The optimal target hemoglobin (Hb) concentration in patients with chronic kidney disease (CKD) remains a matter of considerable uncertainty. Randomized controlled trials have demonstrated that aiming for a “normal” Hb concentration does not improve outcomes but may increase the risk of cardiovascular morbidity and/or mortality.1–4 These results suggest the optimal Hb may differ between the physiologic regulation of red cell formation and the therapeutic correction of anemia using erythropoiesis stimulating agents (ESAs) in CKD patients.
An apparent difference between CKD patients receiving ESAs and normal individuals is Hb variability over time. A potential challenge in determining the clinical relevance of Hb fluctuations lies in the method used to assess Hb variability. Currently available metrics of variability mostly reflect a single aspect of variability (e.g., magnitude, frequency, or duration)5,6 and fail to simultaneously capture all components. Within-patient SD is the simplest measure of Hb variability, but it fails to discern patterns or directionality and cannot account for overall trends. Assessing variation derived from a regression line of Hb values (i.e., “residual SD”) can account for overall trend but does not reflect patterns of variability.7 The “time-in-target” is another easy-to-calculate measure, but as with SD and residual SD, it fails to account for variability patterns. Assessing Hb variability using the categories below, within or above a target range, has the advantage of describing the initial range of Hb measures, direction of change, and amplitude of change8,9; however, further statistical analysis becomes limited because it fails to provide a quantitative measure.
To date, large population-based studies on Hb variability have been based on hemodialysis (HD) patient populations in the United States.7–11 Extrapolation of these data to European populations may not be adequate because of differences in patient characteristics and patterns of care, including higher ESA doses in the United States.12
The Analyzing data, Recognizing excellence, Optimizing outcomes (ARO) CKD research initiative aims to identify risk factors and opportunities for intervention in a large European dialysis cohort using a common database of patients from more than 150 Fresenius Medical Care dialysis centers in eastern and western Europe (EU-FME).13 The aim of this study was to characterize Hb variability in these patients using different measures, to identify predictors of Hb variability, and to evaluate the association between Hb variability and mortality.
Results
Patient Characteristics
Baseline characteristics of the 5037 patients included for analysis and those patients excluded (mainly because contiguous Hb measurements >6 months or other relevant data were not available) are shown in Supplemental Table S1. More than two-thirds of patients included in the analysis underwent dialysis treatment for >6 months (on dialysis for >6 months, herein referred to as “prevalent patients”). Less than half of patients excluded were prevalent, which presumably accounts for the differences in several characteristics between included and excluded patients. Patients included for analysis also had a lower mortality rate compared with those excluded (6 versus 27 deaths/100 population, P < 0.01).
Crossvalidation of Existing Measures of Hb Variability
Because of the limitations of using a single approach to determine Hb variability, we used a panel of different measures to quantify variability over a 6-month period (herein referred to as “exposure period”): within-patient SD, residual SD, time-in-target, and the method of fluctuation across thresholds (target range 11.0 to 12.5 g/dl).5–9 We also derived a novel approach to capture magnitude and frequency of variability as a single, quantitative index by integrating the area under the curve (AUC) between measured Hb values and the mean Hb concentration (Figure 1).
;)
Figure 1: Calculating AUC with the trapezoidal rule. To simultaneously capture the magnitude of variability and the frequency at which the fluctuations occur as a single, quantitative index, the AUC between measured Hb values and the mean Hb concentration was calculated. The letters A through J represent the points along a patient profile for Hb. We sliced Hb profiles into 90-day intervals in this study because (
1) most Hb measures were taken monthly, (
2) at least three measurements of Hb were needed to compute an AUC, and (
3) 3 months provide ample time for patients to respond to ESA therapy and have a measurable effect on Hb. The summed integrations across the two 90-day intervals produce an overall index of variability in units of g/dl × days. The area above the curve and the area below the curve were calculated separately for each 3-month interval. Integration was used to calculate the areas, approximating with the “trapezoidal rule”:
where
y i is the
i th Hb measurement taken on day
d i and
m j is the mean Hb value in quarter
j. To integrate the areas, all points that make up the areas needed to be known, therefore it was necessary to linearly interpolate the days where the Hb profile intersected with the mean Hb reference line (points D, G, I) and the Hb value where the Hb profile intersected with the end of the interval (
i.e., at day 90 and day 180 [points E and J]).
Measures of AUC were positively correlated with within-patient SD (Figure 2A; r = 0.85, P < 0.01) and residual SD (Figure 2B; r = 0.88, P < 0.01). AUC was negatively associated with time-in-target, but the correlation was weak (r = –0.17, P < 0.01) (Figure 2C). Results for AUC were consistent with the method of fluctuation across thresholds (Figure 3): patients who were consistently in target had the lowest median value of AUC (35 g/dl × days), whereas patients in the high amplitude category had the largest median value for AUC (107 g/dl × days) and the widest range (31 to 390 g/dl × days).
Figure 2: AUC is highly correlated with within-person SD (A) and residual SD (B) but not with time spent in target (C) (n = 5037).
Figure 3: Distribution of AUC is consistent with categories of method of fluctuation across thresholds. CT, consistently within the target range; CL, consistently low; CH, consistently high; LAH, low-amplitude fluctuation with high Hb levels; LAL, low-amplitude fluctuation with low Hb levels; HA, high-amplitude fluctuation. Frequency of patients in each group: CT = 228, CL = 376, CH = 292, LAH = 1145, LAL = 1682, and HA = 1314. Values above and below the whiskers have been excluded from the box plot. The whiskers are defined as the upper and lower adjacent values where the upper adjacent value is the largest data value that is less than or equal to the 75th percentile + 1.5 × IQR and the lower adjacent value is the smallest data value that is greater or equal to the 25th percentile − 1.5 × IQR.
Patients in the low-amplitude, high-Hb category were in target for 85 ± 50 days; those in the low-amplitude, low-Hb category were in target for 80 ± 49 days; and high-amplitude cyclers were in target for 73 ± 36 days.
Measures of Hb Variability in Incident and Prevalent Patients
Prevalent patients had a slightly higher mean Hb compared with patients who underwent dialysis therapy for <6 months (herein referred to as “incident patients”) and revealed a more stable Hb level during the exposure period (Table 1). Incident patients experienced a greater level of variability compared with prevalent patients (Table 1) as measured by within-patient SD, residual SD, and AUC. Incident patients spent less time-in-target and were more likely to be in the consistently low Hb category or in the high-amplitude Hb category.
Table 1: Comparison of Hb variability measures by dialysis vintage
Predictors of Hb Variability
Patients in the highest AUC quartile were younger, thinner, and more often had diabetes compared with patients in the lower quartiles (Table 2). There were more incident patients in the higher quartiles of AUC, and they were less likely to have an arteriovenous fistula and more likely to have a change in vascular access type during the exposure period when Hb variability was assessed. Patients in the highest AUC quartile had a lower parathyroid hormone level and more frequently received cardiovascular disease (CVD)-related medication or an ESA. Moreover, they were 3 times more likely to have been hospitalized during the exposure period compared with those in the lowest quartile of AUC. The results for quartiles of SD were similar to those for AUC (data not shown).
Table 2: Patient characteristics by quartile of AUC
Multivariable logistic regression showed that young age (<30 years), low body mass index, incident dialysis vintage, change in dialysis access, catheter use, anemia (Hb < 11 g/dl), ESA use, angiotensin-converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB) use, and hospitalizations were important predictors of high Hb variability as measured by AUC (Table 3). A history of CVD and a higher serum albumin were negatively associated with high Hb variability.
