- Email: [email protected]
The effect of epoetin dose on hematocrit
original article
http://www.kidney-international.org & 2008 International Society of Nephrology
The effect of epoetin dose on hematocrit D Cotter1, Y Zhang1, M Thamer1, J Kaufman2 and MA Herna´n3 1
Medical Technology and Practice Patterns Institute, Bethesda, Maryland, USA; 2Renal Section, Veterans Affairs Boston Healthcare Systems and Boston University School of Medicine, Boston, Massachusetts, USA and 3Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA
Nearly all dialysis patients receive epoetin therapy to treat anemia. Using the United States Renal Data System, we monitored the 14 001 patients aged 65 and older who started dialysis and epoetin treatment in 2003–2004. We estimated the dose–response relationship for the average epoetin dose and hematocrit during a 3-month initiation and subsequent 3-month maintenance phase using a marginal structural model to adjust for measured time-dependent confounding by indication. During the initiation phase, an S-shaped dose–response relationship for average weekly epoetin dose and hematocrit response was found. Average hematocrit levels rose as the epoetin dose was increased from 9000 to approximately 22 500 units per week. At higher doses, the effect of increasing epoetin was minimal with average hematocrit levels plateauing at 38.5%, but this was less evident in the maintenance phase. Among patients who reached this phase, doses required to maintain the hematocrit level were lower than those required to achieve similar hematocrit levels in the initiation phase. The dose–response curve found in our study suggests that published recommendations for starting dose are appropriate, and a starting dose of 7500–15 000 units per week can maintain the hematocrit level in the desired target range of 33–36%. Kidney International (2008) 73, 347–353; doi:10.1038/sj.ki.5002688; published online 14 November 2007 KEYWORDS: erythropoietin; anemia; dialysis; hemoglobin; dose–response; marginal structural model
Correspondence: D Cotter, Medical Technology and Practice Patterns Institute, 4733 Bethesda Avenue, Suite 510, Bethesda, Maryland 20814, USA. E-mail: [email protected] Received 4 May 2007; revised 2 August 2007; accepted 28 August 2007; published online 14 November 2007 Kidney International (2008) 73, 347–353
Anemia affects nearly all end stage renal disease (ESRD) patients, results in reduced quality of life and is associated with decreased survival rates.1–3 In 1987, investigators reported successful use of recombinant human erythropoietin (epoetin or EPO, trade name EPOGEN) in treating the anemia of ESRD patients.4–6 By 2005, nearly all dialysis patients covered by Medicare routinely receive epoetin for treatment of anemia associated with renal disease.7,8 Between 1991 and 2005, the mean administered dose of epoetin increased about fourfold, while the mean monthly hematocrit level increased from 28.5 to 36% among the ESRD population.9 The original phase I–II clinical trials demonstrated that as epoetin dose was increased from 15 to 500 U per kg per dose, there was a progressive increase in the achieved hematocrit level.6 Subsequent studies examining the relationship between epoetin dose and hematocrit (or hemoglobin) response have primarily used administrative databases. We and others have found that patients who receive the highest epoetin doses tend to have low hematocrit levels.1,3,10,11 This inverse association reflects the fact that practitioners are targeting a certain hematocrit level, and patients who do not achieve this level will be given higher doses. Hyporesponsive patients, who receive the highest doses, may have underlying inflammatory disorders that blunt the hematocrit response to epoetin.10 To estimate the true pharmacologic relationship between epoetin dose and hematocrit response, one needs to appropriately adjust for this confounding by indication. Because the epoetin dose is partly determined by measured hematocrit dosages, which are themselves affected by prior dosing decisions, special techniques such as inverse probability weighting of marginal structural models12 are needed. These techniques have been previously applied to studies of vitamin D13 and parenteral iron14 use in dialysis patients as well as to other clinical scenarios. In this paper, we use marginal structural models to estimate the dose–response relationship between epoetin dose and hematocrit response in patients more than 65 years old that initiated dialysis without having received prior epoetin therapy. RESULT Initiation phase
Figure 1 shows the selection process for the patients included in the analysis. The demographics, clinical history, comorbidities, and facility characteristics of the 14 001 patients who 347
original article
D Cotter et al.: Epoetin dose and hematocrit response
met our eligibility criteria are shown in Table 1 overall and by average weekly epoetin dose received in the first 3 months after initiation of dialysis. Twenty percent of all patients had an average weekly dose of less than 12 500 units for the first 3 months, 36% had an average weekly dose between 12 500 and 25 000 units, and 44% had an average weekly dose greater than 25 000 units. Of the 14 001 patients, 3964 were censored (1686 died) during the first 4 months, and 682 did not have a hematocrit value in month 4. Figure 2 shows the estimated relation between epoetin dose averaged over the initiation phase and the average hematocrit level of the population at the end of the initiation phase. As expected, the confidence intervals are widest at each end of the dose–response curve where there are the fewest patients in the study population. Given that a starting dose of 13 500 units per week resulted in an average hematocrit level of 36%, the greatest increases in the average hematocrit level of the population take place with epoetin doses between 9000 and 22 500 units per week. At higher doses of epoetin, the average population hematocrit level plateaus at 38.5%. Maintenance phase
This analysis includes the 10 208 patients who reached the end of the initiation phase. Their characteristics are similar to those reported for the total population in Table 1 (data not shown), except that the mean hematocrit level (s.d.) at the beginning of the maintenance phase was 37.0% (4.6). Figure 3 suggests that patients in the initiation phase received larger epoetin doses compared to patients in the maintenance phase. Of the 10 208 patients who reached the end of the initiation phase, 2533 were censored (1262 died) during months 4–7, and 329 did not have a hematocrit value in month 7. The dose–response curve is shown in Figure 4a (overall), Figure 4b (by hematocrit value achieved in month 3), and Figure 4c (by average epoetin dose administered in the first 3 months). Compared with the initiation phase curve (Figure 2), the overall curve in the maintenance phase (Figure 4a) is shifted to the left and has a less clear plateau. Figure 4b indicates that the hematocrit value achieved in the initiation
phase predicts the response to epoetin in the maintenance phase. Specifically, patients who failed to achieve an initiation phase hematocrit level of 433% might not, on average, achieve hematocrit levels 436% during the maintenance phase despite very large doses of epoetin. The dose–response curve during the maintenance phase depends on the dose received in the initiation phase as well. Compared to patients who received lower doses in the initiation phase, patients who were administered higher epoetin doses in the first 3 months also tended to require higher doses in the subsequent maintenance phase (Figure 4c). The dose–response curves shown in Figures 2 and 4 did not materially change when we further adjusted for profit status (with the exception that at the lower doses, not for profit dialysis units had higher average hematocrit values), baseline hypertension, body mass index, glomerular filtration rate, serum creatinine, and time-dependent iron treatment, blood transfusions, dialysis sessions, and urea reduction ratio; when we increased the number of knots in the cubic spline functions for average epoetin dose; when patients with predialysis epoetin use were included; or when we imposed limits on very high or low inverse probability weights. The epoetin dose necessary to maintain an average target hematocrit level of 33–36% in the population was 7500–15 000 U per week. We also conducted unweighted analyses that showed flatter dose–response curves (data not shown), a finding that might be explained because of inappropriate adjustment for confounding by indication. DISCUSSION
We used longitudinal observational data to study the hematocrit response to epoetin treatment outside of the more controlled Phase II study setting. In the initiation phase, we found the dose–response curve to be S-shaped, plateauing at a population average hematocrit level of approximately 38.5% for epoetin doses greater than 20 000 units per week. The initial starting dose ranges (based on a typical patient weight of 70 kg) recommended by the US Food and Drug Administration (FDA) approved labeling (11 250–22 500
37432 patients .65 years with medicare as primary payer initiated hemodialysis in 2003 35477 patients started outpatient hemodialysis within 90 days of the first ESRD service date
30549 patients started epoetin when starting hemodialysis 19191 patients did not use predialysis epoetin
17430 patients did not have kidney transplantation and prior diagnosis of cancer or HIV
14001 pateints had predialysis hematocrit value and were not censored in the first complete month of dialysis
Figure 1 | Selection of study population from USRDS data. 348
Kidney International (2008) 73, 347–353
original article
D Cotter et al.: Epoetin dose and hematocrit response
Table 1 | Patient and provider baseline characteristics (%) by average epoetin dose (N=14 001) Weekly epoetin dose during initiation phase (U)a Total
o12 500
12 500 to o25 000
25 000 to o37 500
X37 500
No. of patients Age 65 o75 years (%) Male (%) White race (%)
14 001 47.0 50.0 71.0
2779 44.3 48.0 74.5
4991 45.8 48.4 71.3
4191 47.6 51.0 70.1
2040 53.6 52.1 65.4
Primary diagnosis (%) Diabetes Hypertension Glomerulonephritis Cystic kidney
44.0 37.0 4.8 0.7
42.7 37.0 3.9 0.7
43.0 37.7 4.9 0.8
45.2 35.7 4.8 0.6
43.0 36.6 5.4 0.6
Cardiovascular comorbidities (%) Non-cardiovascular comorbidities (%) Hypertension (%)
65.0 61.0 37.0
67.3 61.2 37.0
65.4 59.6 37.7
63.9 61.6 35.7
64.4 60.6 36.6
Geographic region (%) Northeast (Networks 1–5) Southeast (Networks 6–8, 13, 14) Midwest (Networks 9–12) West (Networks 15–18)
27.0 31.0 26.0 16.0
26.1 30.2 22.7 21.1
25.3 30.6 26.6 17.6
26.1 31.2 27.3 15.4
31.8 34.1 24.1 10.00
For profit ownership (%)
77.0
74.5
75.5
80.0
81.1
Chain membership (%) Chain 1 Chain 2 Chain 3 Chain 4 Chain 5 Small chain/non-chain (SMC/NC)
26.0 12.0 14.0 7.5 3.2 35.0
26.2 9.7 10.5 5.7 3.9 42.4
26.6 12.7 11.8 6.4 3.4 37.5
27.0 13.1 16.1 8.5 2.9 30.8
24.4 12.4 17.3 10.2 2.3 29.9
26.7±6.5 3.1±0.6 6.0±2.8 10.7±4.9 30.1±4.9 8497±4996 49.0
26±6.1 3.1±0.6 5.7±2.7 11.3±5.1 31.5±5.2 6134±4617 49.0
26.2±6.2 3.1±0.6 6.0±2.8 10.7±4.9 30.4±4.9 7019±5126 49.0
27.1±6.5 3.1±0.6 6.1±2.8 10.5±4.9 29.5±4.6 9421±2612 48.3
28±7.1 3.0±0.6 6.5±3.1 10.2±4.8 28.8±4.5 13 433±4993 50.5
Body mass index (kg m 2)b Serum albumin (g per 100 ml)b Serum creatinine (mg per 100 ml)b Glomerular filtration rateb,c Predialysis hematocrit (%) Baseline epoetin dose per administration (U) Baseline iron use (% yes)
Notes: Plus–minus values are means±s.d. a Weekly epoetin dose may be an average of 1–3 months, depending on length of follow-up. b Not based on complete sample size due to missing data. c Glomerular filtration rate is estimated value.
units per week),15 and the European Renal Association— European Dialysis and Transplant Association (3750–11 250 units per week)16 are located on the steep portion of the dose–response curve, and thus appear to be appropriate starting doses. The National Kidney Foundation/Kidney Dialysis Outcomes and Quality Initiative (KDOQI) guidelines, which used to recommend 9000–13 500 units per week17, now recommend that ‘ythe initial ESA dose and ESA dose adjustments should be determined by the patient’s Hb level, target Hb level, the observed rate of increase in Hb level, and clinical circumstances.’18 In the maintenance phase of our analysis—comprised of those who survived the initiation phase—population average hematocrit levels are consistently higher for the responders (433% hematocrit level at month 3) than for the less responsive patients (o33% hematocrit level at month 3). This finding is consistent with data from randomized controlled Kidney International (2008) 73, 347–353
trials that have shown that some patients targeted to high hematocrit levels did not achieve the targeted levels despite use of high epoetin dosages,19–24 indicating that high hematocrit targets might not be achievable by some dialysis patients. Recent clinical trials have shown that patients targeted to higher hematocrit levels with higher epoetin doses might have an increased mortality.24,25 As a result of these concerns, a recent FDA black box warning advises health-care providers to use ‘the lowest [epoetin] dose possible to gradually increase the hemoglobin concentration’ and to maintain the hematocrit level below 36%. In contrast to this recommendation, in our study population, the average epoetin dose in the initiation phase was higher than that in the maintenance phase. Historically, Medicare epoetin payment policy administered by the Centers for Medicaid and Medicare Services (CMS) was based on an upper bound of achieved hematocrit level 349
D Cotter et al.: Epoetin dose and hematocrit response
40
40
39
39
38
38
Hematocrit at month 7 (%)
Hematocrit at month 4 (%)
original article
37 36 35 34
37 36 35 34
33 33
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Average epoetin dose in months 1−3 (1000 U week −1)
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
Figure 2 | Dose–response curve and 95% confidence intervals during initiation phase.
40 <33%
33−36%
>36%
45 Initiation phase
Maintenance phase
40
Percentage of patients (%)
35 30
Hematocrit at month 7 (%)
39
25
38 37 36 35 34 33
20
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
15 10
40
5
39
0 <12500
12500 to <25 000 25 000 to <37 500
.37 500
Average epoetin dose (U week −1)
Figure 3 | Distribution of patients by average epoetin doses.
without a limit on dose. CMS’ most recent policy, however, reimburses up to a hematocrit level of 39% (which then triggers a 25% epoetin dose reduction). Recently, the House Ways & Means Committee (W&Ms) has held meetings (December 2006 and June 2007) to express their concern with overuse of epoetin and CMS’ new epoetin policy in overreaching the upper limit of the FDA recommended target hematocrit level (36%). Given these concerns and the results from our analysis, an alternative Medicare policy might be that providers first demonstrate that a dose does not result in the desired hematocrit level before using (and being reimbursed for) higher doses. Such a policy would promote the use of the lowest requisite dose for patients (in our study a starting dose of 13 500 units per week resulted in an average hematocrit level of 36%), but it may be difficult to implement and monitor. An alternative policy would be to include epoetin therapy costs in the ESRD composite rate payment, which currently includes primarily the cost of dialysis and selected laboratory tests, as suggested by the Congressional Budget Office, the Inspector General, the Government 350
Hematocrit at month 7 (%)
-25 000 U week −1
>25 000 U week −1
38 37 36 35 34 33 32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
Figure 4 | Dose-response curve for maintenance phase. (a) Dose–response curve and 95% confidence intervals during maintenance phase. (b) Dose–response curve during maintenance phase disaggregated based on patient hematocrit level at month 3. (c) Dose–response curve during maintenance phase disaggregated based on epoetin dose in the first 3 months.
