 |
||||
| 2008 JASN IMPACT FACTOR 7.505 | HOME AUTHOR INFO EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP | |||
| CURRENT ISSUE | ARCHIVES | JASN Express | ONLINE SUBMISSION | |
|
||||
| 2008 JASN IMPACT FACTOR 7.505 | HOME AUTHOR INFO EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP | |||
| CURRENT ISSUE | ARCHIVES | JASN Express | ONLINE SUBMISSION | |
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Clinical Nephrology |
* Department of Quantitative Health Sciences, Cleveland Clinic Foundation, Cleveland, Ohio; University of Illinois, School of Medicine, Chicago, Illinois; Division of Nephrology, University of California Davis, Sacramento, California; University of Alabama at Birmingham, Birmingham, Alabama; || Department of Community Health, Wright State University, Boonshoft School of Medicine, Dayton, Ohio; ¶ Renal Division, Washington University School of Medicine, St. Louis, Missouri; ** University of California, School of Medicine, San Francisco, California; National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, Maryland; Renal Research Institute, Beth Israel Medical Center, New York, New York; University of Tennessee Health Science Center, Memphis, Tennessee; |||| Vanderbilt University Medical Center, Nashville, Tennessee; ¶¶ University of Texas Southwestern Medical Center, Dallas, Texas; *** Baylor College of Medicine, Houston, Texas
Address correspondence to: Dr. Tom Greene, Cleveland Clinic Foundation, Department of Biostatistics and Epidemiology, Wb-4, 9500 Euclid Avenue, Cleveland, OH 44195. Phone: 216-444-9933; Fax: 216-445-2781; E-mail: tgreene{at}bio.ri.ccf.org
Received for publication March 25, 2005. Accepted for publication August 11, 2005.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
By the mid-1990s, technological advances led to the feasibility of substantially higher dialysis doses, and a randomized trial, the Hemodialysis (HEMO) Study, was performed to compare outcomes between patients who were assigned to a high-dose group with target spKt/V approximately 1.65 and a standard-dose group with target spKt/V approximately 1.25. Completed in 2001, the HEMO Study found no significant differences in mortality or hospitalization between the dose groups (19).
In contrast to the HEMO Study, observational studies continue to suggest that reductions in mortality with increased hemodialysis dose extend to dose levels similar to or higher than the target dose in the high-dose group of the trial (20–24). Because the observational studies had larger sample sizes than the HEMO Study, it is possible that the observational findings may reflect a true dose effect that was too small to be detected by the randomized trial. Alternatively, it is possible that in the observational setting, relationships between dialysis dose and outcome may be subject to bias as a result of confounding from factors linked to both dose and outcome.
Because of the controlled conditions of the HEMO Study, delivered equilibrated Kt/V (eKt/V) values were narrowly distributed within each dose group, making it unlikely that different levels of delivered eKt/V within a dose group could produce detectable biologic effects on mortality. Thus, any relationships of mortality with delivered dose within the individual dose groups more likely are due to other factors, perhaps related to the patient’s clinical status, that affect both dialysis dose and mortality. The goal of this study was to take advantage of the controlled setting of the HEMO Study to determine whether factors related to the patient’s clinical condition did indeed affect delivered dose and, if so, to consider the implications for observational studies.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Implementation of Dose Intervention
During the conduct of the trial, eKt/V was computed using the Daugirdas-Schneditz rate equation, eKt/V = spKt/V – 0.6 K/V + 0.03 (27,28), where spKt/V was determined from the pre- and postdialysis blood urea nitrogen (BUN) using the two-BUN method (29). Conceptually, eKt/V can be regarded as the ratio of the total treatment clearance of urea, Kt, to the kinetically modeled volume of urea distribution. The kinetic modeling methods are presented elsewhere (27,29–31).
The dose intervention was monitored by monthly kinetic modeling and administered with centrally generated dialysis prescription reports. The central prescription reports provided choices of dialyzer model, blood flow, dialysate flow, and treatment time that resulted in a targeted eKt/V of 1.05 or 1.45, depending on the patient’s dose group and running mean kinetic volume (Vm). The running mean Vm was the unweighted average of the kinetic volumes over the last four to six modeled dialyses, with the most extreme value excluded when the coefficient of variation of the kinetic volumes exceeded 10%. The central prescription reports were updated following designated changes in Vm.
