Context.— Although blacks receive lower doses of hemodialysis than whites, their
survival when receiving dialysis treatment is better than that for whites.
Previous studies of the relationship between the dose of dialysis and patient
survival have not controlled for differences in patient characteristics.
Objective.— To examine the association of mortality with the dose of hemodialysis
for clusters of patients categorized by race and sex.
Design.— Retrospective analysis of laboratory data and mortality outcomes from
1994, using a national database of hemodialysis patients.
Patients.— A total of 18,144 black and white patients receiving hemodialysis 3
times weekly who either lived the entire year receiving hemodialysis or died.
Main Outcome Measures.— The fractional reduction of urea in a single dialysis session as the
measured hemodialysis dose (urea reduction ratio [URR]) after controlling
for race, sex, age, and diabetes mellitus. Mortality was determined by strata
of URRs and albumin and creatinine levels.
Results.— Across all age categories, blacks had lower URRs than whites, and men
had lower URRs than women. In an age-adjusted model for evaluating interactions
among URRs, race, sex, and diabetes, the association of URR with mortality
risk was weak among blacks, particularly black men. After adjustment for age
and diabetes, death probability curves were most steep for white women with
URR values less than 60%. The death probability curves were least steep for
black men. There was no meaningful difference between death probability and
albumin or creatinine concentration among the race by sex clusters.
Conclusion.— Using URR, the usual measure of hemodialysis dose, the assumption that
the association between dialysis dose and survival is uniform across demographic
groups appears incorrect. Comparisons of the quality of dialysis patient care
should not rely on URR alone to predict patient survival.
BECAUSE the uremic toxin(s) has not been identified, the quantity of
urea removed during hemodialysis is a clinically accepted surrogate to define
the patient's risk profile. Numerous studies have demonstrated an association
between the dose of hemodialysis and mortality among patients with end-stage
renal disease (ESRD), when all demographic groups of patients are evaluated
together.1-11
The 2 commonly accepted measures of hemodialysis dose are based on the fractional
reduction of blood urea nitrogen concentration during a single hemodialysis
treatment.12,13 The most frequently
used measure of hemodialysis dose is the urea reduction ratio (URR), calculated
by dividing the decrease in blood urea nitrogen (predialysis minus postdialysis
blood urea nitrogen) by the predialysis concentration, expressed as a percentage.1,13,14 Another measure of
hemodialysis dose is based on the pharmacokinetic theory that the fractional
decrease in urea during a dialysis treatment is a mathematical function of
the artificial kidney's clearance of urea (K) times the length of the treatment
(t), divided by the urea distribution volume (V), approximated by the total
body water (TBW).15,16 This ratio,
Kt/V,12-14 can
be calculated from the URR and they are conceptually and mathematically similar.14-16 Patient mortality
is higher when the amount of urea removed (hemodialysis "dose") is low, and
vice versa. Retrospective studies of mortality outcome for patients with ESRD
suggest that the odds of death progressively increase when URR falls lower
than 60% to 65%.1-5
Such findings, and an evidence-based professional consensus, have led 3 national
organizations, including the principal payer of dialysis services, the Health
Care Financing Administration, to advocate a URR of 65% or Kt/V of 1.2 as
thresholds for adequate hemodialysis. These thresholds have recently been
used to profile dialysis providers.12,13,17-19
Systematically, blacks have lower URR measurements than whites, and
the percentage of blacks treated with a URR lower than 65% is greater than
whites.19 Paradoxically, blacks are known to
enjoy better survival receiving hemodialysis than whites.20,21
Despite these observations, and the resultant health policy decisions based
on them, there have been no studies evaluating the relationship of hemodialysis
dose and survival among subgroups of patients, in the manner performed for
the composite patient population.1,2,6,8,10,11
Therefore, in an ESRD population clustered by race and sex, we evaluated hemodialysis
dose-response relationships, measured in terms of URR and mortality. The associations
of the serum albumin and creatinine concentrations, 2 powerful conventional
laboratory predictors of patient mortality,1,4,22-25
served as points of reference. The analytical goal was not to evaluate mortality
differences among patient subgroups after adjustments for other measures,
but to examine the associations of these measures with patient mortality.