Accounting Office and the Medicare Payment Advisory Committee (MedPAC).26–30 The challenge will be how to adjust the composite rate to foster appropriate use (epoetin dose and target hematocrit level). Our results suggest that 12 000–14 000 U per week will result in an average hematocrit level of 36% in patients more than 65 years old within the first 6 months of dialysis treatment. This benchmark might be useful for calculating an epoetin adjustment to the ESRD Kidney International (2008) 73, 347–353
D Cotter et al.: Epoetin dose and hematocrit response
composite rate, and similar analyses could be performed for other populations. This study has several limitations. First, a key assumption of our analysis is that epoetin dosing decisions are based on the hematocrit values recorded in the database, that is, that unmeasured patient characteristics such as weight, iron stores, malnutrition, inflammation, or epoetin responsiveness only affect dosing decisions through their effect on the recorded hematocrit values. Residual confounding by unmeasured factors might underestimate the average hematocrit levels for large epoetin doses (thus shifting the curve downward as perhaps seen in the curve for nonresponders in Figure 4b). Of note, we adjusted for the measured confounders by inverse probability weighting; conventional adjustment methods would not have yielded valid estimates even in the absence of unmeasured confounding and model misspecification. Second, our analysis is restricted to the first 6 months of dialysis treatment of Medicare patients aged 65 and older who had not received epoetin before initiation of dialysis, and thus might not generalize to other populations. Third, we did not have information on the route of epoetin administration. However, we have previously reported that more than 93% of hemodialysis patients received intravenous administration of epoetin.31 Fourth, for the initiation phase data, the time period of analysis might have been too short to achieve a steady state, and this possibility is consistent with the higher doses for similar achieved hematocrit levels compared to the maintenance phase analysis. In conclusion, our results suggest that doses in the range of 7500–15 000 U per week (or 2300–4600 U given thrice weekly) are appropriate to initiate epoetin therapy, that similar doses will on average maintain hematocrit level in the desired target range of 33–36%, and that hyporesponsive patients may not be able to achieve hematocrits 436% despite very large doses of epoetin. As reimbursement policies for epoetin are being re-examined in the United States, the results of our dose–response analysis might help provide a framework for future policy. MATERIALS AND METHODS We estimated the effect of average epoetin dose in the first 3 months on dialysis (initiation phase) on the achieved hematocrit level at month 4 among incident dialysis patients who were epoetin naive. By choosing epoetin-naive anemic patients and limiting the observation period to the first 3 months of treatment, we selected the period in which change in hematocrit levels would be greatest.6 The outcome was chosen at the end of the fourth month period because a 2- to 4-week lag period has been reported for epoetin to affect hematocrit level.32–34 To examine whether the dose–response relationship changes after the initiation phase, we also estimated the relationship between the average epoetin monthly dose during the second 3-month period on dialysis (maintenance phase, months 4–6) and the achieved hematocrit level at the end of month 7. Study data The United States Renal Data System (USRDS) is a national system that includes demographic and clinical data on patients with ESRD Kidney International (2008) 73, 347–353
original article
and their institutional providers of dialysis treatment. Medicare covers 93% of US dialysis patients, and the USRDS Medicare claims database includes data on monthly hematocrit levels and epoetin doses for these patients. (The USRDS website, http://www.usrds.org, ‘Researcher’s Guide to the USRDS Database’ describes the variables, data source, collection methods, and validation studies.) We used the USRDS standard analytic files as of calendar years 2003 and 2004, which are the most recent available data for researchers (as of May 2007). Institutional claims, including treatment information, were used as the primary data set, and merged with variables from patient, medical evidence, and facility files from the USRDS core CD based on unique patient identifiers. We restricted our analysis to patients who were X65 years of age (because epoetin use before the initiation of dialysis can be reliably determined in this group), started hemodialysis and epoetin treatment concurrently within 90 days after their first ESRD service date and had not used epoetin before initiation of dialysis (to ensure we have the first epoetin claim and complete epoetin therapy information), did not have a kidney transplant, HIV, or cancer before starting dialysis (because these patients might respond differently to epoetin therapy), had a predialysis hematocrit measurement, and were not censored during the first complete month on dialysis. In the initiation phase analysis, patients were followed until the last hematocrit measurement in month 4 or censoring due to change of dialysis modality, transplantation, 30 days after change of dialysis provider, gap in outpatient dialysis services (defined as missing dialysis treatment or hospitalization information for 30 consecutive days or longer), or death, whichever came first. In the maintenance phase analysis, patients were followed from month 4 to the last hematocrit measurement in month 7 or censoring, whichever came first. Statistical methods For each phase (initiation and maintenance), we fit a separate regression model to estimate the dose–response relationship between average epoetin dose and hematocrit response. The outcome of these models was hematocrit response at the end of follow-up (month 4 for initiation phase and month 7 for maintenance phase). The covariates were the average epoetin dose during the follow-up (months 1–3 for initiation phase and months 4–6 for maintenance phase), and the following baseline variables measured at the start of the corresponding phase: age, gender, race, primary cause of ESRD, presence of selected comorbidities, geographic region, chain membership, baseline epoetin, and predialysis or month 3 hematocrit levels (for initiation and maintenance phase, respectively). We modeled average epoetin dose as a continuous variable. The main features of the estimated dose–response curve were not sensitive to the functional form used for this continuous variable as long as enough flexibility was allowed. We used cubic splines with five knots located at the 5th, 25th, 50th, 75th, and 95th percentiles which, for the initiation phase, correspond to weekly epoetin doses of 6528, 13 278, 20 005, 25 685, and 47 749 units, respectively. We then plotted the predicted average hematocrit levels for 100 values of epoetin dose between 500 and 50 000 units per week. Point-wise 95% confidence intervals were calculated for each estimate by using a percentile-based nonparametric bootstrap based on 200 full samples (with replacement) from the observed data.35 For the maintenance phase analysis, we also estimated separate dose– response curves by achieved hematocrit level in month 3 (o33%, X33 to o36% and 436%). Because, at any time, high epoetin doses are more likely to be prescribed to patients with a low achieved hematocrit value, the 351
original article
estimates from the above regression model need to be adjusted for this time-dependent confounding by indication. To do so, one could naively add the post-baseline hematocrit values as covariates in the regression model. However, because those hematocrit values are affected by the epoetin treatment (and are on the causal pathway between epoetin treatment and final hematocrit value), their addition to the model would introduce bias.36 We therefore use another method to adjust for time-dependent confounding by indication: inverse probability weighting. That is, each patient in the above regression models received a weight inversely proportional to the estimated probability of having his/her own observed epoetin dose history. Formally, under the assumption that all time-varying predictors of both epoetin therapy and hematocrit value were included in the analysis (as described in the next paragraph), our weighted model estimates the parameters of a marginal structural mean model.12,37,38 Unlike conventional methods for confounding adjustment that add the time-dependent confounders as covariates in the model, the weighted approach appropriately adjusts for timedependent confounders that are affected by prior epoetin therapy (that is, prior hematocrit value and hospitalization),36 and mimics a randomized trial in which subjects were assigned to different values of average epoetin dose. The epoetin weights were estimated by fitting two nested models: a logistic regression model to estimate each patient’s probability of receiving epoetin (7.3% of the patient-months had zero dose) at each month, and a linear regression model to estimate each patient’s density (assumed to be normal) of log epoetin dose among those with non-zero dose at that month. Both models included baseline demographic characteristics and medical conditions and timevarying factors that might affect epoetin dosing. The baseline variables were: age (years) at ESRD onset, race (black, white, or other), gender, initial epoetin dose and number of epoetin administrations, underlying cause of ESRD (diabetes, glomerulonephritis, hypertension, cystic kidney, or other), initial hematocrit value (predialysis for the initiation phase and in month 3 for maintenance phase; cubic splines with five knots), presence of cardiovascular and non-cardiovascular comorbidities,39 US geographic region (northeast, southeast, Midwest, or west), and provider chain status (because we found a wide variation in epoetin dosing strategies and target hematocrit values based on dialysis chain characteristics40). The time-varying (monthly) variables were hematocrit (in five groups: o30; 30 to o33; 33 to o36; 36 to o39, and 39% or greater; most recent hematocrit levels were carried forward for 7.3% with missing hematocrit levels), hospitalization (yes and no), and epoetin dose (cubic spline with five knots). For ease of utility, the monthly epoetin doses found in administrative data were divided by four to obtain approximate weekly doses. An interaction term between hospitalization and hematocrit level was also added because hospitalization might affect the relationship of hematocrit response and epoetin dose.41 A patient’s epoetin weight was computed as the inverse of the product of his/her estimated probabilities and densities over the follow-up. Similar weights were computed to adjust for potential selection bias due to censoring. Both the epoetin and censoring weights were stabilized, and their product was used to fit the weighted regression model. Analyses were done with SAS (version 9.1).
D Cotter et al.: Epoetin dose and hematocrit response
here have been supplied by the United States Renal Data System (USRDS). The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as the official policy or interpretation of the US Government. ACKNOWLEDGMENTS
This work was supported in part by NIH Grants R01-DK066011-01A2 and R01-HL080644. REFERENCES 1.
2.
3. 4. 5. 6.
7. 8.
9.
10. 11.
