Regional variability in anaemia management and haemoglobin in the US

Transcription

1 Nephrol Dial Transplant (2003) 18: Original Article Regional variability in anaemia management and haemoglobin in the US Donal N. Reddan 1, Diane L. Frankenfield 2, Preston S. Klassen 1, Joseph A. Coladonato 1, Lynda Szczech 1, Curtis A. Johnson 3, Anatole Besarab 4, Michael Rocco 5, William McClellan 6, Jay Wish 7 and William F. Owen Jr 1 for CMS s ESRD CPM Workgroup 1 Duke Institute of Renal Outcomes Research and Health Policy, Duke University Medical Center, Durham, NC, 2 Center for Beneficiary Choices, Centers for Medicare & Medicaid Services, Baltimore, MD, 3 University of Wisconsin School of Pharmacy, Madison, WI, 4 West Virginia University, Morgantown, WV, 5 Wake Forest University School of Medicine, Winston-Salem, NC, 6 Emory University, Atlanta, GA and 7 University Hospitals of Cleveland, Cleveland, OH, USA Abstract Background. Regional differences in haemoglobin values and process care measures were examined using data from the Centers for Medicare & Medicaid Services End-Stage Renal Disease (ESRD) Clinical Performance Measures Project. It was posited that regional differences in haemoglobin values are consequent upon differences in components of clinical practice. Methods. A national random sample of 8336 adult, in-centre haemodialysis patients, stratified by the 18 regional ESRD Networks, was drawn. Information was collected for October December Multivariable stepwise linear and logistic regression analyses were performed to identify variables associated with haemoglobin. Linear regression analysis was used to identify variables associated with EpouHb index (mean weight-adjusted treatment level erythropoietin (Epo) dose divided by mean haemoglobin). Results. The percentage of patients with haemoglobin concentration -11 gudl ranged from 34 to 52% across ESRD Networks. In addition to haemoglobin there was significant, non-random variation among ESRD Networks with regard to prescribed Epo dose and administration route, intravenous (IV) iron prescription and dialyser flux (high fluxskuf P 20 mlummhguh) (all P-values ). Higher haemoglobin was associated with older age, male gender, higher serum albumin, higher transferrin saturation, Correspondence and offprint requests to: Dr Donal Reddan, Duke Institute of Renal Outcomes Research and Health Policy, Box 3646, Duke University Medical Center, Durham, NC 27710, USA. redda001@mc.duke.edu The authors wish it to be known that, in their opinion, the first two authors contributed equally to this work. The views expressed in this manuscript are those of the authors and do not necessarily reflect official policy of the Centers for Medicare & Medicaid Services. higher KtuV, lower serum ferritin and lower prescribed Epo dose (all P-values -0.01). Diabetes mellitus as cause of ESRD, high-flux dialyser use, IV iron prescription or subcutaneous Epo prescription were not associated with haemoglobin. Male gender, diabetes as cause of ESRD, older age, higher transferrin saturation and higher albumin concentrations were associated with lower EpouHb index. Prescription of IV iron and IV Epo were associated with higher EpouHb index. Conclusions. Regional mean haemoglobin levels vary considerably across the US and the variation in haemoglobin is explained by both non-modifiable factors and modifiable clinical practice-derived variables. Keywords: anaemia; Centers for Medicare & Medicaid Services; end-stage renal disease; erythropoietin; haemodialysis; iron Introduction Anaemia affects the majority of end-stage renal disease (ESRD) patients and almost all affected patients receive recombinant (biosynthetic) human erythropoietin (Epo) as primary treatment for their anaemia. Because the extent of anaemia correction in ESRD patients has been associated with clinical outcomes, clinical practice guidelines arising from a structured review of available clinical literature advocate a haematocrit range from 33 to 36% [1]. Because absolute or relative iron deficiency [2] is a common cause of relative unresponsiveness to Epo, concomitant clinical practice guideline statements have also been developed for iron administration [1]. The recommendations are for the performance of regular monitoring of the percentage transferrin saturation

2 148 D. N. Reddan et al. (TSAT) and the serum ferritin concentration and maintenance of these values P20% and 100 nguml, respectively. To encourage and assist in the achievement of the target haematocrit, quality improvement projects focused at dialysis facilities have been established and executed through individual dialysis unit providers and nephrologists. Based on results from several national registries and datasets, the haematocrit values for haemodialysis patients in the US have increased over the past 8 years [3,4]. The Centers for Medicare & Medicaid Services (CMS) National ESRD Core IndicatorsuClinical Performance Measures (CPM) Project is one such quality improvement project and reporting system. From 1994 to 1998, the mean haematocrit for a nationally representative sample of adult (P18 years old) haemodialysis patients increased from 31 to 34%. Despite the national improvement in mean haematocrit values, substantial regional differences continue to be observed, especially when reported as the percentage values greater than or equal to the Kidney Disease Outcomes Quality Initiative (KDOQI) benchmark value of 33% [1,4]. Haematocrit values are generally highest in the north-west portion of the US and lowest in the south-eastern US. The basis for this regional variation is debated, i.e. the role of patient-specific variables vs processes of patient care. Several patient-specific variables as well as provider actions influence the achieved haematocrit, the prescribed doses of Epo and the relationship of Epo dose to haematocrit. For example, increasing age, white race, male gender and greater serum albumin and creatinine concentrations have been independently associated with higher haematocrit values [4,5]. In turn, provider behaviours such as the dose and route of administration of Epo [6] and iron [7], the achievement of higher mean TSAT and higher doses of haemodialysis [5], and the use of high-flux haemodialysers [8] have been shown to influence haematocrit. The relative contributions of these individual and aggregate associates of achieved haematocrit and Epo dose have not been examined. It is hypothesized that geographic variation in provider behaviour relating to anaemia management principally accounts for the geographic differences in haemoglobin. We tested this hypothesis using nationally representative data collected on adult (P18 years old) in-centre haemodialysis patients. Subjects and methods The sampling strategy and design of the ESRD CPM project have been described in detail previously [9,10]. Briefly, all Medicare-eligible, adult ESRD patients receiving in-centre haemodialysis on 31 December 1998 were eligible for inclusion in the sample. A random sample of patients, stratified by the 18 ESRD Networks, was identified. The sample size provided a 95% confidence interval (CI) of "5% for ESRD Network-specific estimates. Patient demographic and clinical information was collected for the months October December Data gathered included the patient s prescribed Epo dose and route of administration, the prescribed route of iron administration, the first monthly-recorded haemoglobin (Hb) value, transferrin saturation (TSAT) value, serum ferritin concentration and serum albumin concentration as well as the laboratory technique used to assess the albumin concentration (bromocresol green (BCG) or bromocresol purple (BCP)). The single pool KtuV (calculated using the Daugirdas second generation formula) [11] was generated from first monthly pre-dialysis and post-dialysis blood urea nitrogen (BUN) concentrations, pre-dialysis and postdialysis body weight, and the delivered haemodialysis time at the session the BUN measurements were drawn. The KUf of the haemodialyser membrane was used to classify them as high or low flux. A KUf value P20 mlummhguh was considered high flux and a value -20 mlummhguh low flux. Patients were categorized as dialysing with high-flux dialysers if high-flux dialysers were used for two or more of the 3 months, or if a patient had no data recorded for dialyser type for 2 months but high-flux dialysers were used for one of the 3 months. If the dialyser type was not recorded for 1 month and mixed in the other 2 months, the patient was excluded from multivariable analyses involving dialyser type. Only monthly Hb values with parallel reported prescribed Epo doses were used for analysis. The prescribed Epo route was assigned based on data