or after their initial low eGFR (Figure 1). Of these, 918 (11.8%) had been receiving ESAs before their initial low eGFR, and an additional 2475 patients were excluded because their most recent eGFR before ESA initiation was ≥60 ml/min/1.73m2; most (75%) were under the care of a hematologist-oncologist. After additional exclusion criteria, 1837 patients remained (Figure 1). Of these, 840 (46%) were categorized as having early ESA initiation and 997 (54%) had delayed initiation.
Table 1 compares the characteristics of these two groups at the time of ESA initiation. Compared with the early group, the delayed group was significantly more likely to be Caucasian and to have a high comorbidity index, lack of insurance, and a recent RBC transfusion or acute-care hospitalization. The delayed group was also less likely to have diabetes. Furthermore, the delayed initiation group had a significantly faster mean decline in Hb over the 6 months preceding ESA initiation and had a shorter duration from their last Hb measurement to ESA initiation.
Propensity Score and Matching
After propensity matching, 705 patients with early ESA initiation were successfully matched (within a 0.02 difference) to patients with delayed initiation (76% of total sample matched). Compared with those who were successfully matched, those who were not matched and excluded were younger; more likely to be African American; and to have a history of cancer, a prior blood transfusion, and hospitalization. After matching, there were no significant clinical differences between the early and delayed initiation groups (Table 1).
Association of Early versus Delayed ESA Initiation with Mortality
Over a median 1.8-year follow-up, total mortality was 13.9/100 person-years (early ESA: 13.7/100 person-years; delayed: 14.9/100 person-years). Cumulative mortality did not differ significantly between the propensity-matched groups (Figure 2a; log-rank P = 0.4). The HR in the early versus delayed group was 0.90 [95% confidence interval (CI) = 0.74 to 1.10; Table 2]. No statistical interaction was noted between ESA initiation group and the presence of nephrology subspecialty care, history of recent RBC transfusion, or duration of CKD (P > 0.4).
Early versus Delayed ESA Initiation and Acute-Care Hospitalization
The cumulative proportion (Kaplan–Meier method) without hospitalization was 45.2% within the first year and 19.7% at 5 years (Figure 2b), with a significantly lower proportion in the delayed group. Compared with those with delayed ESA initiation, patients with early initiation had a 17% lower risk of hospitalization (Table 2). The association of early ESA initiation with subsequent initial hospitalization was similar between those not hospitalized in the 6 months before ESA initiation (n = 708; HR = 0.75; 95% CI: 0.61, 0.92) and those who were hospitalized in the 6 months before ESA use (HR = 0.88; 95% CI = 0.75, 1.05; test for interaction: P = 0.3). Likewise, there was no significant difference in the association of early ESA initiation with hospitalization between those with and without prior RBC transfusion, nor between those with different duration of CKD (P > 0.3 for tests of interaction). After excluding those hospitalizations in which RBC transfusion occurred, the difference in hospitalization risk between early and delayed ESA groups was attenuated (HR = 0.90; 95% CI = 0.80, 1.03).
The total hospitalization rate (initial and recurrent) was 96.7/100 person-years. There was no difference in the rate of overall hospitalizations between those with early and delayed ESA initiation (Table 3).
Early versus Delayed ESA Initiation and Blood Transfusion
The cumulative proportion (Kaplan–Meier method) without packed RBC transfusion was 79.0% in the first year and 55.5% within 5 years. Patients with early ESA initiation had a substantially lower cumulative incidence of transfusion compared with propensity-matched patients with delayed initiation (Figure 2c, P < 0.001), with a 29% lower risk (HR = 0.71; Table 2). The association of early ESA initiation and subsequent transfusion did not differ between those with and without pre-ESA transfusion (P = 0.98) or between those with and without pre-ESA hospitalization (P = 0.2 for test of interaction).
The rate of all blood transfusion episodes (first and recurrent) after ESA initiation was 37.7/100 person-years and was 31% lower among those with early initiation (incidence rate ratio = 0.69; Table 3). The association of early ESA initiation with overall rate of transfusion was not different between patients with and without pre-ESA transfusion (test for interaction: P = 0.2).