When followed, the prescription algorithm ensured that the long-term mean achieved eKt/V closely approximated the patient’s target eKt/V. However, variations in achieved eKt/V could occur as a result of fluctuations in Vm or of failure of the clinical sites to use the central prescriptions. The reasons for these failures included an unwillingness of the patient to remain in the dialysis clinic for the required time, inadequate access blood flow, or a desire by the patient or the patient’s physician for a different dose than the study target. Causes of variations in Vm include gain or loss of muscle mass (i.e., total body water) or extracellular fluid and technical problems with the delivery of dialysis, including access recirculation and inaccurate blood pump calibration.
Recalculation of eKt/V
On the basis of data obtained during the trial, the original rate equation was modified to eKt/V = spKt/V – 0.40 K/V (+ 0.01 for patients in the high dose group) for reporting the trial results (31). The modified equation gives eKt/V values approximately 0.06 Kt/V units higher than the original rate equation. However, for maintaining consistency in the protocol, the original rate equation was retained to implement the dose intervention throughout the trial. The modification to the formula for eKt/V did not substantially affect the relative values within each dose group in this analysis.
Measurements
Baseline evaluation included the five case-mix factors: Age, gender, race (black versus nonblack), diabetic status, and years of dialysis before the trial. An anthropometric estimate of urea volume was obtained using the Watson formula (Vant) (32). Comorbidity was summarized using the Karnofsky index (33), the overall score from the Index of Coexisting Disease (ICED) (34,35), and eight subindices from the ICED evaluating specific domains: Congestive heart failure, arrhythmia, ischemia and other cardiac complications, cerebral and peripheral vascular disease, respiratory disease, nervous system disorders, gastrointestinal disease and hepatitis, and malignancy. Serum creatinine was obtained from local biochemistry measurements.
Kinetic modeling sessions included central measurements of pre- and postdialysis BUN and predialysis serum albumin by nephelometry (Spectra East, Rockleigh, NJ). Access type, occurrence of access revisions during the previous month, hypotensive episodes, and pre- and postdialysis weight and BP were ascertained from the dialysis clinic’s flow sheets. Hospitalizations were identified by chart review and patient interview and cross-checked with Medicare records (36).
Data Analysis
All survival analyses were stratified by the 15 clinical centers and conducted with follow-up time censored at transplantation or 4 mo after transfer of patients to nonparticipating dialysis units. Because of its greater familiarity, we use the expression "relative risk" (RR) when referring to hazard ratios in Cox regression analyses.
Predictors of Low eKt/V.
Generalized estimating equations for binomial responses with robust SE (37) were used to relate baseline and follow-up factors to the probability that delivered eKt/V fell into the lowest quintile (defined below) for the patient’s assigned dose group at each modeled dialysis. Potential predictors of low eKt/V were evaluated individually within each dose group, controlling for baseline Vant, the five case-mix factors, and clinical center.
Association of Mortality with Delivered eKt/V at Entry.
Cox regression (38) was used to relate mortality risk to baseline eKt/V, adjusting for Vant and the five case-mix factors. RR were estimated separately in the first year and after 1 yr of follow-up. Baseline eKt/V was defined as the average delivered eKt/V for two baseline kinetic modeling sessions performed while the patient remained on his or her prestudy dialysis prescription.
Association of Mortality with Delivered eKt/V during Follow-Up.
Time-dependent Cox regression was used to relate mortality risk at each follow-up time to the immediately preceding running mean eKt/V. A patient’s running mean eKt/V was computed as the unweighted average of the eKt/V values over the preceding 4-mo period, with the exception of the first 4 mo after randomization, when one to three eKt/V values were averaged depending on the number of previous follow-up kinetic modeling sessions. Twenty patients with no follow-up kinetic modeling sessions were excluded. We refer to the running mean eKt/V as the 4-mo running mean, recognizing that fewer measurements were averaged in the first 3 mo.
Both quintile models and segmented regression models were used. For quintile models, the five quintiles of the 4-mo running mean eKt/V values throughout the follow-up period were first obtained for each dose group. At each follow-up time, the most recent 4-mo running mean eKt/V value then was categorized into one of the five quintiles corresponding to the patient’s assigned dose group. In segmented regression models, separate slopes relating log RR to achieved eKt/V were estimated below and above the median eKt/V level within each dose group.
Covariate Adjustment.
We considered three levels of covariate adjustment: (1) Adjustment for baseline Vant and the five case-mix factors; (2) adjustment for Vant and the case-mix factors plus baseline Karnofsky score and the eight comorbidity indices from the ICED; and (3) adjustment for the above plus baseline levels of predialysis serum albumin, serum creatinine, systolic BP, and equilibrated normalized protein catabolic rate.
Interactions.
The basic models were expanded to include interaction terms to evaluate the association of mortality with eKt/V separately for selected patient subgroups. Models with time-dependent interaction terms were used to evaluate the association of mortality with eKt/V separately when different types of access were used (venous catheter, graft, or fistula) and for time periods with and without access issues, defined as the use of a venous catheter or an access procedure at any time in the preceding 4 mo.
Lagged Analyses.
The analyses relating mortality risk to the running mean eKt/V over the preceding 4 mo may be sensitive to reverse causality resulting from accentuated effects of the patient’s condition on delivered eKt/V in the period immediately preceding death. Thus, the segmented regression analyses were repeated with time lags ranging from 1 to 12 mo between the evaluation of the mortality risk and the 4-mo running mean eKt/V.
In a related analysis, mean eKt/V during the first 4 mo of follow-up were related to mortality during the following year (months 4 to 16). This analysis was repeated for successive periods throughout the first 3 yr, in each case relating mortality in a 1-yr interval to the immediately preceding the 4-mo running mean eKt/V.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Follow-Up Measures of Dose
A total of 62,264 kinetic modeling sessions were performed during follow-up. eKt/V or Vm could not be calculated in 4.9% of sessions as a result of treatment interruptions longer than 15 min (2.5% of sessions) or extreme BUN measurements (2.4% of sessions). The analyses of this report are restricted to the first modeling session within each follow-up month. This resulted in excluding an additional 6.4% of sessions. After all exclusions, the data set included 54,841 kinetic modeling sessions conducted in 1826 of the 1846 randomized patients.
During follow-up, the means (SD) of the patient-averaged values of URR, eKt/V (revised HEMO Study formula), treatment time, and blood flow were 66.3% (2.6%), 1.16 Kt/V units (0.09 Kt/V units), 190 min (23 min), and 324 ml/min (63 ml/min), respectively, in the standard-dose group, and 75.2% (2.6%), 1.53 Kt/V units (0.10 Kt/V units), 219 min (23 min), and 410 ml/min (50 ml/min) in the high-dose group.
Running Mean eKt/V Quintiles
Table 1 summarizes treatment parameters for modeling sessions within the 4-mo running mean eKt/V quintiles in the two dose groups. Prescribed eKt/V (defined by the dialysis prescription Kt taken from the dialysis run-sheet and the running mean Vm just before the current session) was similar to the eKt/V target in eight of the 10 quintiles, excepting the top quintile in the standard-dose group and the bottom quintile in the high-dose group. The prescribed eKt/V was at least 10% lower than the high goal target of 1.45 for 23% of modeling sessions in the bottom quintile of the high-dose group. Much of the variation in delivered eKt/V was related to changes in Vm over time; the mean change in Vm from baseline was 16.0% higher in the first compared with the fifth quintile of the standard-dose group (11.6 versus –4.4%) and 20.2% higher in the first compared with the fifth quintile of the high-dose group (15.1 versus –5.1%).
|
|
|
|
|
Association of Mortality with Lagged eKt/V
Figure 2A relates mortality at each follow-up time point to the running mean eKt/V lagged by varying numbers of months. In the standard-dose group (
), the RR per 0.10 reduction in eKt/V attenuated by approximately half for lag periods of 6 mo or greater. In the high-dose group (
), the strength of the association of mortality with reduced eKt/V largely persisted and remained statistically significant (with P < 0.001) with lags up to 12 mo.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The HEMO Study provides a clear example in which as-treated analyses are contradicted by the intention-to-treat results. In the intention-to-treat analysis, the RR for mortality for the high- versus standard-dose group was 0.96 (95% CI, 0.84 to 1.10), or, conversely, the RR for the standard- versus high-dose group was 1.04 (95% CI, 0.91 to 1.19). The mean difference in achieved eKt/V between the two dose groups was 0.37 Kt/V units, so the intention-to-treat effect represents an average risk increase of 4% per 0.37 decrease in eKt/V, or, equivalently, a 1.1% average increase in risk per 0.10 reduction in eKt/V. The upper limit of the 95% CI for standard- versus high-dose RR similarly represents a maximum risk increase of 4.7% per 0.10 reduction of eKt/V. By contrast, in the as-treated analysis, each 0.10 reduction in achieved eKt/V below the group median was associated with risk increases of 58% in the standard-dose group and 37% in the high-dose group, far exceeding the intention-to-treat upper confidence bound. Furthermore, patients in the lowest achieved eKt/V quintile of the high-dose group experienced a 59% greater risk for mortality than did patients in the standard-dose group, despite their higher achieved eKt/V. Thus, the elevated risk for patients with comparatively low eKt/V levels in each dose group must reflect extraneous factors that are associated with both increased mortality and lower eKt/V.
What these factors are is not entirely clear, and somewhat different factors may be operating in the two dose groups. Whereas almost all patients who were assigned to the HEMO standard-dose group were able to attain the 1.05 target, the mean prescribed eKt/V was at least 10% less than the high-dose target for 23% of the sessions in the lowest eKt/V quintile in the high-dose arm, indicating that a substantial number of patients in that group were unable or unwilling to achieve the target dose (Table 1). Larger anthropometric volume, increases in Vm over time, and markers of comorbidity including hospitalizations and declining serum albumin were associated with lower values of eKt/V in both dose groups (Tables 1 and 2). Black race was also associated with a greater propensity for lower eKt/V, possibly because black individuals have a higher average muscle mass and total body water and hence a higher Vm than white individuals of a similar height and weight (43).
It is well known that mortality is increased in patients who receive dialysis with venous catheters (44,45). Catheters can limit blood flow or cause recirculation that prevents delivery of an adequate eKt/V. It is conceivable that the type of vascular access may have influenced the relationship between eKt/V and mortality in the HEMO Study. However, although use of catheters was a strong predictor of a lowest quintile eKt/V in both dose groups (Table 2), the low eKt/V–mortality association strengthened, rather than weakened, after periods without any recorded access modifications or use of catheters. Other sensitivity analyses (not reported here) indicated that this result persists when we also exclude periods in which a dialysis interruption, shortfall in treatment time, or change in blood or dialysate flow was recorded for one or more modeled dialyses. Thus, access-related issues and other technical factors that influence Vm do not, by themselves, seem to explain the large bias in the as-treated analysis.
The most recent large-scale studies from the USRDS and Dialysis Outcomes and Practice Patterns Study (23,46) have reported estimated dose effects that are close to or greater than the 95% upper confidence bound of 4.7% per 0.10 reduction in eKt/V. Thus, the biases observed in the HEMO Study as-treated analyses are larger than the effects of dose on mortality reported in recent observational studies. This indicates that the magnitudes of the bias estimates from the HEMO trial do not generalize directly to observational data sets. However, this also raises the question of whether estimated dose effects in observational studies may be biased upward, albeit to a lesser extent than in the HEMO as-treated analyses. At least three factors may have led to different results between the HEMO Study as-treated analyses and observational studies: (1) The analytic methods, (2) the controlled conditions of the HEMO trial, and (3) the patients included in the studies.
Regarding analytic methods, the consistency of the HEMO as-treated results when adjusted for a wide range of baseline covariates suggests that the differences are unlikely to be explained by differences in covariate adjustment strategies. However, the results of the as-treated analyses, which related mortality to the mean eKt/V over the immediately preceding 4-mo, may have been especially sensitive to the effects of deteriorating conditions immediately preceding death. By contrast, most observational studies have related mortality over a 1- to 1.5-yr follow-up period to measured dose at the start of the study, resulting in an average lag time of 0.5 to 0.75 yr. In fact, some attenuation was observed in the RR for lower eKt/V when a lag time was introduced in the HEMO as-treated analyses (Figure 2, A and B). However, even with a lag of 1 yr, the RR associated with lower eKt/V remained >1, particularly in the high-dose group, whose conditions more closely reflect current treatment practice.
When dialysis dose is targeted precisely, as in the HEMO Study, variations in Vm and comorbidities may lead to prescription shortfalls accounting for a greater proportion of the variation in delivered dose than in the observational setting, where prescribed dose may depend on variations in practice patterns and other factors. Thus, the confounding effects in the HEMO Study as-treated analyses are likely to have been increased as a result of the controlled conditions of the trial. The strong relationship between baseline eKt/V and mortality in the first year after randomization suggests also that because HEMO patients were selected on the basis of their ability to achieve the high-dose target, a lower delivered eKt/V at baseline was a surrogate for risk factors associated with mortality risk.
Although their effects should be reduced, the confounding relationships involving eKt/V and comorbid conditions associated with dose-target shortfalls seem likely to occur also in the observational setting. The strategy of targeting a URR of at least 75% (corresponding roughly to the HEMO Study high-dose target) is consistent with the practice in many dialysis units at the present time. If such a dose-targeting bias does exist, then regardless of how much the average dose is increased, patients who receive lower amounts of dialysis may always seem to do worse. One explanation for a report that dose thresholds seemed to increase steadily in successive analyses of a national database between 1994 and 1997 (24) is that increasing de facto dose targets over this period may have shifted the dose-targeting bias to higher dose levels. Conceivably, a dose-targeting bias, if unrecognized, might lead to a never-ending cycle of recommendations for ever higher dialysis doses.
The randomized comparisons of the HEMO trial suggested a stronger effect of dose in women than in men (46), a phenomenon also reported from recent observational analyses of achieved dose from the USRDS and international data (47). However, the possibility of a larger dose-targeting bias in women than in men, suggested in Table 4 of this study, indicates the need for continuing caution when interpreting this subgroup result.
The discrepancy of the as-treated analysis with the intention-to-treat analysis in the HEMO trial is in some respects analogous to a similar discrepancy in a previous randomized trial of the effect of normalizing hematocrit level with administration of erythropoietin in hemodialysis patients with congestive heart failure or ischemic heart disease (48). In that study, higher achieved hematocrit levels were associated with lower mortality within the individual treatment arms. However, the intention-to-treat analysis found no significant difference in mortality between the randomized groups.
In summary, the dose-targeting bias that was revealed under the controlled conditions of the HEMO Study emphasizes that caution is necessary when interpreting nonrandomized relationships between dialysis dose and outcome. However, we caution against sweeping generalizations of the implications of these findings for observational research. The large majority of clinical questions will continue to be addressed by observational studies as a result of the limited number of randomized trials that can be conducted. Continued advances in statistical methods and increased ability to identify sources of confounding will lead to continuing improvements in the quality of observational inference.
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  E. Vonesh ON SMALL SOLUTE CLEARANCE AND PATIENT OUTCOMES: EVIDENTIAL PRACTICE OR OBSERVATIONAL TREPIDATION? Perit. Dial. Int., November 1, 2009; 29(6): 623 - 629. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. Miskulin, J. Bragg-Gresham, B. W. Gillespie, F. Tentori, R. L. Pisoni, H. Tighiouart, A. S. Levey, and F. K. Port Key Comorbid Conditions that Are Predictive of Survival among Hemodialysis Patients Clin. J. Am. Soc. Nephrol., November 1, 2009; 4(11): 1818 - 1826. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. Argyropoulos, C.-C. H. Chang, L. Plantinga, N. Fink, N. Powe, and M. Unruh Considerations in the Statistical Analysis of Hemodialysis Patient Survival J. Am. Soc. Nephrol., September 1, 2009; 20(9): 2034 - 2043. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Malyszko, J.S. Malyszko, P. Kozminski, K. Pawlak, and M. Mysliwiec Adipokines, Linking Adipocytes and Vascular Function in Hemodialyzed Patients, May Also Be Possibly Related to CD146, a Novel Adhesion Molecule Clinical and Applied Thrombosis/Hemostasis, July 1, 2008; 14(3): 338 - 345. [Abstract] [PDF]
|
|
|
|
 |
|
 S. G. DEMIRJIAN and S. NURKO Anemia of chronic kidney disease: When normalcy becomes undesirable Cleveland Clinic Journal of Medicine, May 1, 2008; 75(5): 353 - 356. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. X. Garg, T. Greene, and N. W. Levin A well-conducted randomized trial that establishes no benefit of therapy is an important medical advance Nephrol. Dial. Transplant., January 1, 2008; 23(1): 52 - 55. [Full Text] [PDF]
|
|
|
|
 |
|
 D. W. Coyne Epoetin Dosing and Dialysis Facility Ownership Reply JAMA, August 22, 2007; 298(8): 862 - 863. [Full Text] [PDF]
|
|
|
|
|
|
J. Himmelfarb Chronic Kidney Disease and the Public Health: Gaps in Evidence From Interventional Trials JAMA, June 20, 2007; 297(23): 2630 - 2633. [Full Text] [PDF]
|
|
|
|
|
|
D. Van Wyck, K.-U. Eckardt, K. Uhlig, M. Rocco, and A. Levin Appraisal of Evidence and Control of Bias in the Kidney Disease Outcomes Quality Initiative Guideline Development Process Clin. J. Am. Soc. Nephrol., January 1, 2007; 2(1): 8 - 10. [Full Text] [PDF]
|
|
|
|
|
|
H. Singh, J. Reed, S. Noble, J. L. Cangiano, D. B. Van Wyck, and for the United States Iron Sucrose (Venofer) Clini Effect of Intravenous Iron Sucrose in Peritoneal Dialysis Patients Who Receive Erythropoiesis-Stimulating Agents for Anemia: A Randomized, Controlled Trial Clin. J. Am. Soc. Nephrol., May 1, 2006; 1(3): 475 - 482. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
HOME
CURRENT ISSUE
ARCHIVES
JASN Express
ONLINE SUBMISSION
AUTHOR INFO
EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP |
