Changing hemodialysis thresholds for optimal survival

Kidney International, Vol. 59 (2001), pp. 738–745 Changing hemodialysis thresholds for optimal survival LYNDA ANNE SZCZECH, EDMUND G. LOWRIE, ZHENSHENG LI, NANCY L. LEW, J. MICHAEL LAZARUS, and WILLIAM F. OWEN, JR. Duke Institute for Renal Outcomes Research and Health Policy, Duke Clinical Research Institute, Division of Nephrology, Duke University Medical Center, Durham, North Carolina, and Fresenius Medical Care, North America, Lexington, Massachusetts, USA Numerous studies have demonstrated an association between the amount of hemodialysis and mortality among patients with end-stage renal disease (ESRD) [1–8]. In general, patient mortality is higher when the hemodialysis “dose” is low and is lower when the hemodialysis dose is high. The urea reduction ratio (URR) is a commonly used measure of hemodialysis dose and is based on the fractional reduction of blood urea nitrogen (BUN) concentration during a single hemodialysis treatment. It is calculated by dividing the fall of BUN (predialysis minus postdialysis BUN) by the predialysis and is expressed as a percentage [1, 9–11]. Retrospective studies of mortality for ESRD patients suggest that the odds of death increase progressively as the URR falls below 60 to 65% [1–3]. Such findings and a professional consensus have led three national organizations, including the Health Care Financing Administration (HCFA), the principal payer of dialysis services, to advocate a URR of 65% as the threshold for “adequate” hemodialysis and to profile providers accordingly [9, 10, 12–14]. Using URR as a measure of hemodialysis dose, substantial improvement in the amount of hemodialysis has occurred in recent years. As reported in the 1998 National ESRD Core Indicators Report, a profile of patientspecific dialysis practices, the mean URR for the United States increased from 62.7% (deemed inadequate) to 68.0% (deemed adequate) from 1993 to 1997 [14]. Although mean URR provides a gauge of hemodialysis adequacy at the population level, the proportion of patients whose URR is $65% is of greater significance for optimizing individual patient survival. Using this clinical performance benchmark, the percentage of patients receiving a URR $65% rose from 43% in 1993 to 72% in 1997 [14]. The aforementioned outcome studies characterizing the association between the amount of hemodialysis and patient mortality utilized data sets from before 1994 [1–8]. However, the applicability of these studies may be limited because of evolving demographics of the hemodialysis population and the diffusion of newer hemo- Changing hemodialysis thresholds for optimal survival. Background. The urea reduction ratio (URR), a measure quantitating solute removal during hemodialysis, is the fractional reduction of the blood urea concentration during a single hemodialysis treatment. The URR is the principal measure of hemodialysis dose in the United States. Based on studies of patients dialyzed prior to 1994, a minimum URR value of 65% was recommended to optimize survival. Because of new hemodialysis technologies and evolving demographics of the hemodialysis population, the relationship between the amount of hemodialysis and mortality was examined in contemporary cohorts. Methods. This retrospective cohort included .15,000 patients per year receiving hemodialysis during 1994 through 1997. Each patient’s URR was averaged for the three months prior to the beginning of each year. Mortality odds ratios were calculated for patients by URR. To determine the URR value above which no further improvement in mortality was seen (“threshold”), spline functions were tested in logistic regression models, both unadjusted and adjusted for case mix measures. The strength of fit for URR, defined by a range of candidate thresholds from 55 to 75%, was evaluated in increments of 1% for each year using spline functions. Results. The median URR was 63.2, 65.4, 67.4, and 68.1% for 1994 through 1997, respectively. The median length of hemodialysis treatments increased only six minutes from the beginning to the end of the period of analysis. Using spline functions, the threshold URR values were 61.1, 65.0, 68.0, and 71.0% for 1994 through 1997 in models adjusted for case mix. The ratio of median URR to URR threshold decreased from 1.03 in 1994 to 0.97 in 1997. Conclusions. From 1994 to 1997, the median URR and the URR threshold for mortality benefit increased. Although an increased need in the amount of hemodialysis may be a consequence of changes in patients’ demographic characteristics, the likely explanation(s) is a change in the dialysis procedure and/or blood sampling favoring higher URR values without changing the amount of dialysis provided. The recommended minimum URR of 65% appears to be too low to confer an optimal mortality benefit in the context of current practices. Key words: end-stage renal disease, dialysis outcome, Medicare, urea reduction ratio, adequate hemodialysis. Received for publication February 3, 2000 and in revised form June 13, 2000 Accepted for publication August 11, 2000  2001 by the International Society of Nephrology 738 739 Szczech et al: Hemodialysis dose and optimal survival dialysis technologies. For example, the prevalent hemodialysis population includes more patients who are older, who have diabetes mellitus, and who are racial and ethnic minorities [15]. Additionally, prevalent hemodialysis practices have changed with the increased use of bicarbonatebuffered dialysis solutions, high-flux and biocompatible dialysis membranes, and reprocessed dialyzers [14]. The effects of these demographic changes and the diffusion of newer technologies on the relationship between the amount of hemodialysis and mortality have not been evaluated. This study was undertaken to examine the validity of the previously demonstrated relationship between amount of hemodialysis and mortality in the setting of these secular trends. METHODS Data were taken from the routine analytical files of Fresenius Medical Care-North America, Inc. (FMC, Lexington, MA, USA) for the calendar years 1994 through 1997 [1, 5]. The final sample was comprised of all patients receiving three times weekly hemodialysis who were prevalent on January 1 of each year and either lived the entire year on dialysis or died. Dropouts were not included in the sample studied. The URR for each patient was calculated by 100 3 (predialysis BUN 2 postdialysis BUN) 4 predialysis BUN. Prior to the release of the National Kidney Foundation’s Dialysis Outcomes Quality Initiative on Hemodialysis Adequacy in 1997, no specific protocol was specified for the sampling of blood for the measurement of postdialysis BUN. Fresenius endorsed a minimum URR or Kt/V of 65% (URR) or 1.2 [Kt/V (single pool)], respectively, but this was not mandated. The URR and serum albumin concentrations for the last three months of each prior year were averaged. All measurements were determined in a single laboratory (LifeChem Clinical Laboratories, Rockleigh, NJ, USA). Only patients with complete demographic and laboratory data were included in the cohorts. The patients’ demographic and clinical features were compared using the x2 test and the x2 test for trend across years. For patients with a URR ,60%, the odds ratios for mortality were calculated in URR increments of 5%; for patients with a URR $60%, the URR increments were 2.5%, and patients whose URR was .80% comprised the reference group. Odds ratios were calculated in univariate and multivariable analyses, the latter controlled for age, gender, race, and diabetes mellitus (case mix variables). Risk curves of URR versus log transformed odds ratios were plotted for each year. To describe the relationship between URR and mortality odds, spline functions were used [16, 17], as they have been used in prior analyses of dialysis adequacy and mortality [2, 18]. Spline functions mathematically describe complex relationships between continuous vari- ables, when a single analytic description, formula, or function is not valid over the entire range of values. Spline functions transform these relationships into small segments of relatively simple functions connected at juncture points or “knots.” Initially, two-segment splines were used to describe the relationship between threshold URR and mortality odds. To identify a threshold, this analysis assumes that the relationship between URR and mortality has two discrete segments. The first segment demonstrates an inverse relationship between URR and mortality. This inverse relationship continues to a threshold value of URR, which marks the beginning of the second segment, above which no further mortality benefit is observed despite an increasing URR. To assess which URR value defined the spline with the best fit, URR was transformed using each candidate URR threshold by this equation: (URR 2 t)1 5 50,URR 2 t, URR ^ t URR , t where the (URR 2 t)1 term evaluates the difference between the measured URR and the candidate threshold (t, from 55 to 75% in 1% increments) for all URR less than each candidate threshold. The “1” subscript indicates that the term has been truncated during the transformation process performed in this analysis. All URR values $t were assigned a value of 0. For each year, separate logistic regression models estimated the association between mortality and (URR 2 t)1 using each of the 21 candidate threshold URR values [19]. This association was modeled in both a univariate and multivariable manner. For each year, the strength of the association between URR and mortality using each candidate threshold was estimated in separate logistic regression models, unadjusted and adjusted for case mix. The strength of fit for the association between mortality and URR as described by each threshold URR was evaluated using the x2 statistic associated with the (URR 2 t)1 term. The larger the x2 statistic associated with the spline for each threshold URR, the greater the amount of variation in outcome explained by URR using that specific threshold. The x2 statistic for URR as described by each candidate threshold was plotted against the threshold for each year. This analysis was repeated using Kt/V transformed from URR by taking the negative loge of one minus the URR divided by 100 [11, 20]. The routine analytical files of FMC for calendar years 1994 through 1997 do not contain data on the intradialytic ultrafiltration volume. All P values are two-sided, and confidence intervals are 95%. Analyses were performed using SAS (version 6.08; SAS Institute Inc., Cary, NC, USA). 740 Szczech et al: Hemodialysis dose and optimal survival Table 1. Description of final patient cohorts Characteristic Age years Mean Median Female % Caucasian race % Diabetes mellitusa % Height cm Mean Median Weight kg Mean Median Wt/Ht kg/cm Mean Median Albumin g/dL Mean Median Dialysis length hours Meana Median URR % Meana Median Vintage years Mean (SD) Median 1994 N 5 17,141 1995 N 5 20,785 1996 N 5 15,155 1997 N 5 15,197 60.0 6 14.8 62.3 49.3 49.7 39.1 60.2 6 15.0 62.5 49.1 46.1 42.4 59.5 6 14.9 62.0 49.9 46.9 44.9 59.9 6 14.9 62.0 50.0 50.0 47.2 168.0 6 10.8 167.6 167.7 6 11.0 167.6 165.6 6 12.3 165.0 165.5 6 12.8 165.0 70.9 6 17.9 68.5 70.9 6 18.1 68.3 71.5 6 18.2 69.0 72.0 6 18.4 69.3 0.42 6 0.10 0.41 0.42 6 0.10 0.41 0.43 6 0.10 0.41 0.44 6 0.11 0.42 3.81 6 0.42 3.83 3.85 6 0.37 3.87 3.90 6 0.38 3.93 3.85 6 0.38 3.88 3.38 6 0.48 3.50 3.37 6 0.49 3.50 3.43 6 0.46 3.46 3.48 6 0.47 3.46 65.1 6 7.8 65.4 66.7 6 7.3 67.4 67.3 6 7.5 68.1 3.7 6 3.8 2.5 3.7 6 3.6 2.5 3.5 6 3.6 2.4 62.29 6 8.2 63.2 n/a n/a Plus-minus values are means 6 SD; diabetes mellitus describes the presence of diabetes mellitus as a diagnosis or a comorbid condition; vintage refers to the number of years since starting dialysis; serum albumin concentrations were drawn predialysis. a P , 0.0001 for values 1994 to 1997 RESULTS The patients’ demographic features, anthropometric attributes, and distribution of selected laboratory tests are provided in Table 1. Their mean and median ages and composition by gender and race were stable from 1994 to 1997. Also, anthropometric attributes, described by height, weight, and weight/height ratio, were stable. However, the proportion of patients with diabetes mellitus rose over the period of observation from 39.1% in 1994 to 47.2% in 1997 (P , 0.0001). The mean URR increased during the period of observation from 62.9% in 1994 to 67.3% in 1997 (P , 0.0001). The average length of hemodialysis treatments increased only six minutes (P , 0.0001). The serum albumin concentration was stable from 1994 to 1997. The association between log odds ratios for mortality and URR adjusted for age, race, gender, and diabetes mellitus for each year is presented in Figure 1. During each year, as the URR increased, the log odds for mortality decreased. The odds of death declined to an apparent URR threshold, above which increasing URR values did not reduce death risk further. A similar relationship was revealed in the absence of adjustment for case mix measures. To quantitatively evaluate the perception that the URR threshold increased from 1994 to 1997, spline functions were fit in logistic regression models using the range of URR candidate thresholds. Figure 2 plots each URR threshold and the associated x2 statistic for the case mix adjusted model. The larger the x2 statistic, the stronger the association between the individual threshold URR value and mortality. For each year, the curves increased as a smooth, monotonic function to a peak and subsequently decreased in a similar manner. The URR threshold values with the largest x2 statistic were 61, 65, 68, and 71% for 1994, 1995, 1996, and 1997, respectively. For the unadjusted analyses, threshold URR values were 60, 64, 66, and 69%, respectively, for the same years. These findings suggest an increase in apparent threshold for hemodialysis adequacy during this period. In additional analyses, patients with and without diabetes mellitus were evaluated separately. For each year, the plots of URR threshold values and their associated x2 statistic revealed no consistent differences between diabetics and nondiabetics (data not shown). To assess the effect of different modeled relationships between URR and mortality on the increasing threshold for adequacy in dialysis dose, complimentary analyses were performed using two- and three-segment splines. One model assumes that the reduction in death risk declines to a nadir, and any additional increase in URR above this threshold results in a higher mortality (Vshaped relationship). The second model assumes a re- Szczech et al: Hemodialysis dose and optimal survival 741 Fig. 1. Risk curves for urea reduction ratio (URR) and odds ratio of mortality. The logarithm of odds ratio for mortality is plotted against URR stratified by year. verse J-shaped relationship. The initial decline is separated from the terminal increase in mortality odds by interposing a flat middle segment, during which no change is seen with increasing URR. Using either assumption, the threshold URR values for each year were unchanged from the results obtained with two-segment spline analyses (data not shown). The relationship between the median URR and URR threshold by year is illustrated in Figure 3. Both measurements increased from 1994 through 1997. However, the ratio of the median URR to threshold URR decreased from 1.03 in 1994 to 0.97 in 1997. The slope of this relationship indicates that the URR threshold increased faster than the median URR. This analysis was repeated using Kt/V extrapolated from URR as the measurement of dialysis dose. A similar increase in threshold dose was demonstrated over the period of observation (data not shown). DISCUSSION The URR remains the preponderant measure of dialysis dose in the United States [14]. This retrospective study of large cohorts of hemodialysis patients demonstrates that the conventional minimum URR of 65%, recommended to minimize death risk [9, 10, 12–14], may be too low for current hemodialysis practice. The threshold for the 1994 data set of patients was 61%. Assuming variability of 2.4% for a URR determination [21], a minimum URR of 65% was adequate. However, by 1997, the threshold URR value had increased to 71%. This discordance between the URR value associated with optimal mortality benefit (“threshold”), and the conventional benchmark of 65% suggests that some patients may be placed at excessive death risk. Moreover, since 1994, the URR threshold has increased more than the median URR. This secular trend suggests that improvements in mortality may have not been realized fully. These analyses assume that the relationship between URR and mortality has two discrete segments. The first segment has an inverse relationship of decreasing mortality with increasing URR, until a threshold URR value is achieved, which marks the beginning of the second segment above which no further mortality benefit is obtained. Such a relationship has been described by several investigators [1, 3, 5, 22, 23]. Alternative models were derived based on descriptions suggesting that the relationship between URR and mortality may be V or reverse J shaped [2, 3, 24]. The URR threshold values were not changed by modeling using these different assumptions. The increment in URR threshold values was associated temporally with an increase in median URR values. Hypotheses to explain the changing relationship between hemodialysis dose and mortality include temporal changes in patient demographics, the hemodialysis procedure, and/or prevalent blood sampling techniques. For example, if dissimilar demographic groups have differential needs for hemodialysis [4, 5], a change in the case mix of prevalent patients may result in an increased proportion of patients with a need for increased hemodialysis. Some have posited that patients with diabetes mellitus require greater doses of hemodialysis to achieve similar survival [4]. While the proportion of patients with diabetes mellitus in the sample increased between 1994 and 1997, the URR thresholds for patients with and without diabetes mellitus increased over time and were not different by diabetic status. Furthermore, the preponderance of evidence fails to support a need for increased hemodialysis for patients with diabetes mellitus [1, 2, 5]. 742 Szczech et al: Hemodialysis dose and optimal survival Fig. 2. Strength of association between URR and mortality. The x2 statistic for each candidate URR threshold in the logistic regression model of mortality is plotted against the candidate URR threshold stratified by year. Fig. 3. Temporal trends for the prevalent median URR (x-axis) plotted against URR threshold (y-axis) values. Nutrition is a major mortality predictor for ESRD patients, statistically more powerful than the amount of hemodialysis [1, 4, 5, 24]. Recent data suggest that patients’ anthropometric attributes (weight, body surface area, and body mass index) and the volume of urea distribution (total body water) may be surrogate markers for nutritional status among ESRD patients. Higher measurements are associated with improved survival [5, 18, 23, 25]. Lower anthropometric attributes result in a smaller urea distribution volume and typically greater amounts of dialysis [5, 26] but are paradoxically associated with higher death risks [23]. It is unlikely that the findings herein are explained by a trend toward decreasing anthropometric measures among dialysis patients. Neither the cohorts’ anthropometric attributes nor their serum albumin concentrations, a laboratory surrogate of nutrition and/or inflammation, changed from 1994 to 1997 [1, 5, 23, 27, 28]. Another putative explanation to account for the increased dialysis dose thresholds is that the URR does not measure the removal of the critical uremic toxin. A small body of recent evidence suggests that URR does not account for the removal of all toxins associated with mortality, such as larger molecular weight solutes [29]. Therefore, a higher URR would be needed to achieve removal of these toxins. This putative mechanism also seems unlikely because hemodialyzer technology and the type of dialyzer used have evolved. Prevalent hemodialyzers have better solute removal profiles for large molecular weight solutes [14]. If the composition of the cohort with significant residual renal function remained the same during each year, this would introduce another source of variance into the association between dialysis dose and mortality biasing our ability to determine a difference between years toward the null. Alternatively, if the composition of the cohort with significant residual renal function decreased in number with each advancing year, the association between dialysis dose and mortality may be affected shifting the curves to the right. However, as residual renal function declines with increasing time since the initiation of dialysis, the use of only prevalent patients minimizes the potential impact of this limitation. Supporting our ability to compare the cohorts is that the length of time since initiation of dialysis (vintage) had not changed for the years 1995 through 1997. A systematic change in the dialysis technique and/or in the timing for drawing blood samples for the URR could also account for these observations. Although the rate of transfer of urea between intracellular and extracellular water is rapid, inequality in urea concentration Szczech et al: Hemodialysis dose and optimal survival may develop between the two compartments during rapid hemodialysis. When this occurs, urea that has been effectively sequestered intracellularly diffuses back into the extracellular compartment (“urea rebound”). This process begins immediately at the end of hemodialysis and lasts over 30 minutes. Therefore, as blood is sampled over the minutes after the end of hemodialysis, the postdialysis BUN will rapidly increase, and the calculated URR becomes smaller. The converse is also true. The closer the postdialysis BUN is drawn to the end of dialysis, the lower its value and the higher the calculated URR. By not accounting for the sequestered amount of urea, the URR becomes less reflective of total urea removal [10]. Secular changes in the hemodialysis technique, such as the reported increase in the use of higher flux dialysis membranes that result in increased dialysis efficiency [14], may cause more urea rebound. If the use of higher flux membranes is associated with greater rebound, measured URR would be higher, but the amount of hemodialysis delivered would be relatively little changed. Mortality relationships would shift toward higher URR values. Higher URR values would compensate for an overestimation of total urea removal. An alternative, but less frequently used measure of hemodialysis dose, Kt/V, is based on pharmacokinetic theory and attempts to normalize the dose of hemodialysis for the urea distribution volume [10]. It is unclear whether the use of Kt/V, instead of URR as the predominant measure of hemodialysis dose, will minimize the apparent need for higher dialysis doses by its improved kinetic accuracy. This issue may be clarified by the findings from the National Institute of Diabetes, Digestive, and Kidney Diseases Hemodialysis Study, which is based on equilibrated Kt/V measurements and will be available within the next two years [30]. Moreover, the randomized, controlled study design will minimize biases inherent to retrospective studies. Finally, substantial pressure has been exerted to improve dialysis outcomes by increasing the amount of hemodialysis [31–34]. Both URR and Kt/V are calculated centrally by Fresenius’ laboratory services; the aggregate results are reviewed centrally and facility and patient-specific results reviewed and acted upon locally within individual dialysis units. For example, a highly publicized national discussion proposed linking dialysis facility reimbursement to the quality of hemodialysis services measured using URR [35]. In 1993, a longitudinal, nationwide cohort study revealed that only 43% of the patients had a URR $65% [14]. There resulted an urgency to improve URR values without explicit instruction regarding strategies to achieve this goal. Recognizing an opportunity and need to improve patient outcomes, HCFA restructured its method for national quality improvement in 1994 [14, 36]. Although this was a population-based initiative, it was intended to serve as 743 a stimulus for more focused monitoring, profiles, and investigations by the regional peer review organizations, corporate dialysis providers, and individual dialysis units. In the context of substantial pressure to improve dialysis doses, it is noteworthy that two 1995 surveys of greater than 200 hemodialysis centers in the United States and Canada reported over 20 different methods for sampling blood to measure postdialysis BUN [37, 38]. As mentioned earlier, a trend toward the use of more efficient dialyzers has occurred that would favor increased urea rebound [29]. This change in dialyzer type superimposed on wide variability in methods for sampling postdialysis BUN could result in higher prevalent URR values. Moreover, higher URR values achieved in this or another manner would be welcome by the dialysis community as evidence of betterment. However, in this situation, the URR would be accompanied by a rise in the threshold for hemodialysis adequacy. If we assume that these findings are the consequence of variability in postdialysis BUN sampling, this is the first longitudinal analysis to examine the potential quantitative effect of this variability on patient mortality. Recent evidence has suggested that K 3 t is a valid outcome-based measure of hemodialysis dose that is less influenced by anthropometric attributes, which have outcome-associated properties of their own [5, 25]. Among a patient cohort receiving hemodialysis during 1994, their mortality risk profile improved until K 3 t was 41.2 to 47.3 L per treatment in men and 37.0 to 41.2 L per treatment in women [25]. There were no additional improvements in mortality risk for either group as K 3 t increased above these values. Analysis of a patient cohort receiving dialysis during 1998 demonstrated that a K 3 t of at least 50 L per treatment was required to achieve maximal death risk reduction [39]. This comparison suggests that regardless of the measure of dialysis dose, an increasing standard for dialysis adequacy remains evident. Moreover, it underscores that the measurement of dialysis dose rather than the method of calculation have evolved. The same is true for dialysis dose calculated as Kt/V. We acknowledge that the transformation formula used in this analysis does not account for convective solute clearance [11], so it is not surprising that the relationship between Kt/V and URR thresholds is similar. However, this computational limitation does not compromise the relevance of the findings. First, URR remains the principal measure of dialysis dose in the United States [14], so the findings reported herein are germane to the preponderance of clinical dialysis practices. Second, within the FMC system from which the data is derived, and throughout the United States, few practitioners convert URR to Kt/V. Thus, their clinical behavior is driven by the results reported from URR, rather than the Kt/V. Third, to date, there is no peer-reviewed literature that 744 Szczech et al: Hemodialysis dose and optimal survival demonstrates that URR is worse than Kt/V as an outcome predictor for patient mortality. Thus, although some Kt/V calculations offer enhanced accuracy for calculating total solute clearance, its relative value over URR is highly debatable for evaluating mortality relationships. Finally, even if the findings reported herein were abrogated by substituting selected Kt/V formulae in the spline models, the URR relationship persists, and it is the URR relationships that are clinically relevant in the spline models, since this is what most practitioners use rather than Kt/V. In the context of health services, it is worth commenting about the environment that may have contributed to acceptance of higher URR values, indifferent of how they may be achieved. The renal community’s goal was to improve patient health and survival. In this context, the mortality rates for the ESRD program nationally and within FMC were increasing from 1991 to 1993, subsequently stabilizing and improving in recent years. While dose of dialysis is not the only factor that influences mortality among ESRD patients [4, 5], the principal focus for affecting this rate became the surrogate outcome, URR. Concern about the potential for regulatory sanctions and an excessive focus of the measurement alone may have frustrated adherence to fundamental principles of continuous quality improvement (CQI). For example, Dr. W. Edwards Deming, an international leader in CQI concepts and technology, taught that one must thoroughly understand the technical and statistical attributes of a production process for it to be used for quality improvement [40, 41]. His principles may be applied to the method by which the URR is measured and how changes in the measurement method influence the association between the measurement and the clinical outcome demonstrated here. CQI mandates that if a process is altered (change in dialysis membranes and/or blood sampling techniques) that may modify an essential surrogate outcome (URR), its relationship to the ultimate outcome must be reevaluated [42]. Until now, the needed validation for the URR-mortality relationship has not been updated. Applying these principles to the measurement of hemodialysis dose as URR, we need to recalibrate the definition of hemodialysis “adequacy” and readjust our strategy for achieving the newly defined goals. These data suggest the need for uniform blood sampling practices, a change in the benchmark URR, and perhaps an increased appreciation of the fundamental principles of CQI. ACKNOWLEDGMENTS We are grateful to Dr. Robert Wolfe, Dr. Frederich Port, and Dr. Alan Kliger for helpful discussions during the development of this manuscript. The views expressed herein are those of the authors and do not necessarily reflect policy of Fresenius Medical Care, North America. Reprint requests to Lynda Anne Szczech, M.D., M.S.C.E., Duke University Medical Center, Institute for Renal Outcomes and Health Policy Research, Box 3646, Durham, North Carolina 27710, USA. E-mail: [email protected] REFERENCES 1. Owen WF, Lew NL, Liu Y, et al: The urea reduction ratio and serum albumin concentration as predictors of mortality among patients undergoing hemodialysis. N Engl J Med 329:1001–1006, 1993 2. Held PJ, Port FK, Wolfe RA, et al: The dose of hemodialysis and patient mortality. Kidney Int 50:550–556, 1996 3. McClellan WM, Soucie JM, Flanders WD: Mortality in endstage renal disease is associated with facility-to-facility differences in adequacy of hemodialysis. J Am Soc Nephrol 9:1940–1947, 1998 4. Collins AJ, Ma JZ, Umen A, et al: Urea index and other predictors of hemodialysis patient survival. Am J Kidney Dis 23:272–282, 1994 5. Owen WF, Chertow G, Lazarus JM, et al: The dose of hemodialysis: Mortality responses by race and gender. JAMA 280:1–6, 1998 6. Hakim RM, Breyer J, Ismail N, et al: Effects of dose of dialysis on morbidity and morality. Am J Kidney Dis 23:661–669, 1994 7. Acchiardo SR, Hatten KW, Ruvinsky MJ, et al: Inadequate dialysis increases gross mortality rate. ASAIO J 38:M282–M285, 1992 8. Parker T, Husni L, Huang W, et al: Survival of hemodialysis patients in the United States is improved with a greater quantity of dialysis. Am J Kidney Dis 23:670–680, 1994 9. Renal Physicians Association Working Committee on Clinical Practice Guidelines: Clinical Practice Guideline on Adequacy of Hemodialysis. Clinical Practice Guideline, Number 1. Washington, D.C., December, 1993 10. Hemodialysis Adequacy Work Group: NKF-DOQI clinical practice guidelines for hemodialysis. Am J Kidney Dis 30(Suppl 2):S15– S66, 1997 11. Lowrie EG, Lew NL: The