Using an alternative definition of delayed ESA initiation (Hb 6.0 to 9.9 g/dl), the results after propensity matching were not materially different for risk of mortality (HR = 0.84), transfusion (HR = 0.71), or hospitalization (HR = 0.83).
In this retrospective national cohort study of veterans with nondialysis CKD and anemia, patients with early ESA initiation (Hb between 10.0 and 11.0 g/dl at time of ESA initiation) had a 17% lower risk of initial hospitalizations, a 29% lower risk for initial transfusion, and 31% lower rate of all blood transfusions compared with patients with delayed ESA initiation (Hb 8.0 to 9.9 g/dl). The association of early ESA initiation with time to hospitalization and transfusion was not significantly different between those with and without blood transfusion before ESA use. In contrast to these associations observed with transfusion and hospitalization, no significant mortality difference was observed between the groups. These associations with a reduced risk of hospitalization and blood transfusion are especially notable in light of the very high rates of these events in the study population. Because >95% of transfusions occurred during a hospitalization, it is possible that the lower risk of transfusion in the early initiation group accounted for their lower risk of initial all-cause hospitalization. The risk of hospitalizations in which transfusions did not occur was not significantly lower in the early initiation group; however, because of the administrative nature of the data set, we cannot be certain of the indications for the hospitalizations that occurred in this study population.
The approach to treatment of anemia in nondialysis-dependent CKD has changed since the release of the Correction of Hemoglobin and Outcomes in Renal Insufficiency (CHOIR) and Cardiovascular Risk Reduction by Early Anemia Treatment with Epoetin β (CREATE) study results (14,22), in which more complete correction of Hb with ESAs was associated with no cardiovascular benefit (CREATE) or increased cardiovascular risk (CHOIR) compared with partial correction. Current ESA prescribing guidelines recommend against increasing Hb above 12 g/dl. However, within these prescribing parameters, it remains unclear whether it is preferable to wait to initiate ESA treatment until a patient's Hb falls below 10 g/dl or to begin ESAs when anemia is less severe to avoid potential complications of untreated anemia. The results of this observational study suggest that initiating earlier may reduce the risk of blood transfusion—a major goal of ESA therapy—as well as hospitalizations. In contrast, early initiation was not associated with mortality risk. These mortality and transfusion findings are generally consistent with those of the TREAT study, in which nondialysis diabetic CKD patients with Hb < 11 g/dl were randomized to darbepoetin (target Hb 13 g/dl) or placebo. The risk of transfusion was 44% lower in the active treatment group, but no difference in mortality was observed.
Noninterventional studies of the effects of medical interventions are often more susceptible to the effects of confounding because of channeling of specific interventions to patients with worse prognoses. With multivariate regression techniques, residual confounding may still persist. In these situations, matching subjects on their likelihood of treatment given their covariate history (propensity-score matching) may help to minimize residual bias. However, the success of this approach depends on how well balanced the two treatment groups are with respect to known covariates; as part of this matching process, “extreme outliers,” or those subjects for whom there is no comparable patient in a different treatment group, are excluded (21,23,24). In this analysis, there was very close covariate balance (Table 2, right half) between the two matched treatment groups, particularly with respect to important prognostic indicators including age, kidney function, comorbidity, and previous hospitalizations. Propensity-score matching can control against measured confounding but does not directly control confounding by unmeasured factors, including other healthcare treatment decisions in addition to the timing of ESA initiation. Therefore, the possibility of such unmeasured confounding cannot be excluded.
We hypothesized that differential duration of kidney disease between those with early and delayed ESA initiation might be a potentially important threat to validity. If those with early ESA initiation were treated earlier in their course of CKD, then an apparent benefit might be observed even if the natural history of the disease was unchanged (“lead time bias”). However, the propensity-matched cohorts did not differ in duration or severity of CKD, and the association of early ESA initiation with outcomes did not differ by duration of CKD. Other limitations include the possibility of incomplete ascertainment of outcomes and exposures. Although mortality ascertained in the VHA data are highly complete (20), patients may have required blood transfusions and hospitalizations and received ESAs from sources other than the VA that would not be ascertained in this study. However, the two matched treatment groups had high levels of contact with the VA system before ESA initiation and were similar with regards to demographics and connection with the VHA system such that any misclassification of outcomes or ESA use would not be expected to differ greatly. An additional limitation is that information regarding prescribed ESA doses was not available.