12. 13.
14.
15.
16.
17. 18.
19.
20.
21.
22.
DISCLOSURE/CONFLICT OF INTEREST
Dr Kaufman serves as a consultant and has received past grant support from Amgen and F Hoffmann-La Roche. There were no conflicts of interest for any of the other authors. The data reported 352
23.
Collins AJ, Li S, St Peter W et al. Death, hospitalization, and economic associations among incident hemodialysis patients with hematocrit values of 36–39%. J Am Soc Nephrol 2001; 12: 2465–2473. Lowrie EG, Huang WH, Lew NL, Liu Y. The relative contribution of measured variables to death risk among hemodialysis patients. In: Friedman EA (ed). Chapter 13 in Death on Hemodialysis: Preventable Or Inevitable? Kluwer Academic Publishers: Dordrecht; Boston, MA, 1994, pp 121–141. Ma JZ, Ebben J, Xia H et al. Hematocrit level and associated mortality in hemodialysis patients. J Am Soc Nephrol 1999; 10: 610–619. Jacobson LO, Goldwasser E, Fried W et al. Role of the kidney in erythropoiesis. Nature 1957; 179: 633–634. Egrie JC, Browne J, Lai P et al. Characterization of recombinant monkey and human erythropoietin. Prog Clin Biol Res 1985; 191: 339–350. Eschbach JW, Egrie JC, Downing MR et al. Correction of the anemia of end-stage renal disease with recombinant human erythropoietin. Results of a combined phase I and II clinical trial. N Engl J Med 1987; 316: 73–78. Eschbach JW, Adamson JW. Anemia of end-stage renal disease (ESRD). Kidney Int 1985; 28: 1–5. Coladonato JA, Frankenfield DL, Reddan DN et al. Trends in anemia management among US hemodialysis patients. J Am Soc Nephrol 2002; 13: 1288–1295. US Renal Data System. USRDS 2006 Annual Data Report, Atlas of Endstage Renal Disease in the United States. 2006 Avaliable at http:// www.usrds.org/adr.htm,accessed on 8 December 2006. Zhang Y, Thamer M, Stefanik K et al. Epoetin requirements predict mortality in hemodialysis patients. Am J Kidney Dis 2004; 44: 866–876. Madore F, Lowrie EG, Brugnara C et al. Anemia in hemodialysis patients: variables affecting this outcome predictor. J Am Soc Nephrol 1997; 8: 1921–1929. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000; 11: 550–560. Teng M, Wolf M, Ofsthun N et al. Activated injectable vitamin D and hemodialysis survival: a historical cohort study. J Am Soc Nephrol 2005; 15: 1115–1125. Feldman HI, Joffe M, Robinson B et al. Administration of parenteral iron and mortality among hemodialysis patients. J Am Soc Nephrol 2004; 15: 1623–1632. Amgen Inc. EPOGENs Prescribing Information. Thousand oaks, CA, issue date: 03/2007& 1989, 2007. http://www.epogen.com/pdf/ epogen_pi.pdf,accessed on 28 June 2007. Jacobs C, Ho¨rl WH, Macdougall IC et al. European best practice guidelines 9–13: anaemia management. Nephrol Dial Transplant 2000; 15(Suppl 4): S33–S42. NKF-K/DOQI Clinical Practice Guidelines for Anemia of Chronic Kidney Disease: update 2000. Am J Kidney Dis 2001; 37(1 Suppl 1): S182–S238. KDOQI. National Kidney Foundation. II. Clinical practice guidelines and clinical practice recommendations for anemia in chronic kidney disease in adults. Am J Kidney Dis 2006; 47(5 Suppl 3): S16–S85. Besarab A, Bolton WK, Browne JK et al. The effects of normal as compared with low hematocrit values in patients with cardiac disease who are receiving hemodialysis and epoetin. N Engl J Med 1998; 339: 584–590. Parfrey PS, Foley RN, Wittreich BH et al. Double-blind comparison of full and partial anemia correction in incident hemodialysis patients without symptomatic heart disease. J Am Soc Nephrol 2005; 16: 2180–2189. Foley RN, Parfrey PS, Morgan J et al. Effect of hemoglobin levels in hemodialysis patients with asymptomatic cardiomyopathy. Kidney Int 2000; 58: 325–335. Laupacis A. A randomized double-blind study of recombinant human erythropoietin in anaemic hemodialysis patients. Canadian Erythropoietin Study Group. Transplant Proc 1991; 23: 1825–1826. Furuland H, Linde T, Ahlmen J et al. A randomized controlled trial of hemoglobin normalization with epoetin alfa in pre-dialysis and dialysis patients. Nephrol Dial Transplant 2003; 18: 353–361. Kidney International (2008) 73, 347–353
original article
D Cotter et al.: Epoetin dose and hematocrit response
24.
25.
26.
27.
28.
29. 30.
31.
32.
Singh AK, Szczech L, Tang KL, et al., CHOIR Investigators. Correction of anemia with epoetin alfa in chronic kidney disease. N Engl J Med 2006; 355: 2085–2098. Drueke TB, Locatelli F, Clyne N, et al., for the CREATE Investigators. Normalization of hemoglobin level in patients with chronic kidney disease and anemia. N Engl J Med 2006; 355: 2071–2084. United States Government Accountability Office. Report to the chairman, Committee on Ways and Means, House of Representatives. End-stage renal disease: bundling Medicare’s payment for drugs with payments for all ESRD services would promote efficiency and clinical flexibility. GAO07-77; November 2006. GAO. GAO Medicare Dialysis Facilities: beneficiary access stable and problems in payment system being addressed, GAO-04-450. Washington, DC 25 June 2004. Thompson TG Secretary of Health and Human Services, Report to congress: toward a bundled outpatient medicare end stage renal disease prospective payment system. Washington, DC: May 2003. Medicare Payment Advisory Commission (MedPAC). Report to the Congress, Medicare Payment Policy. Washington, DC March 2006. Committee on Ways and Means Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA) y623(e)ESRD Bundled Payment Demonstration. Thamer M, Zhang Y, Kaufman J et al Factors influencing route of administration for epoetin treatment among hemodialysis patients in the United States. Am J Kidney Dis 2006; 48: 77–887. Uehlinger DE, Gotch FA, Sheiner LB. A pharmacodynamic model of erythropoietin therapy for uremic anemia. Clin Pharmacol Ther 1992; 51: 76–89.
Kidney International (2008) 73, 347–353
33.
34.
35.
36. 37.
38.
39.
40.
41.
Eschbach JW, Abdulhadi MH, Browne JK et al. Recombinant human erythropoietin in anemic patients with end-stage renal disease. Results of a phase III multicenter clinical trial. Ann Intern Med 1989; 111: 992–1000. Amgen Inc. EPOGENs (Epoetin alfa) FOR INJECTION. One Amgen Center Drive Thousand Oaks, CA 91320-1799 3xxxxxx – V16, issue date: 19 December 2006& 1989 2006 Amgen Inc., available at http:// www.epogen.com/pdf/epogen_pi.pdf,accessed on 17 January 2007. Efron B, Tibshirani R. Confidence intervals based on bootstrap percentiles. An Introduction to the Bootstrap. Chapman & Hal: New York, 1993: 168–177. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology 2004; 15: 615–625. Herna´n MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 2000; 11: 561–570. Herna´n MA, Brumback B, Robins JM. Estimating the causal effect of zidovudine on CD4 count with a marginal structural model for repeated measures. Stat Med 2002; 21: 1689–1709. Frankenfield DL, Roman SH, Rocco MV et al. Disparity in outcomes for adult Native American hemodialysis patients? Findings from the ESRD Clinical Performance Measures Project, 1996–1999. Kidney Int 2004; 65: 1426–1434. Thamer M, Zhang Y, Kaufman J et al. Dialysis facility ownership and epoetin dosing in patients receiving hemodialysis. JAMA 2007; 297: 1667–1674. Yaqub MS, Leiser J, Molitoris BA. Erythropoietin requirements increase following hospitalization in end-stage renal disease patients. Am J Nephrol 2001; 21: 390–396.
353
http://www.kidney-international.org & 2008 International Society of Nephrology
The effect of epoetin dose on hematocrit D Cotter1, Y Zhang1, M Thamer1, J Kaufman2 and MA Herna´n3 1
Medical Technology and Practice Patterns Institute, Bethesda, Maryland, USA; 2Renal Section, Veterans Affairs Boston Healthcare Systems and Boston University School of Medicine, Boston, Massachusetts, USA and 3Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA
Nearly all dialysis patients receive epoetin therapy to treat anemia. Using the United States Renal Data System, we monitored the 14 001 patients aged 65 and older who started dialysis and epoetin treatment in 2003–2004. We estimated the dose–response relationship for the average epoetin dose and hematocrit during a 3-month initiation and subsequent 3-month maintenance phase using a marginal structural model to adjust for measured time-dependent confounding by indication. During the initiation phase, an S-shaped dose–response relationship for average weekly epoetin dose and hematocrit response was found. Average hematocrit levels rose as the epoetin dose was increased from 9000 to approximately 22 500 units per week. At higher doses, the effect of increasing epoetin was minimal with average hematocrit levels plateauing at 38.5%, but this was less evident in the maintenance phase. Among patients who reached this phase, doses required to maintain the hematocrit level were lower than those required to achieve similar hematocrit levels in the initiation phase. The dose–response curve found in our study suggests that published recommendations for starting dose are appropriate, and a starting dose of 7500–15 000 units per week can maintain the hematocrit level in the desired target range of 33–36%. Kidney International (2008) 73, 347–353; doi:10.1038/sj.ki.5002688; published online 14 November 2007 KEYWORDS: erythropoietin; anemia; dialysis; hemoglobin; dose–response; marginal structural model
Correspondence: D Cotter, Medical Technology and Practice Patterns Institute, 4733 Bethesda Avenue, Suite 510, Bethesda, Maryland 20814, USA. E-mail: [email protected] Received 4 May 2007; revised 2 August 2007; accepted 28 August 2007; published online 14 November 2007 Kidney International (2008) 73, 347–353
Anemia affects nearly all end stage renal disease (ESRD) patients, results in reduced quality of life and is associated with decreased survival rates.1–3 In 1987, investigators reported successful use of recombinant human erythropoietin (epoetin or EPO, trade name EPOGEN) in treating the anemia of ESRD patients.4–6 By 2005, nearly all dialysis patients covered by Medicare routinely receive epoetin for treatment of anemia associated with renal disease.7,8 Between 1991 and 2005, the mean administered dose of epoetin increased about fourfold, while the mean monthly hematocrit level increased from 28.5 to 36% among the ESRD population.9 The original phase I–II clinical trials demonstrated that as epoetin dose was increased from 15 to 500 U per kg per dose, there was a progressive increase in the achieved hematocrit level.6 Subsequent studies examining the relationship between epoetin dose and hematocrit (or hemoglobin) response have primarily used administrative databases. We and others have found that patients who receive the highest epoetin doses tend to have low hematocrit levels.1,3,10,11 This inverse association reflects the fact that practitioners are targeting a certain hematocrit level, and patients who do not achieve this level will be given higher doses. Hyporesponsive patients, who receive the highest doses, may have underlying inflammatory disorders that blunt the hematocrit response to epoetin.10 To estimate the true pharmacologic relationship between epoetin dose and hematocrit response, one needs to appropriately adjust for this confounding by indication. Because the epoetin dose is partly determined by measured hematocrit dosages, which are themselves affected by prior dosing decisions, special techniques such as inverse probability weighting of marginal structural models12 are needed. These techniques have been previously applied to studies of vitamin D13 and parenteral iron14 use in dialysis patients as well as to other clinical scenarios. In this paper, we use marginal structural models to estimate the dose–response relationship between epoetin dose and hematocrit response in patients more than 65 years old that initiated dialysis without having received prior epoetin therapy. RESULT Initiation phase
Figure 1 shows the selection process for the patients included in the analysis. The demographics, clinical history, comorbidities, and facility characteristics of the 14 001 patients who 347
original article
D Cotter et al.: Epoetin dose and hematocrit response
met our eligibility criteria are shown in Table 1 overall and by average weekly epoetin dose received in the first 3 months after initiation of dialysis. Twenty percent of all patients had an average weekly dose of less than 12 500 units for the first 3 months, 36% had an average weekly dose between 12 500 and 25 000 units, and 44% had an average weekly dose greater than 25 000 units. Of the 14 001 patients, 3964 were censored (1686 died) during the first 4 months, and 682 did not have a hematocrit value in month 4. Figure 2 shows the estimated relation between epoetin dose averaged over the initiation phase and the average hematocrit level of the population at the end of the initiation phase. As expected, the confidence intervals are widest at each end of the dose–response curve where there are the fewest patients in the study population. Given that a starting dose of 13 500 units per week resulted in an average hematocrit level of 36%, the greatest increases in the average hematocrit level of the population take place with epoetin doses between 9000 and 22 500 units per week. At higher doses of epoetin, the average population hematocrit level plateaus at 38.5%. Maintenance phase
This analysis includes the 10 208 patients who reached the end of the initiation phase. Their characteristics are similar to those reported for the total population in Table 1 (data not shown), except that the mean hematocrit level (s.d.) at the beginning of the maintenance phase was 37.0% (4.6). Figure 3 suggests that patients in the initiation phase received larger epoetin doses compared to patients in the maintenance phase. Of the 10 208 patients who reached the end of the initiation phase, 2533 were censored (1262 died) during months 4–7, and 329 did not have a hematocrit value in month 7. The dose–response curve is shown in Figure 4a (overall), Figure 4b (by hematocrit value achieved in month 3), and Figure 4c (by average epoetin dose administered in the first 3 months). Compared with the initiation phase curve (Figure 2), the overall curve in the maintenance phase (Figure 4a) is shifted to the left and has a less clear plateau. Figure 4b indicates that the hematocrit value achieved in the initiation
phase predicts the response to epoetin in the maintenance phase. Specifically, patients who failed to achieve an initiation phase hematocrit level of 433% might not, on average, achieve hematocrit levels 436% during the maintenance phase despite very large doses of epoetin. The dose–response curve during the maintenance phase depends on the dose received in the initiation phase as well. Compared to patients who received lower doses in the initiation phase, patients who were administered higher epoetin doses in the first 3 months also tended to require higher doses in the subsequent maintenance phase (Figure 4c). The dose–response curves shown in Figures 2 and 4 did not materially change when we further adjusted for profit status (with the exception that at the lower doses, not for profit dialysis units had higher average hematocrit values), baseline hypertension, body mass index, glomerular filtration rate, serum creatinine, and time-dependent iron treatment, blood transfusions, dialysis sessions, and urea reduction ratio; when we increased the number of knots in the cubic spline functions for average epoetin dose; when patients with predialysis epoetin use were included; or when we imposed limits on very high or low inverse probability weights. The epoetin dose necessary to maintain an average target hematocrit level of 33–36% in the population was 7500–15 000 U per week. We also conducted unweighted analyses that showed flatter dose–response curves (data not shown), a finding that might be explained because of inappropriate adjustment for confounding by indication. DISCUSSION