Table 3: Predictors of Hb variability (AUC): odds ratio of developing “high” variability (defined as > median Hb variability)
Hb Variability and Risk of Mortality
Patients were followed for a median of 12.4 months (interquartile range [IQR]: 7.7 to 17.4 months). Consequent to the open cohort design, prevalent patients had a longer duration of follow-up (16.0 months; IQR: 9.4 to 17.6 months) than incident patients (8.0 months; IQR: 4.0 to 12.1 months). Of the 551 deaths that occurred during this period, 220 (40%) were due to a CVD-related cause. Of these, 91 deaths were due to heart failure, 32 were due to myocardial infarction, and 45 were due to stroke. Overall, we found no statistically significant association between Hb variability and all-cause mortality (Table 4) as measured by SD, residual SD, time-in-target, and AUC. The only exceptions were observed in the crude model for quartile 2 of within-person SD and time-in-target <29 days; however, the relative risk estimates no longer remained statistically significant with multivariable adjustment.
Table 4: Relative risk of mortality by category of Hb variability
In contrast, results obtained using the method of fluctuations across thresholds showed a positive association between movement across the target range of Hb and mortality (Table 4). The crude analysis showed that patients who fell outside of the target range of 11 to 12.5 g/dl had a significantly greater risk of mortality (with the exception of patients in the low-amplitude high-Hb category). Results of Kaplan–Meier analysis (Figure 4) reinforce these findings. After multivariable adjustment, only patients in the consistently low Hb and the low-amplitude, low-Hb groups had a statistically significant increase in risk compared with patients who were consistently on target. When sensitivity analysis was performed using an 11- to 13-g/dl target range,14 a similar pattern of relative risk was observed, although the magnitude was attenuated for all Hb categories; similar results were observed for a 10- to 12-g/dl target range (Supplemental Table S2).
Figure 4: Kaplan–Meier analysis showing patients with consistently low hemoglobin levels have the highest risk of mortality (n = 5037). CT, consistently within the target range; CL, consistently low; CH, consistently high; LAH, low-amplitude fluctuation with high Hb levels; LAL, low-amplitude fluctuation with low Hb levels; HA, high-amplitude fluctuation.
The overall results for CVD-related mortality were consistent with those for all-cause mortality (data not shown). A stratified analysis by dialysis vintage (incident versus prevalent) showed that Hb variability was not associated with an increased risk for all-cause mortality (data not shown). Results stratified by ESA use (ever use/never use) were consistent with those of the overall analysis (data not shown).
Results of sensitivity analysis using a 3- or 12-month period of Hb exposure (instead of 6-month) found no association between Hb variability and all-cause mortality (data not shown). Sensitivity analysis using the second 6 months of the study period as the exposure period (instead of the first 6 months) on the basis of AUC and SD and the method of fluctuations across target found no association between Hb variability and all-cause mortality. Sensitivity analysis in which the period of the first 3 months of Hb measurements was excluded for incident patients was also consistent with the overall results.
In contrast to Hb variability, low serum albumin at baseline was associated with increased mortality risk (Supplemental Table S3).
Discussion
This study provides the first analysis of Hb variability and outcomes involving a large population of HD patients based outside the United States. It confirms that significant fluctuations of Hb values occur in daily practice and provides novel insights into clinical factors that are potentially associated with greater Hb instability. In contrast to previous studies, our results do not confirm that Hb variability is an independent cause of mortality.
Various established measures were used for a comprehensive assessment of Hb variability. We also developed a new method, referred to as the “AUC approach,” to capture magnitude and frequency of the variability in a single quantitative index. An advantage of AUC is that it does not rely on assumptions with respect to the pattern of variability or the distribution of Hb values, whereas its limitations include the nondirectionality. To validate AUC, we compared its results against those based on existing measures of variability. Results showed that AUC was highly correlated with within-patient SD and residual SD, which reinforces the utility of both measures of SD as a metric of Hb variability.
One interesting observation is that the instability of Hb values was larger among incident patients irrespective of the methodology used to measure Hb variability. This could be due to incident patients experiencing a slight increase in Hb levels during the exposure period, presumably because most of them did not receive ESAs before initiation of dialysis.15 The higher level of variability may also be explained by a higher prevalence of comorbidities among incident patients.16
Our findings for residual SD were relatively consistent with previously reported findings. Although residual SD values of 0.65 and 0.77 g/dl in our study for prevalent and incident patients, respectively, are somewhat lower than values reported in the U.S. Medicare HD population (prevalent: 0.75 g/dl; incident: 0.95 g/dl),5 a study in the prevalent North American Fresenius Medical Care population (FMC-NA) found a residual SD of 0.6 g/dl.7 Despite the similarity of residual SD between the European and U.S. populations, the distribution of patients classified under the method of fluctuations across thresholds differs markedly. Three separate analyses of the prevalent U.S. Medicare population reported that <2% of patients were in the consistently low-Hb category; 5.9% to 6.5% in the consistently in-target category; <3% in the consistently high category; 14.8% to 21.3% in the low-amplitude, low-Hb category; 28.9% to 36.4% in the low-amplitude, high-Hb category; and approximately 40% in the high-amplitude Hb category.5,8,9 Prevalent patients in our study were more frequently found in the consistently low, consistently high, and low-amplitude, low-Hb categories (6.9%, 6.9%, and 32.9%, respectively) and less frequently in the low-amplitude, high-Hb and high-amplitude categories (24.9% and 23.1%, respectively), indicating less frequent fluctuations into the above target range and thus possibly a more conservative approach to raising Hb levels in Europe.
Various parameters known to be associated with severity of renal anemia and/or higher ESA dose requirements are being considered as possible causes of Hb variability.6 Interestingly, the analysis presented here could only confirm a role for some of these factors and the mechanisms by which some predictors affect Hb stability are not immediately apparent. Hospitalizations, change in vascular access, and not having an arteriovenous fistula were associated with greater variability and could affect blood loss, erythropoietic responsiveness, or ESA dosing. A significant association between hospitalizations and Hb variability was also observed in other studies.17 The observation that ESA use itself is a predictor of Hb variability could be because patients not receiving ESA can still adjust their residual endogenous erythropoietin formation. It could also indicate that the pharmacokinetics and pharmacodynamics of ESAs and dosing adjustments contribute to greater Hb variability. Recently it was reported that among nondialysis CKD patients, Hb variability is increased in ESA users.18
As in a previous study, higher serum albumin levels were inversely associated with Hb variability,17 which is consistent with low serum albumin being a surrogate for comorbidity and inflammation. In contrast, C-reactive protein values were identical across quartiles of AUC and SD. Why lower age and body mass index predicted greater Hb variability is less clear, but a similar association with age has been reported previously.17 Equally unclear is the association between the absence of CVD history or antihypertensive medication use and Hb variability.
It is important to note that no consistent algorithm for dose adjustments of ESAs was used in the participating facilities. However, recommended target ranges were different, ranging from between 9 and 12 g/dl (Slovak Republic) to between 11 and 13 g/dl (Czech Republic). Adjustment for these different Hb target ranges as part of sensitivity analysis did not modify the study conclusions (data not shown).
Irrespective of the complexity of the underlying mechanisms, we were unable to confirm that Hb variability is an independent risk factor for all-cause or CVD mortality. In contrast, Yang et al.7 previously found an association of Hb variability and mortality in a large FMC-NA cohort prevalent in 1996. In this cohort, the magnitude of this association became larger when the analysis was restricted to subgroups of Hb to address time-dependent confounding.19 However, an analysis of a smaller, more recent incident cohort by the same investigators found that Hb variability was not associated with decreased survival.10 The investigators concluded that there may be important differences between incident and prevalent cohorts with respect to the effect of Hb variability or perhaps that changes in anemia management over time might account for the different role of Hb variability. The results of this study were stratified by dialysis vintage; however, we found no association between Hb variability and mortality in either patient group.
Although Hb variability per se was not associated with decreased survival, patients in the consistently low-Hb category and those in the low-amplitude, low-Hb category had an increased risk for mortality. These observations are consistent with a recent analysis by Gilbertson et al.,9 who found the number and timing of Hb values <11 g/dl and falling Hb values were associated with increased mortality. The cause and effect of this relationship cannot easily be established given the absence of randomized controlled trials comparing the currently recommended Hb target range with a lower range.20 On the other hand, it appears noteworthy in light of the concern about the risks associated with high-Hb targets that in the study presented here and in previous analyses9,21 those patients who had achieved Hb levels consistently >12.5 g/dl did not show an increase in mortality risk. When different Hb ranges (11 to 13 g/dl or 10 to 12 g/dl) than 11 to 12.5 g/dl were used to analyze patient risks, the relative risk in the consistently low Hb category remained elevated; however, the relative risk for all other Hb categories (including consistently above target) remained near 1.0.
The limitations of this study include the observational nature, with the inherent difficulties to determine cause-and-effect relationships, and the fact that the results are based on a clinical database that was not originally established as a research tool.22 A more specific limitation is that the database is from a single private dialysis provider (EU-FME). Thus, the generalizability to patients under the care of other providers and in countries without EU-FME centers remains unclear. Another limitation is the large percentage of patients (44%) who were excluded from analysis as a result of the open cohort design (52% of excluded patients were incident), which may have resulted in selection bias. The absence of information on iron administration is another limitation because iron therapy may have influenced Hb variability or patient survival. Furthermore, the median observation period was approximately 1 year, which precludes any conclusions on longer-term mortality. Although the study included >5000 patients, the numbers of events in subgroups of patients were relatively small. Nevertheless, study characteristics were sufficient to confirm that low serum albumin, an established risk marker,23 predicts poor outcomes.
In conclusion, Hb variability occurs in the European HD population to a similar extent as in U.S. HD populations. It is related to various patient characteristics, comorbidities, and hospitalizations but is not an independent risk factor for all-cause or CVD mortality.
Concise Methods
Study Population
The entire population consisted of randomly selected HD patients treated between January 2005 and December 2006 at an EU-FME facility in one of 11 countries: Czech Republic, France, Hungary, Italy, Poland, Portugal, Slovak Republic, Slovenia, Spain, Turkey, and the United Kingdom (n = 11,153).13 We excluded patients from centers where most data on actual blood flow or Kt/V were missing (n = 1352). U.K. patients were excluded because of missing information on medications (n = 838). From the remaining 8963 patients, 5427 patients were selected who had at least 6 months of contiguous monthly Hb measurements over the exposure period. Patients were excluded if they had a bleeding episode or a blood transfusion (n = 390) during follow-up, resulting in a study cohort of 5037 patients. Patients were classified as incident if they had received HD therapy for <6 months at the time of enrollment.
Assessment of Hb Variability
Hb variability was assessed using within-patient SD, residual SD, time-in-target, and the method of fluctuation across thresholds over a 6-month period of evaluation.7–9 An 11.0- to 12.5-g/dl target was used for time-in-target and the method of fluctuation across thresholds. Sensitivity analysis was performed using 10- to 12-g/dl and 11- to 13-g/dl targets. We also derived a novel approach to capture the magnitude of variability and the frequency at which the fluctuations occur as a quantitative index referred to as the AUC approach (Figure 1).
Statistical Analysis
Unadjusted comparisons were performed using a paired or independent t test, χ2 test, or Wilcoxon rank-sum test as appropriate. Poisson regression was used to assess mortality rates. Predictors of high variability (defined as AUC > 50th percentile: 68.2 g/dl × days) were evaluated using logistic regression.
Cox regression was used to examine the association between Hb variability and mortality. Subjects began to accrue risk for up to 18 months after the Hb variability was assessed (i.e., the exposure period) until death or a censoring event. Patients were considered lost to follow-up if they left a dialysis facility and did not return within 45 days. Sensitivity analysis evaluated the effect of changing the duration of Hb exposure (3 and 12 months). We reanalyzed the data using the second 6 months of Hb variability measurements. Furthermore, we evaluated the effect of excluding the first 3 months of incident patients' follow-up. Time-to-event analysis was performed using Kaplan–Meier methods to evaluate survival to 18 months.
All statistical analyses were performed using Stata (version 10.0, College Station, TX) and were reproduced independently by a second statistician using SAS (version 9.0, Cary, NC).