Data were taken from the routine analytical files of Fresenius Medical
Care–North America (Lexington, Mass) for 1994.23,24
Patients not classified as black or white were excluded, leaving 18,144 patients
who received hemodialysis treatments 3 times weekly, and either lived the
entire year receiving dialysis or died. Values for the URR and serum albumin
and creatinine concentrations measured during the last 3 months of 1993 were
averaged for each patient. All measurements were performed by a single laboratory
(LifeChem Clinical Laboratories, Rockleigh, NJ). Total body water26 and body surface area27
were calculated using accepted formulas.
Two complementary strategies were used. Both focused on the association
of mortality with URR, albumin, and creatinine within clusters of patients
categorized by race and sex.
The independent target measures of URR, albumin, and creatinine were
treated as continuous variables.28 The process
had 3 parts. First, the logisticform of the relationship between odds of death
and each measure was evaluated as previously reported.4,23,24
Odds ratios (ORs) for death, adjusted for age, sex, race, and diabetes mellitus,
were compared with a reference stratum for each to visually evaluate the form
of the mortality relationships among all patients. Separate models, in which
each target was treated as either a first (eg, URR, a linear form) or second
(eg, URR and URR2, a quadratic form of URR) order polynomial adjusted
for age, race, sex, and diabetic status, were also evaluated. Second, each
of the 3 primary targets was evaluated using a logistic regression model that
included race, sex, age, and diabetes as well as interaction terms between
the target and the race, sex, and diabetes clusters. Third, separate URR,
albumin, and creatinine logistic models were evaluated for each race by sex
cluster adjusting for age and diabetes. First-order models were used for albumin
and creatinine; the second order model was used for URR.
Four ordered strata, each for URR, albumin, and creatinine, were constructed
based on clinically relevant and statistically practical values for each variable.
Mortality gradients between the strata of URR, albumin, and creatinine were
evaluated by χ2 analysis. The χ2 was partitioned
into linear and residual components, allowing evaluation of whether there
existed a linear trend difference among the strata and/or if there existed
differences not explained by a linear trend.29,30
The patients' median age was 63.2 years (black men, 60.0 years; black
women, 64.1 years; white men, 62.6 years; white women, 66.7 years). Thirty-nine
percent of the patients were diagnosed as having diabetes (black men, 29.4%;
black women, 46.6%; white men, 36.7%; white women, 44.3%). Blacks had lower
mortality rates than whites. Table 1
shows summary statistics for race by sex clusters. The differences in weight
(F=537; P<.001), TBW (F=4140; P<.001), and body surface area (F=1196; P<.001)
among clusters were significant. All these values were higher in blacks than
whites and in men than women. Similarly, URR values differed significantly
among clusters (F=538; P<.001), highest among
white women and lowest among black men. The URR was inversely associated with
body weight (r=−0.33; P<.001),
TBW (r=−0.38; P<.001),
body surface area (r=−0.36; P<.001), creatinine (r=−0.22; P<.001), and albumin (r=−0.016; P<.04). Albumin and creatinine concentrations were higher
in men and blacks. The association of creatinine and albumin with TBW was
direct (rcreatinine=0.35, P<.001; ralbumin=0.16, P<.001).
Logistic modeling using continuous rather than stratified variables
revealed significant first-order effects for URR, serum albumin, and creatinine
concentrations. However, significant second-order effects were observed for
URR only. Since including the second-order URR2 term better described
what was observed clinically, and was statistically significant, the remaining
logistic evaluations of URR included a second-order term.