We used longitudinal observational data to study the hematocrit response to epoetin treatment outside of the more controlled Phase II study setting. In the initiation phase, we found the dose–response curve to be S-shaped, plateauing at a population average hematocrit level of approximately 38.5% for epoetin doses greater than 20 000 units per week. The initial starting dose ranges (based on a typical patient weight of 70 kg) recommended by the US Food and Drug Administration (FDA) approved labeling (11 250–22 500
37432 patients .65 years with medicare as primary payer initiated hemodialysis in 2003 35477 patients started outpatient hemodialysis within 90 days of the first ESRD service date
30549 patients started epoetin when starting hemodialysis 19191 patients did not use predialysis epoetin
17430 patients did not have kidney transplantation and prior diagnosis of cancer or HIV
14001 pateints had predialysis hematocrit value and were not censored in the first complete month of dialysis
Figure 1 | Selection of study population from USRDS data. 348
Kidney International (2008) 73, 347–353
original article
D Cotter et al.: Epoetin dose and hematocrit response
Table 1 | Patient and provider baseline characteristics (%) by average epoetin dose (N=14 001) Weekly epoetin dose during initiation phase (U)a Total
o12 500
12 500 to o25 000
25 000 to o37 500
X37 500
No. of patients Age 65 o75 years (%) Male (%) White race (%)
14 001 47.0 50.0 71.0
2779 44.3 48.0 74.5
4991 45.8 48.4 71.3
4191 47.6 51.0 70.1
2040 53.6 52.1 65.4
Primary diagnosis (%) Diabetes Hypertension Glomerulonephritis Cystic kidney
44.0 37.0 4.8 0.7
42.7 37.0 3.9 0.7
43.0 37.7 4.9 0.8
45.2 35.7 4.8 0.6
43.0 36.6 5.4 0.6
Cardiovascular comorbidities (%) Non-cardiovascular comorbidities (%) Hypertension (%)
65.0 61.0 37.0
67.3 61.2 37.0
65.4 59.6 37.7
63.9 61.6 35.7
64.4 60.6 36.6
Geographic region (%) Northeast (Networks 1–5) Southeast (Networks 6–8, 13, 14) Midwest (Networks 9–12) West (Networks 15–18)
27.0 31.0 26.0 16.0
26.1 30.2 22.7 21.1
25.3 30.6 26.6 17.6
26.1 31.2 27.3 15.4
31.8 34.1 24.1 10.00
For profit ownership (%)
77.0
74.5
75.5
80.0
81.1
Chain membership (%) Chain 1 Chain 2 Chain 3 Chain 4 Chain 5 Small chain/non-chain (SMC/NC)
26.0 12.0 14.0 7.5 3.2 35.0
26.2 9.7 10.5 5.7 3.9 42.4
26.6 12.7 11.8 6.4 3.4 37.5
27.0 13.1 16.1 8.5 2.9 30.8
24.4 12.4 17.3 10.2 2.3 29.9
26.7±6.5 3.1±0.6 6.0±2.8 10.7±4.9 30.1±4.9 8497±4996 49.0
26±6.1 3.1±0.6 5.7±2.7 11.3±5.1 31.5±5.2 6134±4617 49.0
26.2±6.2 3.1±0.6 6.0±2.8 10.7±4.9 30.4±4.9 7019±5126 49.0
27.1±6.5 3.1±0.6 6.1±2.8 10.5±4.9 29.5±4.6 9421±2612 48.3
28±7.1 3.0±0.6 6.5±3.1 10.2±4.8 28.8±4.5 13 433±4993 50.5
Body mass index (kg m 2)b Serum albumin (g per 100 ml)b Serum creatinine (mg per 100 ml)b Glomerular filtration rateb,c Predialysis hematocrit (%) Baseline epoetin dose per administration (U) Baseline iron use (% yes)
Notes: Plus–minus values are means±s.d. a Weekly epoetin dose may be an average of 1–3 months, depending on length of follow-up. b Not based on complete sample size due to missing data. c Glomerular filtration rate is estimated value.
units per week),15 and the European Renal Association— European Dialysis and Transplant Association (3750–11 250 units per week)16 are located on the steep portion of the dose–response curve, and thus appear to be appropriate starting doses. The National Kidney Foundation/Kidney Dialysis Outcomes and Quality Initiative (KDOQI) guidelines, which used to recommend 9000–13 500 units per week17, now recommend that ‘ythe initial ESA dose and ESA dose adjustments should be determined by the patient’s Hb level, target Hb level, the observed rate of increase in Hb level, and clinical circumstances.’18 In the maintenance phase of our analysis—comprised of those who survived the initiation phase—population average hematocrit levels are consistently higher for the responders (433% hematocrit level at month 3) than for the less responsive patients (o33% hematocrit level at month 3). This finding is consistent with data from randomized controlled Kidney International (2008) 73, 347–353
trials that have shown that some patients targeted to high hematocrit levels did not achieve the targeted levels despite use of high epoetin dosages,19–24 indicating that high hematocrit targets might not be achievable by some dialysis patients. Recent clinical trials have shown that patients targeted to higher hematocrit levels with higher epoetin doses might have an increased mortality.24,25 As a result of these concerns, a recent FDA black box warning advises health-care providers to use ‘the lowest [epoetin] dose possible to gradually increase the hemoglobin concentration’ and to maintain the hematocrit level below 36%. In contrast to this recommendation, in our study population, the average epoetin dose in the initiation phase was higher than that in the maintenance phase. Historically, Medicare epoetin payment policy administered by the Centers for Medicaid and Medicare Services (CMS) was based on an upper bound of achieved hematocrit level 349
D Cotter et al.: Epoetin dose and hematocrit response
40
40
39
39
38
38
Hematocrit at month 7 (%)
Hematocrit at month 4 (%)
original article
37 36 35 34
37 36 35 34
33 33
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Average epoetin dose in months 1−3 (1000 U week −1)
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
Figure 2 | Dose–response curve and 95% confidence intervals during initiation phase.
40 <33%
33−36%
>36%
45 Initiation phase
Maintenance phase
40
Percentage of patients (%)
35 30
Hematocrit at month 7 (%)
39
25
38 37 36 35 34 33
20
32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
15 10
40
5
39
0 <12500
12500 to <25 000 25 000 to <37 500
.37 500
Average epoetin dose (U week −1)
Figure 3 | Distribution of patients by average epoetin doses.
without a limit on dose. CMS’ most recent policy, however, reimburses up to a hematocrit level of 39% (which then triggers a 25% epoetin dose reduction). Recently, the House Ways & Means Committee (W&Ms) has held meetings (December 2006 and June 2007) to express their concern with overuse of epoetin and CMS’ new epoetin policy in overreaching the upper limit of the FDA recommended target hematocrit level (36%). Given these concerns and the results from our analysis, an alternative Medicare policy might be that providers first demonstrate that a dose does not result in the desired hematocrit level before using (and being reimbursed for) higher doses. Such a policy would promote the use of the lowest requisite dose for patients (in our study a starting dose of 13 500 units per week resulted in an average hematocrit level of 36%), but it may be difficult to implement and monitor. An alternative policy would be to include epoetin therapy costs in the ESRD composite rate payment, which currently includes primarily the cost of dialysis and selected laboratory tests, as suggested by the Congressional Budget Office, the Inspector General, the Government 350
Hematocrit at month 7 (%)
-25 000 U week −1
>25 000 U week −1
38 37 36 35 34 33 32 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 Average epoetin dose in months 4 −6 (1000 U week −1)
Figure 4 | Dose-response curve for maintenance phase. (a) Dose–response curve and 95% confidence intervals during maintenance phase. (b) Dose–response curve during maintenance phase disaggregated based on patient hematocrit level at month 3. (c) Dose–response curve during maintenance phase disaggregated based on epoetin dose in the first 3 months.
Accounting Office and the Medicare Payment Advisory Committee (MedPAC).26–30 The challenge will be how to adjust the composite rate to foster appropriate use (epoetin dose and target hematocrit level). Our results suggest that 12 000–14 000 U per week will result in an average hematocrit level of 36% in patients more than 65 years old within the first 6 months of dialysis treatment. This benchmark might be useful for calculating an epoetin adjustment to the ESRD Kidney International (2008) 73, 347–353
D Cotter et al.: Epoetin dose and hematocrit response
composite rate, and similar analyses could be performed for other populations. This study has several limitations. First, a key assumption of our analysis is that epoetin dosing decisions are based on the hematocrit values recorded in the database, that is, that unmeasured patient characteristics such as weight, iron stores, malnutrition, inflammation, or epoetin responsiveness only affect dosing decisions through their effect on the recorded hematocrit values. Residual confounding by unmeasured factors might underestimate the average hematocrit levels for large epoetin doses (thus shifting the curve downward as perhaps seen in the curve for nonresponders in Figure 4b). Of note, we adjusted for the measured confounders by inverse probability weighting; conventional adjustment methods would not have yielded valid estimates even in the absence of unmeasured confounding and model misspecification. Second, our analysis is restricted to the first 6 months of dialysis treatment of Medicare patients aged 65 and older who had not received epoetin before initiation of dialysis, and thus might not generalize to other populations. Third, we did not have information on the route of epoetin administration. However, we have previously reported that more than 93% of hemodialysis patients received intravenous administration of epoetin.31 Fourth, for the initiation phase data, the time period of analysis might have been too short to achieve a steady state, and this possibility is consistent with the higher doses for similar achieved hematocrit levels compared to the maintenance phase analysis. In conclusion, our results suggest that doses in the range of 7500–15 000 U per week (or 2300–4600 U given thrice weekly) are appropriate to initiate epoetin therapy, that similar doses will on average maintain hematocrit level in the desired target range of 33–36%, and that hyporesponsive patients may not be able to achieve hematocrits 436% despite very large doses of epoetin. As reimbursement policies for epoetin are being re-examined in the United States, the results of our dose–response analysis might help provide a framework for future policy. MATERIALS AND METHODS We estimated the effect of average epoetin dose in the first 3 months on dialysis (initiation phase) on the achieved hematocrit level at month 4 among incident dialysis patients who were epoetin naive. By choosing epoetin-naive anemic patients and limiting the observation period to the first 3 months of treatment, we selected the period in which change in hematocrit levels would be greatest.6 The outcome was chosen at the end of the fourth month period because a 2- to 4-week lag period has been reported for epoetin to affect hematocrit level.32–34 To examine whether the dose–response relationship changes after the initiation phase, we also estimated the relationship between the average epoetin monthly dose during the second 3-month period on dialysis (maintenance phase, months 4–6) and the achieved hematocrit level at the end of month 7. Study data The United States Renal Data System (USRDS) is a national system that includes demographic and clinical data on patients with ESRD Kidney International (2008) 73, 347–353
original article
and their institutional providers of dialysis treatment. Medicare covers 93% of US dialysis patients, and the USRDS Medicare claims database includes data on monthly hematocrit levels and epoetin doses for these patients. (The USRDS website, http://www.usrds.org, ‘Researcher’s Guide to the USRDS Database’ describes the variables, data source, collection methods, and validation studies.) We used the USRDS standard analytic files as of calendar years 2003 and 2004, which are the most recent available data for researchers (as of May 2007). Institutional claims, including treatment information, were used as the primary data set, and merged with variables from patient, medical evidence, and facility files from the USRDS core CD based on unique patient identifiers. We restricted our analysis to patients who were X65 years of age (because epoetin use before the initiation of dialysis can be reliably determined in this group), started hemodialysis and epoetin treatment concurrently within 90 days after their first ESRD service date and had not used epoetin before initiation of dialysis (to ensure we have the first epoetin claim and complete epoetin therapy information), did not have a kidney transplant, HIV, or cancer before starting dialysis (because these patients might respond differently to epoetin therapy), had a predialysis hematocrit measurement, and were not censored during the first complete month on dialysis. In the initiation phase analysis, patients were followed until the last hematocrit measurement in month 4 or censoring due to change of dialysis modality, transplantation, 30 days after change of dialysis provider, gap in outpatient dialysis services (defined as missing dialysis treatment or hospitalization information for 30 consecutive days or longer), or death, whichever came first. In the maintenance phase analysis, patients were followed from month 4 to the last hematocrit measurement in month 7 or censoring, whichever came first. Statistical methods For each phase (initiation and maintenance), we fit a separate regression model to estimate the dose–response relationship between average epoetin dose and hematocrit response. The outcome of these models was hematocrit response at the end of follow-up (month 4 for initiation phase and month 7 for maintenance phase). The covariates were the average epoetin dose during the follow-up (months 1–3 for initiation phase and months 4–6 for maintenance phase), and the following baseline variables measured at the start of the corresponding phase: age, gender, race, primary cause of ESRD, presence of selected comorbidities, geographic region, chain membership, baseline epoetin, and predialysis or month 3 hematocrit levels (for initiation and maintenance phase, respectively). We modeled average epoetin dose as a continuous variable. The main features of the estimated dose–response curve were not sensitive to the functional form used for this continuous variable as long as enough flexibility was allowed. We used cubic splines with five knots located at the 5th, 25th, 50th, 75th, and 95th percentiles which, for the initiation phase, correspond to weekly epoetin doses of 6528, 13 278, 20 005, 25 685, and 47 749 units, respectively. We then plotted the predicted average hematocrit levels for 100 values of epoetin dose between 500 and 50 000 units per week. Point-wise 95% confidence intervals were calculated for each estimate by using a percentile-based nonparametric bootstrap based on 200 full samples (with replacement) from the observed data.35 For the maintenance phase analysis, we also estimated separate dose– response curves by achieved hematocrit level in month 3 (o33%, X33 to o36% and 436%). Because, at any time, high epoetin doses are more likely to be prescribed to patients with a low achieved hematocrit value, the 351
original article
estimates from the above regression model need to be adjusted for this time-dependent confounding by indication. To do so, one could naively add the post-baseline hematocrit values as covariates in the regression model. However, because those hematocrit values are affected by the epoetin treatment (and are on the causal pathway between epoetin treatment and final hematocrit value), their addition to the model would introduce bias.36 We therefore use another method to adjust for time-dependent confounding by indication: inverse probability weighting. That is, each patient in the above regression models received a weight inversely proportional to the estimated probability of having his/her own observed epoetin dose history. Formally, under the assumption that all time-varying predictors of both epoetin therapy and hematocrit value were included in the analysis (as described in the next paragraph), our weighted model estimates the parameters of a marginal structural mean model.12,37,38 Unlike conventional methods for confounding adjustment that add the time-dependent confounders as covariates in the model, the weighted approach appropriately adjusts for timedependent confounders that are affected by prior epoetin therapy (that is, prior hematocrit value and hospitalization),36 and mimics a randomized trial in which subjects were assigned to different values of average epoetin dose. The epoetin weights were estimated by fitting two nested models: a logistic regression model to estimate each patient’s probability of receiving epoetin (7.3% of the patient-months had zero dose) at each month, and a linear regression model to estimate each patient’s density (assumed to be normal) of log epoetin dose among those with non-zero dose at that month. Both models included baseline demographic characteristics and medical conditions and timevarying factors that might affect epoetin dosing. The baseline variables were: age (years) at ESRD onset, race (black, white, or other), gender, initial epoetin dose and number of epoetin administrations, underlying cause of ESRD (diabetes, glomerulonephritis, hypertension, cystic kidney, or other), initial hematocrit value (predialysis for the initiation phase and in month 3 for maintenance phase; cubic splines with five knots), presence of cardiovascular and non-cardiovascular comorbidities,39 US geographic region (northeast, southeast, Midwest, or west), and provider chain status (because we found a wide variation in epoetin dosing strategies and target hematocrit values based on dialysis chain characteristics40). The time-varying (monthly) variables were hematocrit (in five groups: o30; 30 to o33; 33 to o36; 36 to o39, and 39% or greater; most recent hematocrit levels were carried forward for 7.3% with missing hematocrit levels), hospitalization (yes and no), and epoetin dose (cubic spline with five knots). For ease of utility, the monthly epoetin doses found in administrative data were divided by four to obtain approximate weekly doses. An interaction term between hospitalization and hematocrit level was also added because hospitalization might affect the relationship of hematocrit response and epoetin dose.41 A patient’s epoetin weight was computed as the inverse of the product of his/her estimated probabilities and densities over the follow-up. Similar weights were computed to adjust for potential selection bias due to censoring. Both the epoetin and censoring weights were stabilized, and their product was used to fit the weighted regression model. Analyses were done with SAS (version 9.1).
D Cotter et al.: Epoetin dose and hematocrit response
here have been supplied by the United States Renal Data System (USRDS). The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as the official policy or interpretation of the US Government. ACKNOWLEDGMENTS
This work was supported in part by NIH Grants R01-DK066011-01A2 and R01-HL080644. REFERENCES 1.
2.
3. 4. 5. 6.
7. 8.
9.
10. 11.