urea reduction ratio (URR): A simple method for evaluating hemodialysis treatment. Contemp Dial Nephrol 11–20, 1991 12. American Association of Kidney Patients Advisory: Inadequate Dialysis Increases the Risk of Premature Death. Tampa, 1995 13. Know Your Number. Baltimore, Health Care Financing Administration; Health Standards, Quality Bureau, 1995 14. Health Care Financing Administration: 1999 Annual Report End Stage Renal Disease Core Indicators Project. Baltimore, Department of Health, Human Services, Health Care Financing Administration Office of Clinical Standards and Quality, December 1999 15. U.S. Renal Data System: USRDS 1999 Annual Data Report. The National Institutes of Health National Institute of Diabetes, Digestive, Kidney Diseases, Bethesda, April 1999 16. Shikin EV, Plis AI: Handbook on Splines for the User. Boca Raton, CRC Press, 1995 17. Deboor C: A Practical Guide to Splines. New York, Springer and Herlock, 1978 18. Wolfe RA, Ashby VB, Daugirdas JT, et al: Body size, dose of hemodialysis and mortality. Am J Kidney Dis 35:80–88, 1999 19. Hosmer DW, Lemeshow S: Applied Logistic Regression. New York, John Wiley & Sons, 1989 20. Lowrie EG, Chertow G, Lazarus JM, et al: The urea {clearance 3 dialysis time} product (Kt) as an outcome-based measure of hemodialysis dose. Kidney Int 56:729–737, 1999 21. Kloppenburg WD, Stegeman CA, Hooyschuur M, et al: Assessing dialysis adequacy and dietary intake in the individual hemodialysis patient. Kidney Int 55:1961–1969, 1999 22. Gotch FA, Levin NW, Port FK, et al: Clinical outcome relative to the dose of dialysis is not what you think: The fallacy of the mean. Am J Kidney Dis 30:1–15, 1997 23. Chertow GM, Owen WF, Lazarus JM, et al: Exploring the reverse J-shaped curve between urea reduction ratio and mortality. Kidney Int 56:1872–1878, 1999 24. Lowrie EG, Zhu S, Lew NL: Primary associates of mortality among dialysis patients: Trends and reassessment of Kt/V and urea reduction ratio as outcome-based measures of dialysis dose. Am J Kidney Dis 32(Suppl):S16–S31, 1998 Szczech et al: Hemodialysis dose and optimal survival 25. Fleischmann E, Teal N, Dudley J, et al: Influence of excess weight on mortality and hospital stay in 1346 hemodialysis patients. Kidney Int 55:1560–1567, 1999 26. Frankenfield DL, McClellan WM, Helgerson SD, et al: Relationship between urea reduction ratio, demographic characteristics, and body weight for patients in the 1996 National ESRD Core Indicators Project. Am J Kidney Dis 33:584–591, 1999 27. Owen WF, Lowrie EG: C-reactive protein as an outcome predictor for maintenance hemodialysis patients. Kidney Int 54:627–636, 1998 28. Yeun JY, Kaysen GA: Factors influencing serum albumin in dialysis patients. Am J Kidney Dis 32(Suppl):S118–S125, 1998 29. Leypoldt JK, Cheung AK, Carroll CE, et al: Effect of dialysis membranes and middle molecule removal on chronic hemodialysis patient survival. Am J Kidney Dis 33:349–355, 1999 30. Eknoyan G, Levey AS, Beck GJ, et al: The hemodialysis (HEMO) study: Rationale for selection of interventions. Semin Dial 9:24–33, 1996 31. Hull AR: Lessons from Japan can go beyond building automobiles. Nephrol News Issues 5:48, 1992 32. Held PJ, Blagg CR, Liska DW, et al: The dose of hemodialysis according to dialysis prescription in Europe and the United States. Kidney Int 38:S16–S21, 1992 33. Eichenwald K: Death and deficiency in kidney treatment. NY Times Section A1, 1995 745 34. Brink S: Dialysis by the numbers. US News World Report 116:58, 1994 35. Congressman Fortney Stark (D-Ca): HR# 1067, 1995 36. Jencks S, Wilensky G: The Health Care Quality Improvement Initiative: A new approach to quality assurance in medicare. JAMA 268:900–903, 1993 37. Beto JA, Bansal VK, Ing TS, et al: Variation in blood sample collection for determination of hemodialysis adequacy. Am J Kidney Dis 31:135–141, 1998 38. Owen WF, Meyer K, Schmitt G, et al: Methodological limitations of the ESRD Core Indicators Project: An ESRD Network’s experience with implementing an ESRD quality survey. Am J Kidney Dis 30:349–355, 1997 39. Li Z, Lew NL, Lazarus JM, et al: Comparing the urea reduction ratio and the urea product as outcome-based measures of hemodialysis dose. Am J Kidney Dis 35:598–605, 2000 40. Deming WE: Out of the Crisis. Cambridge, MIT Centers for Advanced Engineering, 1986 41. Deming WE: The New Economics for Industry, Government Education (2nd ed). Cambridge, MIT Centers for Advanced Engineering, 1994 42. The Food and Drug Administration Modernization Act of 1997 (FDAMA), November 21, 1997. Pub. L. 105–115 b 410(c)(4), 21 U.S.C. b 383(c)(4)