Table 2 summarizes the age-adjusted
model, evaluating interactions between URR and the race, sex, and diabetes
clusters. There was a strong interaction between race and URR (race×URR;
race×URR2), suggesting that the significant association of
mortal risk with URR was manifest mainly among whites. In contrast, the interactions
among URR and sex and diabetes were not significant (sex×URR; sex×URR2; diabetes× URR; diabetes×URR2). The association
between URR and mortality was not significant (P=.10)
after adjustment for the cluster variables and interactions, suggesting that
the association of URR with mortal risk was weak among blacks, particularly
men.
Table 3 shows a similar
model evaluating possible interactions with the serum albumin concentration.
The inverse association of albumin with death risk was highly meaningful (OR,
0.19). There were significant interactions of albumin with sex and diabetes
(sex×albumin; diabetes×albumin), each serving to increase the
OR for death associated with it (ORsex×albumin,
0.23; ORdiabetes×albumin, 0.25). The race interaction
nearly achieved statistical significance (ORrace×albumin, 0.23). However, interactions with albumin were small, and the effect
of albumin was highly meaningful among all clusters. A similar analysis of
creatinine concentration was performed (data not shown). Creatinine was also
inversely associated with odds of death (OR, 0.89; 95% confidence interval,
0.92-0.88 per 1 mg/dL increase); there were no significant interactions with
race, sex, or diabetes.
Figure 1 shows death probability
curves associated with URR, albumin, and creatinine that resulted from separate
logistic models evaluated for each race by sex cluster, after adjustment for
age and diabetes. A steep increase in the risk of death occurred among white
women as URR values fell below 60%. To a lesser extent, an increased death
risk was also seen at similar URR values among white men. The association
of URR with mortal risk was least steep among black men and women. The increase
in risk at low albumin concentrations was steeper among men than women. The
form of the relationships between mortality and creatinine did not differ
between the race by sex clusters.
Table 4 evaluates the strata
of URR, albumin, and creatinine among all patients. The gradients along the
strata for albumin and creatinine were monotonic and inverse. The death rate
for each stratum of higher concentrations of albumin or creatinine was lower
than for adjacent strata of lower concentrations. This was not observed for
URR. Statistical tests revealed highly significant linear trends for each
measure. Residual differences among the strata were smaller or did not exist.
Figure 2 illustrates the relationship
of URR to death rate for each race by sex cluster. Neither the linear (χ2=0.99) nor residual (χ2=0.54) differences between the
strata were significant for black men. The linear value (χ21=3.81; P=.05), but not the residual value
(χ22=3.12; P=.21), was
significant for black women. The relationship between URR and death rate was
monotonic and highly significant among white women. Among white men it was
nonmonotonic; both linear and residual χ2 statistics were significant.
If only the lower 3 strata (URR≤55% through 60%-65% for white men) were
considered, the linear difference among strata was significant (χ2=18.42; P<.001), while the residual component
was not (χ2=0.03; P=.05). Thus, the
significant residual χ2 statistic was caused by the slightly
greater death rate observed in the highest URR strata.
Figure 3 shows death rates
associated with albumin concentration among the race by sex clusters. The
relationships were monotonic in all clusters. The linear χ2
value was highly significant for all 4 clusters (P<.001)
and residual χ2 value was only highly significant for black
and white men. For creatinine concentration, the relationship to death rate
was also monotonic, inverse, and similar among the clusters (data not shown).
The linear χ2 value was highly significant for all 4 clusters
(P<.001) but residual χ2 value
was not (P>.05). Further partitioning of the race
by sex clusters by diabetes revealed similar trends for all 3 primary target
variables (data not shown).
A hypothesis offered for the enhanced survival enjoyed by blacks receiving
dialysis is that blacks and whites have differential sensitivity to the hemodialysis
dose.21 The goal of the current analysis was
to evaluate the assumed null hypothesis of no difference among the races and
sexes with respect to their mortality response to URR.1,2
Many mathematical models can be used to describe relationships among even
a limited number of variables. All such models must be regarded as taken from
a number of possible models, and it is difficult to say with certainty which
representation best reflects reality. Therefore, we supplemented logistic
modeling with evaluation of simple contingency tables of the race by sex clusters
stratified by URR, albumin, or creatinine. The decline in survival at lower
URR was higher among whites than blacks, especially among white women. Such
differences among the race by sex clusters were much less for the other predictors
(serum albumin and creatinine concentrations) of survival in patients with
ESRD.1,6,22-24
Therefore, the data suggest that the hypothesis of no difference is incorrect.
An arguable criticism of the current analysis is that URR is an imprecise
measure of the total urea clearance during a single hemodialysis session.
The URR does not account for convective clearance of solute achieved by fluid
removal during hemodialysis.16,31,32
Black patients may have larger weight gains than white patients, so affect
more solute clearance than is appreciated by URR. However, there is no evidence
that blacks routinely have greater weight gains, especially across all age
groups. An additional issue is that URR does not account for the contribution
of residual renal function. Arguably, blacks with ESRD may have greater residual
renal function, so they require less clearance from hemodialysis. There is
no evidence to confirm or dispute this hypothesis.
Differential racial sensitivity to hemodialysis dose may contribute
to the survival advantage of blacks with ESRD. If blacks are less sensitive
to lower doses of hemodialysis, black patients with URR values lower than
65% will not exhibit an exaggerated mortality in comparison with whites. A
fundamental biologic question is the explanation for this difference. The
data presented herein suggest that one component is that the prevalent measure
of the dose of hemodialysis is flawed. Blacks were observed to have greater
weight, TBW, body surface area, serum albumin levels, and creatinine concentrations
than whites. Therefore, lower URRs among blacks and men were a consequence
of their larger urea distribution volume. Since TBW was directly associated
with albumin and creatinine concentrations, urea distribution volume may be
viewed as a nutritional surrogate. Such nutritional surrogates are powerful
and independent mortality outcome predictors for dialysis patients.1,4,11,21-25
We propose that the nutritional-derived effect of urea distribution volume
on death risk may supersede its mortality impact by way of the hemodialysis
dose.33 As viewed by segregating them into
their components, a low URR or mathematical function of the artificial kidney's
clearance of urea times the length of the treatment, divided by the urea distribution
volume (Kt/V) may have 2 etiologies that result in different mortality effects.
A low URR may be due to a low mathematical function of the artificial kidney's
clearance of urea times the length of the treatment (eg, too little solute
removal), which would enhance the patient's death risk. Alternatively, a low
URR may be due to a large urea distribution volume (eg, improved muscle mass
and nutrition), and the latter would confer a lower death risk. We propose
these complex clinical, laboratory, and statistical linkages may contribute
to the observed differential death risk in small white women and large black
men, respectively. This statistical and clinical chain of logic suggests that
different doses of hemodialysis are not required for blacks and whites. Instead,
the conventional measure of dialysis dose needs to be refined.
Urea reduction ratio is illustrative of the difficulty encountered establishing
national clinical performance measures and clinical performance goals. The
reiterative and evolutionary nature of the science behind clinical practice
recommendations mandates that they be dynamic. The 1997 Balanced Budget Amendment
mandates the development of "methods to measure and report on the quality
of renal dialysis services provided under Medicare."34
Because of this mandate, URR was offered.35
A fundamental assumption, supporting the use of URR as a national clinical
performance measure, is that its statistical relationship to patient mortality
is the same for all patient subgroups.1 Using
this chain of logic, manipulating URR is a simple way to predictably affect
dialysis patient mortality.12,13
However, based on the representative patient database presented herein, manipulating
URR alone may not be sufficient to improve mortality. A differential mortality
response to changes in dialysis dose may account for the lesser improvement
in survival probability for blacks in comparison with whites from 1984 to
1994, a period during which the amount of dialysis increased for all racial
groups.19,36
The observation of different patient outcomes by race and sex confounds
the application of URR as a core indicator or clinical performance measure
for hemodialysis.19 Patient mortality may not
be within immediate physician control by effecting this process measure. Furthermore,
quality comparisons of dialysis care that use URR as a clinical performance
measure should be circumspect of the previous assumption that lower URR values
routinely mean that patients' outcomes will be compromised.1,12,13
External comparisons of URR profiles of different nephrologists and dialysis
units19,35 should likewise be
undertaken with consideration of these constraints. However, lower hemodialysis
doses should not be prescribed based on race or sex differences. Validation
of these findings from prospective trials and using other databases with more
precise measures of dialysis dose, such as the National Institutes of Health's
HEMOdialysis Study,37 are needed. Finally,
because higher URR frequently requires increasing the duration of hemodialysis,
and this is a major quality-of-life issue for many patients,38
correct advice about the consequences of a low URR must be based on accurate
information about death risk in the context of the patient's demographic profile.
1.Owen WF, Lew NL, Liu Y.
et al. The urea reduction ratio and serum albumin concentration as predictors
of mortality in patients undergoing hemodialysis.
N Engl J Med.1993;329:1001-1006.Google Scholar 2.Held PJ, Port FK, Wolfe RA.
et al. The dose of hemodialysis and patient mortality.
Kidney Int.1996;50:550-556.Google Scholar 3.Hakim RM, Depner TA, Parker TF. Adequacy of hemodialysis: in depth review.
Am J Kidney Dis.1992;20:107-123.Google Scholar 4.Lowrie EG. Chronic dialysis treatment.
Am J Kidney Dis.1994;24:255-266.Google Scholar 5.Lowrie EG. Thoughts about judging dialysis treatment.
Semin Nephrol.1996;16:242-262.Google Scholar 6.Collins AJ, Ma JZ, Umen A, Keshaviah P. Urea index and other predictors of hemodialysis patient survival.
Am J Kidney Dis.1994;23:272-282.Google Scholar 7.Hakim RM. Assessing the adequacy of dialysis.
Kidney Int.1990;37:822-832.Google Scholar 8.Hakim RM, Breyer J, Ismail N, Schulman G. Effects of dose of dialysis on morbidity and mortality.
Am J Kidney Dis.1994;23:661-669.Google Scholar 9.Acchiardo SR, Hatten KW, Ruvinsky MJ.
et al. Inadequate dialysis increases gross mortality rate.
ASAIOJ.1992;38(3):M282-M285.Google Scholar 10.Ifudu O, Chan E, Brezsnyak WF.
et al. Interdialytic weight gain and missed dialysis treatments in long-term
hemodialysis patients.
Dial Transplant.1995;24:292-302.Google Scholar 11.Parker T, Husni L, Huang W, Lew N, Lowrie EG.and Dallas Nephrology Associates. Survival of hemodialysis patients in the United States is improved
with a greater quantity of dialysis.
Am J Kidney Dis.1994;23:670-680.Google Scholar 12.Renal Physicians Association Working Committee on Clinical Practice
Guidelines. Clinical practice guideline on adequacy of hemodialysis. In: Clinical Practice Guideline, Number 1.
Washington, DC: Renal Physicians Association; December 1993.
13.Hemodialysis Adequacy Work Group. NKF-DOQI clinical practice guidelines for hemodialysis.
Am J Kidney Dis.1997;30(suppl 2):S15-S66.Google Scholar 14.Lowrie EG, Lew NL. The urea reduction ratio (URR).
Contemp Dial Nephrol.1991;2:11-20.Google Scholar 15.Gotch FA. Kinetic modeling in hemodialysis. In: Nissenson AR, Fine RN, Gentile DE, eds. Clinical
Dialysis. 2nd ed. Norwalk, Conn: Appleton & Lange; 1990:118-146.
16.Depner TA. Prescribing Hemodialysis: A Guide to Urea Modeling. Norwell, Mass: Kluwer Academic Publishers; 1991.
17.American Association of Kidney Patients. Inadequate dialysis increases the risk of premature death. In: American Association of Kidney Patients Advisory. Tampa, Fla: American Association of Kidney Patients; 1995:1-5.
18.Department of Health and Human Services. Know Your Number. Baltimore, Md: Health Care Financing Administration and the Health
Standards and Quality Bureau; 1995.
19.Department of Health and Human Services. ESRD Core Indicators Project, 1996 Annual Report. Baltimore, Md: Health Care Financing Administration and the Health
Standards and Quality Bureau; 1997.
20.Bloembergen WE, Port FK, Mauger EA, Wolfe RA. Causes of death in dialysis patients.
J Am Soc Nephrol.1994;5:1231-1242.Google Scholar 21.Owen WF, Price D. African-Americans on maintenance dialysis.
Adv Ren Replace Ther.1997;4:3-12.Google Scholar 22.Lowrie EG, Lew NL. Death risk in hemodialysis patients.
Am J Kidney Dis.1990;15:458-482.Google Scholar 23.Lowrie EG. Conceptual model for a core pathobiology of uremia with special reference
to anemia, malnourishment, and mortality among dialysis patients.
Semin Dial.1997;10:115-129.Google Scholar 24.Lowrie EG, Huang WH, Lew NL, Liu Y. The relative contribution of measured variables to death risk among
hemodialysis patients. In: Friedman E, ed. Death on Hemodialysis: Preventable
or Inevitable? Norwell, Mass: Kluwer Academic Publishers; 1994:121-141.
25.Lowrie EG, Lew NL, Huang WH. Race and diabetes as death risk predictors in hemodialysis patients.
Kidney Int.1992;38:S22-S31.Google Scholar 26.Chertow GM, Lowrie EG, Lew NL, Lazarus JM. Development of a population-specific regression equation to estimate
total body water in hemodialysis patients.
Kidney Int.1997;51:1578-1582.Google Scholar 27.DuBois D, DuBois EF. A formula to estimate the approximate surface area if height and weight
be known.
Arch Intern Med.1916;17:863-871.Google Scholar 28.Hosmer DW, Lemeshow S. Applied Logistic Regression. New York, NY: John Wiley & Sons Inc; 1989.
29.Finestein AR. Multivariable Analysis. New Haven, Conn: Yale University Press; 1996:164-187.
30.Armitage P. Tests for linear trends in proportions and frequencies.
Biometrics.1955;11:375-386.Google Scholar 31.Sherman RA, Cody RP, Rogers ME, Solanchick JC. Accuracy of the urea reduction ratio in predicting dialysis delivery.
Kidney Int.1995;47:319-321.Google Scholar 32.Depner TA. Estimation of Kt/V from the URR for varying levels of dialytic weight
loss: a bedside graphic aid.
Semin Dial.1993;6:242.Google Scholar 33.Owen WF, Lowrie EG. CRP as an outcome predictor for maintenance hemodialysis patients.
Kidney Int.1998;54:627-636.Google Scholar 34. Not Available Balanced Budget Act of 1997, §4558 (1997).
35.Health Care Financing Administration. HCFA publication 60 A/B, change request No. 218. Health Care Financing Administration: Baltimore, Md. 1997.
36.US Renal Data System. USRDS 1997 Annual Data Report. Bethesda, Md: The National Institutes of Health and National Institute
of Diabetes and Digestive and Kidney Diseases; 1996.
37.Eknoyan G, Levey AS, Beck GJ.
et al. Hemodialysis (HEMO) Study.
Semin Dial.1996;9:24-33.Google Scholar 38.Hornberger JC. The hemodialysis prescription and quality adjusted life expectancy.
J Am Soc Nephrol.1993;4:1004-1020.Google Scholar