Disclosures
Kai-Uwe Eckardt has received consulting or lecture fees from Affymax, Amgen, Johnson & Johnson, Kirin, Roche, and Sandoz Hexal. Florian Kronenberg received consulting fees from Amgen. Pedro Aljama has received research grants and participated on advisory boards for Amgen, Janssen-Cilag, and Roche. Stefan Anker has received consultancy fees for Amgen, Vifor International, and Fresenius Kabi and honoraria for lectures from Amgen and Vifor International. Bernard Canaud has received grants for research from Fresenius, Baxter, Bellco, and the Public Ministry of Health and honoraria as an invited speaker from Amgen, Roche, Janssen-Cilag, Shire, and Genzyme. Peter Stenvinkel is a member of the Scientific Advisory Board at Gambro and has given lectures at meetings organized by Amgen, Baxter, Genzyme, Roche, and Astra Zeneca. He has received a research grant and consulting fees from Amgen. Guntram Schernthaner has received consulting fees for advisory board meetings from Amgen and Roche. Iain C. Macdougall has received lecture and consulting fees from Affymax, Amgen, Ortho Biotech, Roche, Shire, Vifor Pharma, and grant support from Affymax, Amgen, Ortho Biotech, and Roche. Joseph Kim, Elizabeth Ireland, Bruno Fouqueray, and Bart Molemans are employees of Amgen.
Funding for the ARO research initiative was provided by Amgen (Europe) GmbH, Zug, Switzerland. We are grateful to the ARO Steering Committee for their valuable comments. The authors wish to thank Susan Wieting and Caterina Hatzifoti (Amgen Europe) for providing editorial assistance. ARO CKD Research Initiative Steering Committee members: Pedro Aljama, Stefan D. Anker, Bernard Canaud, Charles Chazot, Angel L.M. de Francisco, Tilman Drüeke, Kai-Uwe Eckardt, Jürgen Floege, Bruno Fouqueray, Joseph Kim, Florian Kronenberg, Iain C. Macdougall, Daniele Marcelli, Bart Molemans, Jutta Passlick-Deetjen, Guntram Schernthaner, Peter Stenvinkel, and David C. Wheeler
Published online ahead of print. Publication date available at www.jasn.org.
Supplemental information for this article is available online at http://www.jasn.org/.
References
1. Besarab A, Bolton WK, Browne JK, Egrie JC, Nissenson AR, Okamoto DM, Schwab SJ, Goodkin DA: 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 339: 584–590, 1998
2. Singh AK, Szczech L, Tang KL, Barnhart H, Sapp S, Wolfson M, Reddan D: Correction of anemia with epoetin alfa in chronic kidney disease. N Engl J Med 355: 2085–2098, 2006
3. Drueke TB, Locatelli F, Clyne N, Eckardt KU, Macdougall IC, Tsakiris D, Burger HU, Scherhag A: Normalization of hemoglobin level in patients with chronic kidney disease and anemia. N Engl J Med 355: 2071–2084, 2006
4. 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
5. Arneson TJ, Zaun D, Peng Y, Solid CA, Dunning S, Gilbertson DT: Comparison of methodologies to characterize haemoglobin variability in the US Medicare haemodialysis population. Nephrol Dial Transplant 24: 1378–1383, 2009
6. Kalantar-Zadeh K, Aronoff GR: Hemoglobin variability in anemia of chronic kidney disease. J Am Soc Nephrol 20: 479–487, 2009
7. Yang W, Israni RK, Brunelli SM, Joffe MM, Fishbane S, Feldman HI: Hemoglobin variability and mortality in ESRD. J Am Soc Nephrol 18: 3164–3170, 2007
8. Ebben JP, Gilbertson DT, Foley RN, Collins AJ: Hemoglobin level variability: Associations with comorbidity, intercurrent events, and hospitalizations. Clin J Am Soc Nephrol 1: 1205–1210, 2006
9. Gilbertson DT, Ebben JP, Foley RN, Weinhandl ED, Bradbury BD, Collins AJ: Hemoglobin level variability: Associations with mortality. Clin J Am Soc Nephrol 3: 133–138, 2008
10. Brunelli SM, Lynch KE, Ankers ED, Joffe MM, Yang W, Thadhani RI, Feldman HI: Association of hemoglobin variability and mortality among contemporary incident hemodialysis patients. Clin J Am Soc Nephrol 3: 1733–1740, 2008
11. Fishbane S, Berns JS: Evidence and implications of haemoglobin cycling in anaemia management. Nephrol Dial Transplant 22: 2129–2132, 2007
12. Pisoni RL, Bragg-Gresham JL, Young EW, Akizawa T, Asano Y, Locatelli F, Bommer J, Cruz JM, Kerr PG, Mendelssohn DC, Held PJ, Port FK: Anemia management and outcomes from 12 countries in the Dialysis Outcomes and Practice Patterns Study (DOPPS). Am J Kidney Dis 44: 94–111, 2004
13. DeFrancisco ALM, Aljama P, Kim J, Marcelli D. An epidemiological study of haemodialysis patients based on the European Fresenius Medical Care (FMC) haemodialysis network: The ARO Research Initiative [Abstract]. NDT Plus 1: ii384–ii385, 2008
14. Locatelli F, Aljama P, Barany P, Canaud B, Carrera F, Eckardt KU, Horl WH, Macdougall IC, Macleod A, Wiecek A, Cameron S. Revised European Best Practice Guidelines for the management of anaemia in patients with chronic renal failure. Nephrol Dial Transplant 19: ii1–ii47, 2004
15. Thilly N, Stengel B, Boini S, Villar E, Couchoud C, Frimat L: Evaluation and determinants of underprescription of erythropoiesis stimulating agents in pre-dialysis patients with anaemia. Nephron Clin Pract 108: c67–c74, 2008
16. van Manen JG, van Dijk PC, Stel VS, Dekker FW, Cleries M, Conte F, Feest T, Kramar R, Leivestad T, Briggs JD, Stengel B, Jager KJ: Confounding effect of comorbidity in survival studies in patients on renal replacement therapy. Nephrol Dial Transplant 22: 187–195, 2007
17. Berns JS, Elzein H, Lynn RI, Fishbane S, Meisels IS, Deoreo PB: Hemoglobin variability in epoetin-treated hemodialysis patients. Kidney Int 64: 1514–1521, 2003
18. Boudville NC, Djurdjev O, Macdougall IC, de Francisco AL, Deray G, Besarab A, Stevens PE, Walker RG, Urena P, Inigo P, Minutolo R, Haviv YS, Yeates K, Aguera ML, MacRae JM, Levin A: Hemoglobin variability in nondialysis chronic kidney disease: Examining the association with mortality. Clin J Am Soc Nephrol 4: 1176–1182, 2009
19. Brunelli SM, Joffe MM, Israni RK, Yang W, Fishbane S, Berns JS, Feldman HI: History-adjusted marginal structural analysis of the association between hemoglobin variability and mortality among chronic hemodialysis patients. Clin J Am Soc Nephrol 3: 777–782, 2008
20. KDOQI Clinical Practice Guidelines and Clinical Practice Recommendations for Diabetes and Chronic Kidney Disease. Am J Kidney Dis 49: S12–S154, 2007
21. Sturm G, Lamina C, Zitt E, Lhotta K, Lins F, Freistätter O, Neyer U, Kronenberg F. Sex-specific association of time-varying haemoglobin values with mortality in incident dialysis patients. Nephrol Dial Transplant 2010, in press
22. Jager KJ, Stel VS, Wanner C, Zoccali C, Dekker FW: The valuable contribution of observational studies to nephrology. Kidney Int 72: 671–675, 2007
23. Lowrie EG, Lew NL: Death risk in hemodialysis patients: The predictive value of commonly measured variables and an evaluation of death rate differences between facilities. Am J Kidney Dis 15: 458–482, 1990