over 3 months. Patients prescribed intravenous (IV) Epo for P1 month and not prescribed any subcutaneous (SC) Epo were categorized as having been prescribed IV Epo. Patients prescribed both IV and SC Epo were excluded from multivariable analyses that used route of Epo administration as a variable. Patients were categorized as having been prescribed IV iron if they were prescribed IV iron at any time during the 3 months of data collection. For repeated measures, such as the dose of Epo, Hb and serum ferritin, the averages of the values for each patient recorded during the 3 month abstraction period were used for subsequent analysis. An EpouHb index for each patient was calculated by dividing the average per dialysis treatment weight-adjusted Epo dose by the average Hb over 3 months. Statistical analysis Bivariate analyses, including chi-square test, Student s t-test and hierarchical analysis of variance (ANOVA), were conducted to determine associations between variables. Mean values are presented as means"sd. Associations by race were limited to black and white race only, due to small numbers of patients of other races. Analyses that included the serum albumin concentration were restricted to the subset of patients with serum albumin values reported by the BCG assay only (84% of the 8336 patients). A twotailed P-value was considered significant. Linear correlations were performed to test for associations between Hb values and the variables of interest. Multivariable linear and logistic regression analyses were performed separately to adjust simultaneously for potential confounding variables and to identify independent variables associated with the Hb concentration. The linear regression analysis was then repeated with EpouHb index as the dependent variable. The initial linear regression model was developed with the mean Hb concentration as the dependent variable. The model was developed using a forward stepwise process with entry criteria of 0.1. The variables in the stepwise selection process included all demographic variables, diabetes mellitus as the cause of ESRD vs all other aetiologies

3 Variability in anaemia management combined, duration of haemodialysis, mean prescribed Epo dose and route of administration, prescription of IV iron, mean TSAT, mean serum ferritin concentration, dose of haemodialysis, vascular access type, dialyser KUf category, mean serum albumin concentration (BCG method only) and ESRD Network. Appropriate interactions were tested. A final model was subsequently generated using the variables noted to be significant in the stepwise analysis and other variables thought to be clinically relevant. Subsequently, a linear regression model was generated with the EpouHb index as the dependent variable. This model was generated using the same pool of independent predictor variables as the earlier model predicting Hb concentration. Potential predictor variables and other clinically relevant variables significantly associated with mean Hb concentration P11 gudl (P-0.01) in bivariate analyses were introduced into logistic regression models. Mean Hb concentration P11 gudl was the dependent variable of interest. Predictor variables assessed included patient demographics, diabetes as the cause of ESRD vs all other aetiologies combined, duration of haemodialysis, dose of dialysis (mean KtuV and mean urea reduction ration (URR) separately), mean prescribed Epo dose and route of administration, prescription of IV iron, mean TSAT, mean serum ferritin and albumin concentrations (BCG method only), type of vascular access, dialyser KUf category and ESRD Network. Interaction terms were tested. Linear regression analyses were performed using SAS software v (SAS Institute Inc., Cary, NC) and logistic regression analyses were performed using SPSS v. 8.0 [12]. Results There were 8336 patients (94% response rate) in the sample for analysis. Tied or parallel monthly Hb and prescribed Epo doses were available for 7592 (91%) of the patients. There were no significant differences in patient demographic characteristics between patients with tied monthly Hb and Epo values available and those without. The final sample closely resembled the entire Medicare ESRD population with its slight preponderance of men, a disproportionate number of whites and diabetes being the principal reported cause of ESRD [13] (Table 1). The median age was 62 years and the median duration of dialysis was 2.1 years. Most patients dialysed through a prosthetic graft or autologous fistula. The median delivered (single pool) KtuV was 1.4. Eighty-nine percent of patients were prescribed IV Epo; 10% of patients were prescribed SC Epo and 1% were prescribed both routes. Median Epo dose for IV and SC routes was 54.0 and 45.8 unitsukg, respectively. The distribution of prescribed Epo doses by route of administration is shown in Figure 1. The prescribed route of Epo administration varied significantly among ESRD Networks with 71 98% of the patients prescribed IV Epo (chi-square 374.5, P ) (Figure 2). Mean prescribed Epo dose also varied significantly by ESRD Network, ranging from 56.1 to 90.7 unitsukg (F-statistic 9.64, P ). The mean Hb for all patients was 11.1"1.2 gudl. There was no significant difference in mean Hb Table 1. Patient characteristics Variable n (%) Number of patients 8336 (100) Gender a Male 4449 (53) Female 3878 (47) Race White 4167 (50) Black 3145 (38) Otheruunknown 1024 (12) Hispanic ethnicity 994 (12) Reported primary cause of ESRD Diabetes mellitus 3423 (41) Hypertension 2127 (26) Glomerulonephritis 1027 (12) Other 1759 (21) Vascular access a Autologous arteriovenous fistula 2180 (26) Prosthetic arteriovenous graft 4261 (51) Percutaneous tunnelled catheter 1564 (19) Intravenous Epo 6703 (89) Mean ("SD), median Mean age (years) 60.2 ("15.2), 62.0 Duration of ESRD therapy (years) 3.3 ("3.7), 2.1 Mean albumin (BCG method) (gudl) 3.8 ("0.4), 3.8 Mean KtuV 1.4 ("0.27), 1.4 Mean URR (%) 68.3 ("7.8), 69.4 EpouHb index (unitsukgudoseugudl) 6.44 ("5.62), 4.76 Mean Hb (gudl) 11.1 ("1.2), 11.2 a Numbers do not add up to total n secondary to missing data. 149 concentration by prescribed route of Epo administration; 11.1"1.2 gudl for IV Epo and 11.1"1.2 gudl for SC Epo (Ps0.637). The frequency distribution of Hb values by prescribed route of Epo administration for the US is illustrated in Figure 3 and this figure demonstrates that Hb values were normally distributed. Hierarchical analysis of variance of Hb concentrations across all ESRD Networks revealed significant non-random variation (i.e. variation not due to chance), with mean Hb values ranging from 10.8 to 11.4 gudl (F-statistic 4.74, P-0.001). The proportion of patients across Networks with Hb values P11.0 gudl ranged from 48 to 66% (chi-square 77.8, P-0.001). Fig. 1. Distribution of Epo dose by route of administration. Closed bars, SC; open bars, IV. Median Epo dose was 45.8 and 54.0 unitsukg for SC and IV Epo, respectively.

4 150 D. N. Reddan et al. Fig. 2. Percentage of patients receiving IV Epo by ESRD Network. Grey bar, overall percentage across all ESRD Networks. P Fig. 3. Haemoglobin distribution. Closed squares, SC; open squares, IV. Mean Hb 11.07"1.21 gudl for SC and 11.09"1.21 gudl for IV. Figure 4 describes the details of the relationship between Epo dose, iron utilization and mean Hb. Mean Hb achieved was lower across increasing quartiles of Epo dose and this relationship was similar among patients prescribed and not prescribed IV iron. There was significant variability among ESRD Networks in the percentage of patients prescribed IV iron, ranging from 47 to 70% across Networks (chisquare 173.0, P ). There was also significant variability across ESRD Networks with regard to mean TSAT, ranging from 19.2 to 32.5% (F-statistic 21.45, P ) and there was considerable variability among ESRD Networks in terms of use of high-flux haemodialysers (chi-square 575.1, P-0.001). At one extreme is ESRD Network 3, in which only 31% of patients were dialysed with high-flux dialysers in contrast to ESRD Network 18 in which 76% were dialysed with high-flux dialysers. The other ESRD Networks had percentages intermediate to these two. For the final linear model, 72% of the patients were included (5475 of 7592). The variables found to be significant in the stepwise selection process having a positive relationship with Hb concentration included increasing age, male gender, lower mean Epo doses, higher mean TSAT, lower mean serum ferritin concentration, higher mean serum albumin concentrations, greater mean haemodialysis doses and receiving care in ESRD Networks 4 or 12. The parameter estimates and significance levels can be seen in Table 2. There were no significant interaction terms found during the linear modelling procedure. Seventy-two percent of the patients (5454 of 7592) were included in the final logistic regression model predicting a mean Hb P11 gudl. The significant predictors in the final logistic regression model are depicted in Table 3. Increasing age, male gender, duration of haemodialysis -1 year, lower mean doses of Epo, higher mean TSAT, higher mean KtuV, higher mean serum albumin and receiving care in ESRD Network 4 (all P-values -0.01) were significantly associated with experiencing a mean Hb P11 gudl. There were no significant interaction terms found during the logistic regression procedure. The mean EpouHb index for the population studied was 6.44"5.62 unitsukgudoseugudl. When a multivariable linear regression model was generated with EpouHb index as the dependent variable, male gender, diabetes as cause of ESRD, older age, higher transferrin saturation, higher albumin concentrations and Table 2. Significant variables in the final linear model with Hb as the dependent variable Variable Parameter P-value F-value type III SS Age (years) Male gender Mean serum albumin (gudl) a Mean TSAT Mean serum ferritin (nguml) Mean Epo dose (unitsukg) Mean dose of dialysis (KtuV) ESRD Network ESRD Network Fig. 4. Mean Hb by quartile of Epo dose with and without parenteral iron administration. a Modelling was conducted on the subset of patients with serum albumin values tested by the BCG method only. Factors entered into the model but found to be insignificant included race (black vs white only), duration of haemodialysis as renal replacement therapy, prescription of IV iron, prescription of IV Epo, diabetes mellitus as a cause of ESRD vs all other causes combined, type of vascular access, dialyser KUf category and other ESRD Networks.

5 Variability in anaemia management Table 3. Final logistic regression model predicting a mean Hb P11 gudl a Variable OR (95% CI) P-value Age (years) 1.01 (1.006, 1.014) Male gender 1.20 (1.08, 1.37) Duration of dialysis (years) (2q yearssreferent) (0.60, 0.89) (1.24, 1.77) (0.83, 1.12) Mean Epo dose (unitsukg) (quartile 4 (highest)sreferent) Quartile (3.09, 4.34) (2.00, 2.76) (1.43, 1.97) Mean TSAT (%) 1.01 (1.009, 1.02) Mean serum ferritin (nguml) (0.9996, ) Mean KtuV 1.81 (1.43, 2.29) Mean serum albumin (gudl) 2.40 (2.05, 2.81) ESRD Network ( ) a Modelling was conducted on the subset of patients with serum albumin values tested by the BCG method only. Factors entered into the model but found to be insignificant included race (black vs white only), ethnicity (Hispanic vs non-hispanic), diabetes mellitus as a cause of ESRD vs all other causes combined, prescription of IV Epo, prescription of IV iron, type of vascular access, dialyser KUf category and other ESRD Networks. receiving care in ESRD Networks 11 or 15 were associated with lower EpouHb index (Table 4). Prescription of IV iron and IV Epo and receiving care in Network 4 were associated with higher EpouHb index. Discussion Based on a nationally representative sample of ESRD patients receiving in-centre haemodialysis, significant regional differences were observed in both mean Hb concentration and the percentage of patients with Table 4. Predictors of EpouHb index Variable Parameter P-value F-value type III SS Male gender Diabetes mellitus a Age (years) TSAT Prescription of IV Epo Prescription of IV iron Serum albumin (gudl) b ESRD Network ESRD Network ESRD Network a Diabetes as cause of ESRD. b Modelling was conducted on the subset of patients with serum albumin values tested by the BCG method only. Factors entered into the model but found to be insignificant included duration of dialysis (years), race (black vs white only), dialyser KUf category, serum ferritin, dose of dialysis (KtuV) and other ESRD Networks. 