The study population, representing active outpatients in the VHA, is overwhelmingly male; however, there is no reason to expect the association of timing of ESA initiation and blood transfusion to differ between men and women. On the other hand, the study presented here includes a large national cohort of patients with CKD and anemia, reflecting the real-world practices of providers in the VHA. As such, it offers potentially greater external validity than RCTs, which typically enroll a selected population with lower morbidity and mortality compared with the general population with the disease being studied (25,26). For example, 1-year mortality in this study was 17.6% versus <10% in the TREAT study, despite similar subject age.
In conclusion, among a large national cohort of ESA-treated nondialysis-dependent CKD patients with anemia, ESA initiation at a Hb 10.0 to 11.0 g/dl was associated with a significantly lower risk of initial hospitalization and lower rate of subsequent blood transfusions compared with ESA initiation at a Hb level 8.0 to 9.9/dl after controlling for measured confounding by propensity-score matching.
Drs. Gandra, Bradbury, and Chiou are employees of Amgen, Inc. Dr. Fox is a consultant for Amgen, Inc.
This research was funded by a grant from Amgen, Inc. Drs. Seliger, Fink, Hsu, and Walker had complete and sole access to the research data; Dr. Seliger performed the statistical analysis.
Published online ahead of print. Publication date available at www.cjasn.org.
1. Al-Ahmad A, Rand WM, Manjunath G, Konstam MA, Salem DN, Levey AS, Sarnak MJ: Reduced kidney function and anemia as risk factors for mortality in patients with left ventricular dysfunction. J Am Coll Cardiol 38: 955–962, 2001
2. Astor BC, Coresh J, Heiss G, Pettitt D, Sarnak MJ: Kidney function and anemia as risk factors for coronary heart disease and mortality: The Atherosclerosis Risk in Communities (ARIC) study. Am Heart J 151: 492–500, 2006
3. Gerson A, Hwang W, Fiorenza J, Barth K, Kaskel F, Weiss L, Zelikovsky N, Fivush B, Furth S: Anemia and health-related quality of life in adolescents with chronic kidney disease. Am J Kidney Dis 44: 1017–1023, 2004
4. Goicoechea M, de Vinuesa SG, Gomez-Campdera F, Luno J: Predictive cardiovascular risk factors in patients with chronic kidney disease (CKD). Kidney Int Suppl 93: S35–S38, 2005
5. Jurkovitz CT, Abramson JL, Vaccarino LV, Weintraub WS, McClellan WM: Association of high serum creatinine and anemia increases the risk of coronary events: Results from the prospective community-based Atherosclerosis Risk in Communities (ARIC) study. J Am Soc Nephrol 14: 2919–2925, 2003
6. Levin A, Thompson CR, Ethier J, Carlisle EJ, Tobe S, Mendelssohn D, Burgess E, Jindal K, Barrett B, Singer J, Djurdjev O: Left ventricular mass index increase in early renal disease: Impact of decline in hemoglobin. Am J Kidney Dis 34: 125–134, 1999
7. Odden MC, Whooley MA, Shlipak MG: Association of chronic kidney disease and anemia with physical capacity: The Heart and Soul Study. J Am Soc Nephrol 15: 2908–2915, 2004
8. Perlman R, Finkelstein F, Liu L, Roys E: Quality of life in chronic kidney disease (CKD): A cross-sectional analysis in the Renal Research Institute–CKD study. Am J Kidney Dis 45: 658–666, 2005
9. National Kidney Foundation: K/DOQI Clinical Practice Guidelines for Anemia of Chronic Kidney Disease, 2006. Am J Kidney Dis 47[ Suppl 3]: S11–S145, 2006
10. The U.S. Recombinant Human Erythropoietin Predialysis Study: Double-blind, placebo-controlled study of the therapeutic use of recombinant human erythropoietin for anemia associated with chronic renal failure in predialysis patients. The U.S. Recombinant Human Erythropoietin Predialysis Study Group. Am J Kidney Dis 18: 50–59, 1991
11. Leaf DE, Goldfarb DS: Interpretation and review of health-related quality of life data in CKD patients receiving treatment for anemia. Kidney Int 75: 15–24, 2008
12. Fink J, Blahut S, Reddy M, Light P: Use of erythropoietin before the initiation of dialysis and its impact on mortality. Am J Kidney Dis 37: 348–355, 2001
13. Xue JL, St Peter WL, Ebben JP, Everson SE, Collins AJ: Anemia treatment in the pre-ESRD period and associated mortality in elderly patients. Am J Kidney Dis 40: 1153–1161, 2002
14. Singh AK, Szczech L, Tang KL, Barnhart H, Sapp S, Wolfson M, Reddan D; CHOIR Investigators: Correction of anemia with Epoetin alfa in chronic kidney disease. N Engl J Med 355: 2085–2098, 2006
15. Pfeffer MA, Burdmann EA, Chen CY, Cooper ME, de Zeeuw D, Eckardt KU, Feyzi JM, Ivanovich P, Kewalramani R, Levey AS, Lewis EF, McGill JB, McMurray JJ, Parfrey P, Parving HH, Remuzzi G, Singh AK, Solomon SD, Toto R; TREAT Investigators: A trial of darbepoetin alfa in type 2 diabetes and chronic kidney disease. N Engl J Med 361: 2019–2032, 2009
16. Cowper DC, Hynes DM, Kubal JD, Murphy PA: Using administrative databases for outcomes research: Select examples from VA Health Services Research and Development. J Med Syst 23: 249–259, 1999
17. Maynard C, Chapko MK: Data resources in the Department of Veterans Affairs. Diabetes Care 27[ Suppl 2]: B22–B26, 2004
18. National Kidney Foundation: K/DOQI Clinical Practice Guidelines for Chronic Kidney Disease: Evaluation, Classification, and Stratification: Part 4. Definition and classification of stages of chronic kidney disease. Am J Kidney Dis 39: S46, 2002
19. Levey AS, Greene T, Kusek J, Beck G: A simplified equation to predict glomerular filtration rate from serum creatinine (Abstract) J Am Soc Nephrol 11: 155A, 2000
20. Arnold N, Sohn N, Maynard C, Hynes D: VIREC Technical Report 2: VANDI Mortality Data Merge Project, Hines, IL, VA Information Resource Center, 2006
21. Rubin DB: Estimating causal effects from large data sets using propensity scores. Ann Intern Med 127: 757–763, 1997
22. Drüeke TB, Locatelli F, Clyne N, Eckardt KU, Macdougall IC, Tsakiris D, Burger HU, Scherhag A; CREATE Investigators: Normalization of hemoglobin level in patients with chronic kidney disease and anemia. N Engl J Med 355: 2071–2084, 2006
23. Kurth T, Walker AM, Glynn RJ, Chan KA, Gaziano JM, Berger K, Robins JM: Results of multivariable logistic regression, propensity matching, propensity adjustment, and propensity-based weighting under conditions of nonuniform effect. Am J Epidemiol 163: 262–270, 2006
24. Rubin DB: On principles for modeling propensity scores in medical research. Pharmacoepidemiol Drug Saf 13: 855–857, 2004
25. Steg PG, López-Sendón J, Lopez de Sa E, Goodman SG, Gore JM, Anderson FA Jr, Himbert D, Allegrone J, Van de Werf F; GRACE Investigators: External validity of clinical trials in acute myocardial infarction. Arch Intern Med 167: 68–73, 2007
26. Rothwell PM: External validity of randomised controlled trials: “To whom do the results of this trial apply?” Lancet 365: 82, 2005