12. 13.
14.
15.
16.
17. 18.
19.
20.
21.
22.
DISCLOSURE/CONFLICT OF INTEREST
Dr Kaufman serves as a consultant and has received past grant support from Amgen and F Hoffmann-La Roche. There were no conflicts of interest for any of the other authors. The data reported 352
23.
Collins AJ, Li S, St Peter W et al. Death, hospitalization, and economic associations among incident hemodialysis patients with hematocrit values of 36–39%. J Am Soc Nephrol 2001; 12: 2465–2473. Lowrie EG, Huang WH, Lew NL, Liu Y. The relative contribution of measured variables to death risk among hemodialysis patients. In: Friedman EA (ed). Chapter 13 in Death on Hemodialysis: Preventable Or Inevitable? Kluwer Academic Publishers: Dordrecht; Boston, MA, 1994, pp 121–141. Ma JZ, Ebben J, Xia H et al. Hematocrit level and associated mortality in hemodialysis patients. J Am Soc Nephrol 1999; 10: 610–619. Jacobson LO, Goldwasser E, Fried W et al. Role of the kidney in erythropoiesis. Nature 1957; 179: 633–634. Egrie JC, Browne J, Lai P et al. Characterization of recombinant monkey and human erythropoietin. Prog Clin Biol Res 1985; 191: 339–350. Eschbach JW, Egrie JC, Downing MR et al. Correction of the anemia of end-stage renal disease with recombinant human erythropoietin. Results of a combined phase I and II clinical trial. N Engl J Med 1987; 316: 73–78. Eschbach JW, Adamson JW. Anemia of end-stage renal disease (ESRD). Kidney Int 1985; 28: 1–5. Coladonato JA, Frankenfield DL, Reddan DN et al. Trends in anemia management among US hemodialysis patients. J Am Soc Nephrol 2002; 13: 1288–1295. US Renal Data System. USRDS 2006 Annual Data Report, Atlas of Endstage Renal Disease in the United States. 2006 Avaliable at http:// www.usrds.org/adr.htm,accessed on 8 December 2006. Zhang Y, Thamer M, Stefanik K et al. Epoetin requirements predict mortality in hemodialysis patients. Am J Kidney Dis 2004; 44: 866–876. Madore F, Lowrie EG, Brugnara C et al. Anemia in hemodialysis patients: variables affecting this outcome predictor. J Am Soc Nephrol 1997; 8: 1921–1929. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000; 11: 550–560. Teng M, Wolf M, Ofsthun N et al. Activated injectable vitamin D and hemodialysis survival: a historical cohort study. J Am Soc Nephrol 2005; 15: 1115–1125. Feldman HI, Joffe M, Robinson B et al. Administration of parenteral iron and mortality among hemodialysis patients. J Am Soc Nephrol 2004; 15: 1623–1632. Amgen Inc. EPOGENs Prescribing Information. Thousand oaks, CA, issue date: 03/2007& 1989, 2007. http://www.epogen.com/pdf/ epogen_pi.pdf,accessed on 28 June 2007. Jacobs C, Ho¨rl WH, Macdougall IC et al. European best practice guidelines 9–13: anaemia management. Nephrol Dial Transplant 2000; 15(Suppl 4): S33–S42. NKF-K/DOQI Clinical Practice Guidelines for Anemia of Chronic Kidney Disease: update 2000. Am J Kidney Dis 2001; 37(1 Suppl 1): S182–S238. KDOQI. National Kidney Foundation. II. Clinical practice guidelines and clinical practice recommendations for anemia in chronic kidney disease in adults. Am J Kidney Dis 2006; 47(5 Suppl 3): S16–S85. Besarab A, Bolton WK, Browne JK et al. The effects of normal as compared with low hematocrit values in patients with cardiac disease who are receiving hemodialysis and epoetin. N Engl J Med 1998; 339: 584–590. Parfrey PS, Foley RN, Wittreich BH et al. Double-blind comparison of full and partial anemia correction in incident hemodialysis patients without symptomatic heart disease. J Am Soc Nephrol 2005; 16: 2180–2189. Foley RN, Parfrey PS, Morgan J et al. Effect of hemoglobin levels in hemodialysis patients with asymptomatic cardiomyopathy. Kidney Int 2000; 58: 325–335. Laupacis A. A randomized double-blind study of recombinant human erythropoietin in anaemic hemodialysis patients. Canadian Erythropoietin Study Group. Transplant Proc 1991; 23: 1825–1826. Furuland H, Linde T, Ahlmen J et al. A randomized controlled trial of hemoglobin normalization with epoetin alfa in pre-dialysis and dialysis patients. Nephrol Dial Transplant 2003; 18: 353–361. Kidney International (2008) 73, 347–353
original article
D Cotter et al.: Epoetin dose and hematocrit response
24.
25.
26.
27.
28.
29. 30.
31.
32.
Singh AK, Szczech L, Tang KL, et al., CHOIR Investigators. Correction of anemia with epoetin alfa in chronic kidney disease. N Engl J Med 2006; 355: 2085–2098. Drueke TB, Locatelli F, Clyne N, et al., for the CREATE Investigators. Normalization of hemoglobin level in patients with chronic kidney disease and anemia. N Engl J Med 2006; 355: 2071–2084. United States Government Accountability Office. Report to the chairman, Committee on Ways and Means, House of Representatives. End-stage renal disease: bundling Medicare’s payment for drugs with payments for all ESRD services would promote efficiency and clinical flexibility. GAO07-77; November 2006. GAO. GAO Medicare Dialysis Facilities: beneficiary access stable and problems in payment system being addressed, GAO-04-450. Washington, DC 25 June 2004. Thompson TG Secretary of Health and Human Services, Report to congress: toward a bundled outpatient medicare end stage renal disease prospective payment system. Washington, DC: May 2003. Medicare Payment Advisory Commission (MedPAC). Report to the Congress, Medicare Payment Policy. Washington, DC March 2006. Committee on Ways and Means Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA) y623(e)ESRD Bundled Payment Demonstration. Thamer M, Zhang Y, Kaufman J et al Factors influencing route of administration for epoetin treatment among hemodialysis patients in the United States. Am J Kidney Dis 2006; 48: 77–887. Uehlinger DE, Gotch FA, Sheiner LB. A pharmacodynamic model of erythropoietin therapy for uremic anemia. Clin Pharmacol Ther 1992; 51: 76–89.
Kidney International (2008) 73, 347–353
33.
34.
35.
36. 37.
38.
39.
40.
41.
Eschbach JW, Abdulhadi MH, Browne JK et al. Recombinant human erythropoietin in anemic patients with end-stage renal disease. Results of a phase III multicenter clinical trial. Ann Intern Med 1989; 111: 992–1000. Amgen Inc. EPOGENs (Epoetin alfa) FOR INJECTION. One Amgen Center Drive Thousand Oaks, CA 91320-1799 3xxxxxx – V16, issue date: 19 December 2006& 1989 2006 Amgen Inc., available at http:// www.epogen.com/pdf/epogen_pi.pdf,accessed on 17 January 2007. Efron B, Tibshirani R. Confidence intervals based on bootstrap percentiles. An Introduction to the Bootstrap. Chapman & Hal: New York, 1993: 168–177. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology 2004; 15: 615–625. Herna´n MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 2000; 11: 561–570. Herna´n MA, Brumback B, Robins JM. Estimating the causal effect of zidovudine on CD4 count with a marginal structural model for repeated measures. Stat Med 2002; 21: 1689–1709. Frankenfield DL, Roman SH, Rocco MV et al. Disparity in outcomes for adult Native American hemodialysis patients? Findings from the ESRD Clinical Performance Measures Project, 1996–1999. Kidney Int 2004; 65: 1426–1434. Thamer M, Zhang Y, Kaufman J et al. Dialysis facility ownership and epoetin dosing in patients receiving hemodialysis. JAMA 2007; 297: 1667–1674. Yaqub MS, Leiser J, Molitoris BA. Erythropoietin requirements increase following hospitalization in end-stage renal disease patients. Am J Nephrol 2001; 21: 390–396.
353