151 mean Hb exceeding the recommended minimum value of 11.0 gudl [1]. Previous work has suggested that both patient-specific variables and provider practices can affect the final Hb concentration [5]. In addition to confirming observations about non-modifiable factors [5], the additional information provided by this analysis is a clarification of the influence of modifiable factors on achieved Hb targets. The finding that between 46 and 58% of patients in individual Networks had Hb concentrations equal to or higher than 11 gudl suggests that a slightly higher proportion of patients reached target Hb levels than the 43.6% reported at the end of the European Survey on Anaemia Management (ESAM) study in France. The ESAM study also reported that unsatisfactory prescription and monitoring of the effects of iron supplementation accounted for a large part of the under-achievement of Hb targets [14]. We describe significant variability in Hb accounted for by patient-specific and clinical practice variables. Patient-specific variables included age, gender and albumin. Variables that were considered modifiable clinical practice variables included mean weekly Epo dose and mean dialysis dose. Variables describing iron stores, TSAT and serum ferritin could also be considered modifiable clinical practice variables and were explanatory. However, the complex interactions between measures of iron stores and serum ferritin, in particular with inflammation, make it difficult to define them as completely modifiable. The absence of specific iron dosing information is a limitation of these analyses and despite the observation of non-random regional differences in prescription of IV iron, the prescription of IV iron was not found to be a significant predictor of Hb. The positive association of TSAT with Hb, however, suggests the possibility of an association of increased Hb with successful iron repletion. That route of Epo administration was predictive of EpouHb index but not predictive of Hb probably reflects appropriate adjustment by nephrologists for the reduced efficacy of IV relative to SC Epo. That higher Epo doses predicted lower mean Hb could possibly be explained by indication bias, in that relatively higher doses of Epo may be given to poorly responsive Epo-resistant patients. This analysis bears the typical limitations associated with observational analyses. Bias by indication is likely for some elements, such as the prescription of parenteral iron and SC Epo. Facility-specific data describing practice patterns or other characteristics were not available. Despite these limitations, these findings are strengthened by the large number of subjects and observations, the use of an established and validated nationally representative sample, confirmation of conventional associations using this dataset and the use of complimentary multivariable analyses with comparable results. We also used this dataset to examine the relationships between Epo doses and achieved Hb concentration, by determining predictors of the EpouHb index.

6 152 D. N. Reddan et al. Previous analyses have reported reduced Epo doses or enhanced responsiveness for patients receiving SC rather than IV Epo [15,16] and IV rather than oral iron [17,18]. These data confirm that the use of SC Epo is associated with more efficient Epo dosing. The finding that higher TSAT also leads to reduced EpouHb index is also consistent with these earlier findings. The finding that prescription of IV iron predicted a higher EpouHb index could be explained by indication bias and may have been less evident had iron dosing data been available. Erythropoietin dosing on a population basis has been increasing in recent years and a relationship has been established between increased doses and increased Hb targets [19]. This observation obviously has substantial cost implications. These findings provide evidence to support the development of focused quality improvement projects that rely upon the continued development and application of clinical practice guidelines and performance measures. By encouraging administration of SC Epo, more iron use and greater attention to evaluating Epo hyporesponsiveness, Epo effectiveness may be enhanced. Moreover, Hynes et al. [20] suggest that use of SC Epo could lead to substantial cost savings. More importantly, further exploration of the relative ineffectiveness of existing quality-improvement projects already implemented in many health systems across the world is warranted. Possible explanations for such ineffectiveness include provider disbelief of recommendations, inadequate CPM dissemination, incomplete prospective monitoring and feedback of results or insufficient systems to support anduor implement changes [21]. The absence of a clear correlation between reimbursement and quality of care could also be playing an important role [22]. Linking reimbursement with quality is one example of a possible ecological health system intervention that could lead to improvement in quality of care. Further refinements in the CPM process that could lead to improved interpretation and implementation of CPMs might be achieved by the expansion of the process to collect data at a dialysis-unit level, in addition to the current patient-specific data collection. In conclusion, we have shown that regional mean Hb levels vary considerably across the US. The variation in Hb is explained by both non-modifiable factors and modifiable clinical practice-derived variables. Considerable opportunity for improvement exists with respect to the management of anaemia by enhanced implementation of appropriate clinical performance measures. Acknowledgements. These data were presented previously in abstract form at the American Society of Nephrology meeting 2001 in San Francisco. References 2. Macdougall IC, Cavill I, Hulme B et al. Detection of functional iron deficiency during erythropoietin treatment: a new approach. BMJ 1992; 304: Cotter DJ, Thamer M, Kimmel PL, Sadler JH. Secular trends in recombinant erythropoietin therapy among the U.S. hemodialysis population: Kidney Int 1998; 54: Frankenfield D, Johnson CA, Wish JB, Rocco MV, Madore F, Owen WF. Anemia management of adult hemodialysis patients in the US results: from the 1997 ESRD Core Indicators Project. Kidney Int 2000; 57: Madore F, Lowrie EG, Brugnara C et al. Anemia in hemodialysis patients: variables affecting this outcome predictor. J Am Soc Nephrol 1997; 8: Kaufman JS, Reda DJ, Fye CL et al. Subcutaneous compared with intravenous epoetin in patients receiving hemodialysis. Department of Veterans Affairs Cooperative Study Group on Erythropoietin in Hemodialysis Patients. N Engl J Med 1998; 339: Fudin R, Jaichenko J, Shostak A, Bennett M, Gotloib L. Correction of uremic iron deficiency anemia in hemodialyzed patients: a prospective study. Nephron 1998; 79: Depner TA, Rizwan S, James LA. Effectiveness of low dose erythropoietin: a possible advantage of high flux hemodialysis. ASAIO Transactions 1990; 36: M223 M Health Care Financing Administration. Annual Report ESRD Clinical Performance Measures Project. Department of Health and Human Services Health Care Financing Administration Office of Clinical Standards and Quality, Baltimore, MD, December Frankenfield DL, McClellan WM, Helgerson SD, Lowrie EG, Rocco MV, Owen WF, Jr. Relationship between urea reduction ratio, demographic characteristics, and body weight for patients in the 1996 National ESRD Core Indicators Project. Am J Kidney Dis 1999; 33: Daugirdas JT. Second generation logarithmic estimates of single-pool variable volume KtuV: an analysis of error. JAm Soc Nephrol 1993; 4: Norusis MJ. SPSS for Windows, Advanced Statistics, Release 8.0. Chicago: Prentice Hall; U.S. Renal Data System. USRDS 1999 Annual Data Report. National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, Jacobs C. Treatment of anemia in dialysis patients in France ( ) (Study by ESAM-France). Nephrologie 2002; 23: Besarab A. Optimizing epoetin therapy in end-stage renal disease: the case for subcutaneous administration. Am J Kidney Dis 1993; 22: Paganini EP, Eschbach JW, Lazarus JM et al. Intravenous versus subcutaneous dosing of epoetin alfa in hemodialysis patients. Am J Kidney Dis 1995; 26: Macdougall IC, Tucker B, Thompson J, Tomson CR, Baker LR, Raine AE. A randomized controlled study of iron supplementation in patients treated with erythropoietin. Kidney Int 1996; 50: Fishbane S, Frei GL, Maesaka J. Reduction in recombinant human erythropoietin doses by the use of chronic intravenous iron supplementation. Am J Kidney Dis 1995; 26: Coladonato JA, Frankenfield DL, Reddan DN et al. Trends in anemia management among US hemodialysis patients. J Am Soc Nephrol 2002; 13: Hynes DM, Stroupe KT, Greer JW et al. Potential cost savings of erythropoietin administration in end-stage renal disease. Am J Med 2002; 112: Cabana MD, Rand CS, Powe NR et al. Why don t physicians follow clinical practice guidelines? A framework for improvement. J Am Med Assoc 1999; 282: Powe NR, Thamer M, Hwang W et al. Cost-quality tradeoffs in dialysis care: a national survey of dialysis facility administrators. Am J Kidney Dis 2002; 39: NKF-KuDOQI Clinical Practice Guidelines for Anemia of Chronic Kidney Disease: Update Am J Kidney Dis 2001; 37: S182 S238 Received for publication: Accepted in revised form: