 |
||||
| 2007 JASN IMPACT FACTOR 7.111 | HOME AUTHOR INFO EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP | |||
| CURRENT ISSUE | ARCHIVES | JASN Express | ONLINE SUBMISSION | |
|
||||
| 2007 JASN IMPACT FACTOR 7.111 | HOME AUTHOR INFO EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP | |||
| CURRENT ISSUE | ARCHIVES | JASN Express | ONLINE SUBMISSION | |
The Division of Nephrology, Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland.
Correspondence to Dr. Jeffrey C. Fink, University of Maryland Medical Systems, 22 South Greene Street, N3W143, Baltimore, MD 21201. Phone: 410-328-5720; Fax: 401-328-5685; E-mail: jfink{at}medicine.umaryland.edu
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
, which was only minimally
diminished after adjusting for individual-level covariates (adjusted ,
0.14; P < 0.0001). The variation in URR attributable to the center
effect, quantified by R2, was greater than that related to
individual-level dialysis factors (facility- and individual-level dialysis
covariates R2, 23.6 and 11.3%, respectively). Initiatives to
improve the delivery of dialysis in patients with end-stage renal disease
should be directed at facility policies governing dialysis care, along with
patient-specific problems, because center effects have a major influence on
dialysis adequacy. |
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
30% or 
90%. Individual URR values were calculated with the
following formula:
 |
The information recorded, along with URR and dialysis facility, included several other variables that were designated as either case-mix factors or dialysis-specific factors. The case-mix factors included age, race, gender, and presence of diabetes. The dialysis factors included access type, blood flow, time on dialysis, dialyzer urea clearance, and the predialysis BUN value. Access type was categorized into those who had an in-dwelling venous catheter, autologous arteriovenous fistula, or polytetrafluoroethylene arteriovenous graft. The manufacturer's estimate of urea clearance at a blood flow of 200 ml/min was used for all of the dialyzers in the study. The multivariate analyses used a reduced sample (n = 3162) that included only patients who had complete information on all covariates and dialyzed in units with at least 4 other patients who also were included in the reduced sample.
The study sample was contrasted to a data set designed to simulate the null hypothesis, which stated that patients are not influenced by a center effect in dialysis adequacy. The simulated data set included the same patients but replaced their actual URR values with random URR values. Each random URR value was drawn from a normal distribution with a mean and SD derived from the actual data and then was assigned to every patient who had no change in his or her actual facility identifier. All of the random URR values were independent from one another, and in the random sample a selected patient was no more likely to have a value similar to a neighbor at his or her facility than to any other patient in the population.
Analysis
The presence and significance of a center effect on dialysis adequacy was
determined by (1) using the within-center correlation in URR values
across all facilities, quantified with the parameter , and (2)
measuring the extent of the between-center variation in facility mean URR
values, which is related to (see the Statistical Analyses section). The
implication of the null hypothesis was that there was no within-center
correlation in URR values, with = 0, and a minimal between-center
variation in facility means around the overall network mean URR. To
demonstrate the significance of within-center correlation and between-center
variation in URR values, the calculations and between-center variation
for the actual URR values were compared with the same parameters calculated
from the random data sample. The estimated across all centers in the
sample were also adjusted for patient-specific case-mix and dialysis factors
to determine to what extent the within-center correlation in URR values was
confounded by these covariates. Finally, because a portion of the
within-center correlation may have been related to facility policies on
postdialysis BUN sampling, units were grouped by timing of postdialysis BUN
sampling and estimates were made within each stratum.
The relative strength of the center effect versus case-mix and dialysis-specific factors as predictors of URR values was compared. The proportion of total variance in URR explained by each set of covariates was determined using R2 estimates. The R2 parameter was derived from a multivariate linear model and provided an estimate of that fraction of total variance that was attributable to the covariates included in the model.
Statistical Analyses
The details of the statistical model have been reported elsewhere but are
outlined briefly here (3). A
large sample of URR values approximates the distribution of all URR values
with an overall (grand) mean designated by µij =
Yij/n. Yij is observation j in center
i, and the naive variance is 
2/n,
where n is the number of observations in the sample. The
naive variance, which does not account for the within-center
correlation, can be broken down into its components:
 |
b2 is the between-center variance and
e2 is the within-center variance. These terms can
be used to calculate the average intraclass or within-center correlation of
observations among dialysis centers in the sample:
 |
When patients within dialysis centers are independent and noncorrelated,
observations from a center are equivalent to one in a series of random samples
from the total population and each mean should approximate the grand mean of
the entire population. In this scenario, b2 is
small and approaches 0. However, if patients within dialysis centers are
highly correlated, then the observations from each center no longer represent
a random sample of the population; therefore, the mean of each dialysis center
differs from the grand mean, the variance b2 is
large, and approaches 1
(2).
A one-way ANOVA table can be used to decompose the elements of variation
for a sample of URR values (Table
1) (4). The ANOVA
table components can then be used to estimate the mean
(4,5):
 |
|
The SAS PROC GLM procedure was used for this analysis (Statistical Analysis
Software, Cary, NC). Estimates for were also made using random effect
models and the method of restricted maximum likelihood found in the PROC MIXED
procedure of SAS (6). Because
the results from PROC MIXED were not significantly different from those
derived with PROC GLM, only the latter are reported in the results.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
Table 4 provides
estimates of the within-center correlation in URR values and the proportion of
variation, R2, in URR explained by each set of variables. There was
a strong within-center correlation across all centers in both the reduced and
the total samples, which was in contrast to the null estimate obtained
for the individuals with random URR values. Adjusting for case-mix or
dialysis-specific factors only minimally diminished the estimate across
centers. The strength of the within-center correlation, as estimated by ,
was not diminished within groups of facilities with common policies for
sampling of postdialysis BUN levels. In the analysis of the reduced sample,
the proportion of variance in URR attributable to facilities was 23.6% and
greater than that related to either the case-mix or dialysis covariates alone
(unadjusted R2, 12.3 and 11.3%, respectively). Inclusion of all
covariates, in addition to centers in the final multivariate model, accounted
for 37.8% of the total variance in URR values found in the sample.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
.
Furthermore, adjusting for predialysis BUN, which is likely to be influenced
by the frequency of dialysis sessions per week, had little effect on the
within-center correlation. The facilities as a set of independent variables
accounted for a larger proportion of variation in URR than either the case-mix
or individual-level dialysis factors. As physicians and policy makers devise strategies to improve overall dialysis adequacy in patients with ESRD, they need to choose interventions that have the greatest effectiveness because of finite resources. The results of this study suggest that variations in the adequacy of a hemodialysis population may be more attributable to policies and procedures that are unique to the dialysis facility than to individual variations in dialysis factors; therefore, initiatives to improve efficacy may be better directed to the facility. Although centers were found to be significant predictors of variations in URR, as indicated by the within-center correlation and between-center variation in adequacy, it is worth noting that these were characteristics of a network rather than of individuals. Hence, even units with superior aggregate results for adequacy were likely to have a significant fraction of individuals who did not meet the minimum criteria for adequate dialysis. These patients, despite the performance of their unit, probably required modification of those individual factors that accounted for their poor adequacy.
Several studies have examined the most common patient-specific factors that impede the delivery of adequate dialysis (7,8,9,10). Few studies have acknowledged the role of facilities in the outcomes of patients with ESRD. McClellan et al. (11) demonstrated that when the facility was treated as the unit of analysis, there was a broad between-center variation in both dialysis adequacy and mortality rates (12, 13). In a report that used data from the U.S. Renal Data System, there was a higher mortality rate found for patients who received dialysis at for-profit dialysis units compared with similar individuals who received dialysis at not-for-profit dialysis units (14). A previous study by Fink et al. (15) showed that a significant factor that accounted for regional variations in dialysis adequacy within a network was the difference in achieved adequacy attributable to dialysis centers in the region.
The present study was a unique effort that addressed problems of dialysis adequacy in two ways. First, the study quantified the degree of within-center correlation after adjusting for several individual-level dialysis characteristics that were not available in previous studies of center effects in dialysis patients. Second, the analysis compared the relative influence of center- versus individual-level characteristics within one large sample of patients with ESRD. These findings lend support to the notion that quality improvement efforts should be directed at the center and their associated practice patterns along with attention to individual patient problems that affect adequacy. In fact, there is growing evidence that quality improvement initiatives directed at facilities have a positive impact on the overall performance of a network in dialysis adequacy (16).
Studies that are based on retrospective data analyses such as this have several limitations that should be considered when interpreting the results. It is possible that the method of sampling may have introduced a bias that led to similarities in URR values within centers that could not have been adjusted for in the analysis. This study, however, confirmed that there was a strong within-center correlation in dialysis adequacy on an independent sample of patients with ESRD with a distinct sampling strategy from that used in the previous study of within-center correlations (3), thus minimizing the likelihood of a sampling bias. The final model derived in the study accounted for only one third of the total variance in URR values found in the Network 5 sample. This suggests that there probably were other important predictors of dialysis adequacy that as of yet have not been identified; however, it is not likely that there was a single unmeasured covariate that could have explained a predominant proportion of the residual variance in URR values. There may be explanations for the strong within-center correlation and between-center variation in dialysis adequacy other than the center effect. These include the possibility that there was an affinity among patients with similar characteristics for certain centers and interactions between individuals that could have led to similar outcome measurements. It would be difficult, however, to theorize a causal connection between such group behavior patterns and URR measurements that are a function of the efficacy of a given dialysis session.
The conclusion of this study was that there are likely to be center effects on hemodialysis adequacy that should be considered in efforts to improve delivery of adequate dialysis within a region. It remains to be determined whether the center effect that led to the measured within-center correlation and between-center variation in URR measurements were the same across all centers. It is probable that the characteristics that distinguish some centers as superior performers are different from those properties that lead others to have inferior results, and these characteristics should be identified. Because there are clear national standards for dialysis adequacy, strong within-center correlations and between-center variations in URR values drawn from a population of dialysis patients are likely to be an undesirable property for a network. Further work needs to be conducted to evaluate whether quality improvement efforts directed at center practice patterns can use changes in center effects as a gauge of the efficacy of those efforts.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  J. C. Fink, V. D. Hsu, M. Zhan, L. D. Walker, C. D. Mullins, C. Jones-Burton, P. Langenberg, and S. L. Seliger Center Effects in Anemia Management of Dialysis Patients J. Am. Soc. Nephrol., February 1, 2007; 18(2): 646 - 653. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. C. Fink, M. Zhan, S. A. Blahut, M. Soucie, and W. M. McClellan Measuring the Efficacy of a Quality Improvement Program in Dialysis Adequacy with Changes in Center Effects J. Am. Soc. Nephrol., September 1, 2002; 13(9): 2338 - 2344. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. C. Winkelmayer, R. J. Glynn, M. A. Mittleman, R. Levin, J. S. Pliskin, and J. Avorn Comparing Mortality of Elderly Patients on Hemodialysis versus Peritoneal Dialysis: A Propensity Score Approach J. Am. Soc. Nephrol., September 1, 2002; 13(9): 2353 - 2362. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. S. Brimble, J. S. Onge, D. J. Treleaven, and E. J. Carlisle Comparison of volume of blood processed on haemodialysis adequacy measurement sessions vs regular non-adequacy sessions Nephrol. Dial. Transplant., August 1, 2002; 17(8): 1470 - 1474. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
HOME
CURRENT ISSUE
ARCHIVES
JASN Express
ONLINE SUBMISSION
AUTHOR INFO
EDITORIAL BOARD SUBSCRIBE FEEDBACK ALERTS HELP |
