The Official Journal of the

National Kidney Foundation

VOL 40, NO 5, NOVEMBER 2002

AJKD
META-ANALYSIS
American Journal of
Kidney Diseases

Acute Renal Failure in the Intensive Care Unit: A Systematic Review

of the Impact of Dialytic Modality on Mortality and Renal Recovery
Marcello Tonelli, MD, Braden Manns, MD, and David Feller-Kopman, MD

● Background: There is controversy about which dialytic modality should be used for the treatment of acute renal
failure (ARF) in the intensive care unit. We performed a systematic review and meta-analysis to determine the
relative risks (RRs) of mortality and renal recovery associated with intermittent hemodialysis (IHD) therapy
compared with continuous renal replacement therapy (CRRT) in critically ill adults with ARF. Methods: Four
databases (MEDLINE, Cochrane Library, Database of Abstracts and Reviews, and Science Citation Index), hand
searching of conference proceedings and journals, manual review of bibliographies from identified articles, and
contact with experts were used. All randomized trials (published or unpublished in any language) that compared
mortality between intermittent and continuous treatments were eligible. Trials for which an RR for mortality could
not be calculated or with multiple experimental interventions were excluded. Data were extracted separately by two
authors and recorded on a standardized form. Disagreements were resolved by consensus. Results: Six eligible
trials were identified; four of these provided data on renal outcomes. RR (mortality) for IHD was 0.96 (95%
confidence interval [CI], 0.85 to 1.08; P ⴝ 0.50), RR (renal death) was 1.02 (95% CI, 0.89 to 1.17; P ⴝ 0.78), and RR
(dialysis dependence) in survivors was 1.19 (95% CI, 0.62 to 2.27; P ⴝ 0.60; all compared with continuous therapy).
Several sensitivity analyses did not change these results. Of the outcomes studied, the risk for dialysis dependence
in survivors would be most sensitive to the addition of new trials. Conclusions: In comparison to IHD therapy, CRRT
does not improve survival or renal recovery in unselected critically ill patients with ARF. Future studies should
focus on well-defined subgroups of such patients using lessons learned from the trials in this meta-analysis. The
high cost of chronic dialysis therapy and the relative instability of the RR for dialysis dependence suggest that
future trials also should evaluate differences in renal recovery between dialytic modalities. Am J Kidney Dis 40:
875-885.
© 2002 by the National Kidney Foundation, Inc.

INDEX WORDS: Acute renal failure (ARF); hemodialysis (HD); meta-analysis.

Editorial, p. 1097

From the Division of Nephrology, University of Alberta,

A CUTE RENAL failure (ARF) occurs in 20%
to 25% of patients admitted to the inten-
sive care unit (ICU) and is associated with ad-
verse patient outcomes and high health care
costs.1-3 For instance, a significant proportion of
these patients will require dialysis therapy, with
in-hospital mortality rates averaging 40% to
65%.4 Of patients who survive an episode of
ARF in the ICU, 5% to 30% will remain on
long-term dialysis therapy without renal recov-
ery.5
ARF in the ICU can be managed with intermit-
Edmonton, Alberta, Canada; Department of Medicine, Divi-
sion of Nephrology, Calgary Regional Health Authority,
Calgary, Alberta, Canada; and the Divisions of Pulmonary
and Critical Care Medicine, Beth Israel Deaconess Medical
Center, Boston, MA.
Received April 5, 2002; accepted in revised form May 30,
2002.
Address reprint requests to Marcello Tonelli, MD, Divi-
sion of Nephrology, University of Alberta, 11-103E Clinical
Science Bldg, 8440 112 St, Edmonton, Alberta, Canada T6B
2B7. E-mail: mtonelli@ualberta.ca
© 2002 by the National Kidney Foundation, Inc.
0272-6386/02/4005-0001$35.00/0
doi:10.1053/ajkd.2002.36318

American Journal of Kidney Diseases, Vol 40, No 5 (November), 2002: pp 875-885 875
876
tent hemodialysis (IHD) therapy, in which inten-
sive dialysis is performed for a few hours at
variable intervals, or continuous renal replace-
ment therapy (CRRT), during which treatment is
performed continuously at lower blood flow
rates.6 Although CRRT has several theoretical
advantages compared with IHD,7 including en-
hanced hemodynamic stability, increased solute
removal, and greater ultrafiltration capacity, indi-
vidual randomized trials have not supported its
superiority. Consequently, IHD continues to be
widely used.8,9
Unfortunately, several such trials have not
been published as full articles or appear to have
baseline imbalances in illness severity between
treatment groups despite randomization.10,11 At
present, large practice variations exist with re-
spect to the use of IHD and CRRT to manage
ARF in the ICU,9,12,13 perhaps because of a lack
of evidence about which therapy is superior.
We performed a systematic review and meta-
analysis of published and unpublished random-
ized trials to quantify the relative risks (RRs) for
mortality and renal recovery associated with IHD
compared with CRRT in the treatment of ARF in
the ICU. We also performed meta-regression to
control for baseline differences in illness severity
between treatment groups. Our goal is to synthe-
size the available information on this topic, with
extensive use of sensitivity analysis to test the
robustness of conclusions reached.
METHODS
Searching
We searched MEDLINE using the terms (kidney failure-
acute OR “acute renal failure”) and (hemodialysis or haemo-
dialysis or hemofiltration or haemofiltration), using data-
bases from 1969 to January 2002. We also searched the
Cochrane Library and the Database of Abstracts and Re-
views using only the terms (kidney failure-acute OR “acute
renal failure”). We manually reviewed reference lists from
review articles identified in the search and examined ab-
stracts of the major North American nephrology meeting
(American Society of Nephrology) between 1990 and 2001.
We used the Science Citation Index to identify investiga-
tions citing articles of interest. Tables of contents in four
major nephrology journals and three major critical care
journals were reviewed from January 2000 to January 2002
for articles that might be pertinent. Finally, we reviewed
reference lists of investigations meeting our inclusion crite-
ria for other potentially relevant studies. Any trial deemed
worthy of manual review was recorded, as well as the
reasons for subsequent exclusion (if applicable).
TONELLI, MANNS, AND FELLER-KOPMAN
Selection and Data Abstraction
The search strategy and data abstraction were defined by a
prospective protocol. To minimize effects of publication
bias, all studies that compared mortality between contempo-
raneous groups of critically ill patients treated with CRRT
and IHD for ARF were eligible for inclusion, whether
published or unpublished. We did not restrict the search to
the English language. Exclusion criteria were designed to
reduce the risk for selection bias. Trials for which an RR for
mortality or renal recovery could not be determined, that
involved multiple experimental interventions, or that in-
cluded children were excluded. Trials with a nonrandomized
design were eligible for sensitivity analyses (discussed later),
but were not used in assessment of the primary or secondary
end points.
If multiple publications existed by the same investigator,
the studies were carefully reviewed and/or the investigator
was contacted to ensure that no data were analyzed in
duplicate. Some studies did not contain all the required
information; a note was made of this in the log to facilitate a
later inquiry to the investigators. All data were extracted
separately by two authors, and results were compared;
disagreements were resolved by consensus with the aid of a
third party. Extracted data were recorded on a standardized
form.
At least three attempts were made to contact the corre-
sponding and/or first investigator on every included random-
ized controlled trial (RCT). Methods included mailed or
faxed letters, E-mail, and telephone calls. Not all investiga-
tors provided the information requested. Data used in meta-
analysis of RCTs was confirmed by five of the six original
investigators. In some cases, updated information was avail-
able for inclusion.
Outcome Measures
The primary outcome was the pooled estimate of the RR
for mortality for patients with ARF in the ICU treated with
IHD compared with CRRT, obtained from RCTs only. The
definition of mortality varied among studies (some used
in-hospital rates, and others used mortality at ICU dis-
charge). We used these outcomes interchangeably in our
analysis because the same definition was used for both arms
of any given study. Although this may affect the estimate of
the absolute risk for death in patients requiring dialysis
therapy for ARF, it should not affect the difference in
outcomes between treatments. If both in-hospital and ICU-
discharge mortality rates were available, the former was
used. Intention-to-treat results were used when available.
Secondary outcomes included RRs for renal death (death
or dialysis dependence at study end) and dialysis depen-
dence among survivors. Renal death and the presence or
absence of dialysis dependence among survivors were deter-
mined at hospital or ICU discharge, as available.
Validity Assessment
We attempted to assess the quality of each eligible study,
using a previously validated instrument.14 Unfortunately,
many studies had missing data or were only published in
abstract form. Even after including information obtained
directly from the investigators, we could not calculate these
DIALYTIC MODALITY IN ACUTE RENAL FAILURE
scores appropriately in most cases. We therefore abandoned
the attempt to use a composite scoring system to rate study
quality. Instead, we provided qualitative details (when avail-
able) on methods likely to influence the internal validity of
randomized trials.15
Assessment of Trial Heterogeneity
We plotted mortality rates for IHD against those for
CRRT to visually inspect their distribution. The Q statistic
was calculated to provide a numerical measure (and signifi-
cance test) of heterogeneity for each analysis.16 This statistic
tests the null hypothesis that the underlying effect measured
by each of the pooled studies is equivalent. For P less than
0.05 for Q, this assumption is invalid and heterogeneity
exists.
Quantitative Data Synthesis
We first performed a funnel plot of sample size against RR
(mortality) for nonrandomized trials.17 A funnel plot was not
produced for randomized trials because of the small number
of studies and because several of these studies were unpub-
lished, making the plot difficult to interpret. Next, data from
the 18 eligible studies were combined using fixed and
random-effect (DerSimonian and Laird18) models. P is re-
ported for fixed-effects models if the Q statistic suggested no
heterogeneity of effect; in all other cases, random-effects
models were used.
In addition to the primary and secondary outcomes (based
on RCTs only), we also evaluated the effect of study quality
on RR (mortality) by performing the modeling procedure in
two additional groups: nonrandomized trials alone and all
eligible trials. Because nonrandomized trials may be more
likely to favor IHD (because sicker patients often undergo
CRRT in clinical practice), this served as a form of sensitiv-
ity analysis for RR (mortality).19 We did not perform this
analysis for renal outcomes because of the small number of
nonrandomized studies with the requisite data.
We used random-effects meta-regression20 to control for
baseline differences in mean Acute Physiology and Chronic
Health Evaluation II (APACHE II) scores between treatment
arms. Although RCTs do not normally require adjustment
for baseline factors, it appears that significant differences in
illness severity between groups existed in at least one RCT,21
leading to the suggestion that the results are difficult to
interpret.22 To address this issue in the form of a sensitivity
analysis, we performed meta-regression in both RCTs alone
and in all eligible trials.
To explore the possible impact of CRRT technique (hemo-
dialysis, hemofiltration, hemodiafiltration) on outcomes, we
performed sensitivity analyses combining only trials using
equivalent forms of CRRT. To evaluate the possibility of an
era effect, in which outcomes associated with each modality
change according to the time the study was conducted, we
performed an additional analysis combining results only
from trials that enrolled patients in 1995 or later. Meta-
regression was used to formally test for interaction between
these variables and the effect of dialysis modality on mortal-
ity. Finally, we estimated the size of a hypothetical RCT that
would significantly change our results (fail-safe N)23 using
fixed-effects models.
877
All statistical analysis was performed using Intercooled
Stata 6.0 (Stata Corp, College Station, TX). The level of
statistical significance is set at P less than 0.05. Values for
RR are expressed as a point estimate with 95% confidence
intervals (CIs) and P. All RRs refer to the risk for IHD
compared with CRRT.
RESULTS
Trial Flow and Study Characteristics
Abstracts of 2,028 studies were reviewed. Of
these, 116 studies (13 RCTs) were deemed wor-
thy of further exploration and retrieved for re-
view. Eighteen studies (6 RCTs) satisfied inclu-
sion and exclusion criteria and were included in
the analysis.21,24-40 Two potentially eligible stud-
ies were excluded because the CRRT group also
underwent IHD41 or involved children.42 A num-
ber of seemingly relevant studies were multiple
publications and therefore excluded. The remain-
ing studies either did not compare IHD and
CRRT directly or did not provide data on mortal-
ity or renal recovery.
Six studies were RCTs,21,25,33,35,37,40 of which
3 studies were published only in abstract
form25,33,37; 2 studies, as full articles21,40; and 1
study, as a thesis.35 APACHE II scores were
available for 11 studies. Information on indi-
vidual studies is listed in Tables 1 and 2. Four
studies used intention-to-treat analyses, and 1
study did not35 (this information was not avail-
able in one case33). Allocation concealment was
adequate in 1 study,21 inadequate in a second
study,40 and not assessed in the remainder.
Two RCTs excluded patients with severe hypo-
tension.21,35 Timing of and indications for dialy-
sis were determined clinically in most cases; one
study each followed a detailed protocol for adjust-
ment of vasopressors40 and dialysis dose.37 Deliv-
ered dialysis dose was equivalent between arms
in two studies,25,37 apparently greater for CRRT
in one study,21 and not available for the remain-
der. Dialysis membranes used in each arm were
made of the same material in three cases25,37,40
and appeared to be of similar biocompatibility in
one additional trial.35
Quantitative Data Synthesis
Meta-analysis (mortality). The funnel plot
of N against effect size showed an absence of
small nonrandomized studies that found RR (mor-
tality) less than 1, suggesting at least some de-
 878
Table 1. Study Characteristics of RCTs

IHD IHD CRRT CRRT CRRT

Publication Country IHD APACHE IHD Renal IHD Dialysis CRRT APACHE CRRT Renal Dialysis
Reference Year of Origin IHD Technique No. II Mortality Death Dependence CRRT Technique N II Mortality Death Dependence

Simpson and 1993 UK CUP membrane 58 NA 0.83 NA NA CAVHDF, CVVHDF, 65 NA 0.71 NA NA

Allison33 PS membranes
Kierdorf35 1994 Germany PMMA membrane 6-7 52 24.8 ⫾ 5.8 0.64 NA NA CVVHF, PAN 48 26.0 ⫾ 4.0 0.63 NA NA
⫻ wk membrane
Sandy et al37 1998 US PS membrane, 40 24.5 ⫾ 7.7 0.60 0.87 0.69 CVVHD, PS 39 21.4 ⫾ 6.1 0.71 0.821 0.36
dialysis dose ⫽ membrane
CRRT 3.5 ⫻ wk
John et al40 2001 Germany PS membrane, 20 33 ⫾ 4 0.70* 0.8* 0.33* CVVHF, PS 10 34 ⫾ 5 0.70* 0.8* 0.33*
dialytic technique membrane,
not standardized dialytic technique
not standardized
Mehta et al21 2001 US CUP, CA, PMMA, PS, 82 20.6 0.48 0.634 0.07 CAVHDF/CVVHDF, 84 25.3 0.66 0.667 0.14
PAN membranes 5 PS or PAN
⫻ wk membrane
Uehlinger 2001 Switzerland PS membrane, 55 NA 0.51 0.49 0.02 CVVHDF, PAN 71 NA 0.47 0.48 0.02
et al25 dialysis dose ⫽ membrane
CRRT 5-7 ⫻ wk

TONELLI, MANNS, AND FELLER-KOPMAN

NOTE. Values expressed as mean ⫾ SD when appropriate. All outcomes are at hospital discharge, unless otherwise stated. Renal death indicates death or need for dialysis at
end of study; dialysis dependence refers to proportion of survivors requiring dialysis at study end.
Abbreviations:CVVHDF, continuous venovenous hemodiafiltration; CAVHDF, continuous arteriovenous hemodiafiltration; CVVHD, continuous venovenous hemodialysis; NA,
not available; CUP, cuprophane; CA, cellulose acetate; PMMA, polymethylmethacrylate; PS, polysulfone; PAN, polyacrylonitrite.
*This study was not designed to assess the impact of dialytic technique on mortality or renal recovery. Data provided by investigators. End of follow-up was at ICU discharge.
DIALYTIC MODALITY IN ACUTE RENAL FAILURE 879

Table 2. Study Characteristics of Nonrandomized Controlled Trials

Publication IHD IHD APACHE IHD CRRT CRRT APACHE CRRT

Reference Year Country of Origin No. II Mortality No. II Mortality

Mauritz et al27 1986 Germany 22 NA 0.91 36 NA 0.75

Paganini26 1988 US 47 27 0.81 27 27 0.82
Bastien et al28 1991 France 32 19.8 0.75 34 22.5 0.50
McDonald and Mehta30 1991 US 10 NA 0.70* 22 NA 0.77*
Kierdorf32 1991 Switzerland 73 NA 0.93 73 NA 0.78
Kruczynski et al29 1993 Canada 23 28 0.82 12 26.2 0.25
van-Bommel et al36 1995 The Netherlands 34 22.2 0.41 60 26.5 0.57
Neveu et al34 1996 France 141 NA 0.58 28 NA 0.89
Rialp et al31 1996 Spain 21 23.3 0.68 43 24.5 0.76
Bellomo et al38 1999 Australia 47 25.7 0.57 47 29.4 0.53
Swartz et al39 1999 US 137 NA 0.41 90 NA 0.68
Ji et al24 2002 China 92 13.2 0.36 101 21.0 0.41

Abbreviation: NA, not available.

*Among patients undergoing only one dialytic modality.

gree of publication bias (Fig 1). Inspection of
RRs for mortality in individual studies showed
no correlation with year of publication, country
of origin, study method (retrospective versus
prospective), or CRRT technique (arteriovenous
versus venovenous, hemofiltration versus hemo-
diafiltration). However, as a group, effect sizes of
RCTs were more homogeneous than nonrandom-
ized studies.
We first analyzed the six RCTs to evaluate the
primary end point. The Q statistic suggested no
significant heterogeneity across studies (P ⫽
0.09). The fixed-effect RR (mortality) was 0.96
(95% CI, 0.85 to 1.08; P ⫽ 0.50), indicating no
significant difference between IHD and CRRT
(Fig 2). Because of the borderline significance of
the Q statistic, we repeated this analysis using a
Fig 1. Funnel plot of N versus effect size in terms of
RR (mortality). There is an absence of small nonran-
domized trials that found a benefit of IHD, suggesting
publication bias.
random-effects model, which did not change the
results: RR (mortality) 0.97 (95% CI, 0.82 to
1.15; P ⫽ 0.74).
Limiting the analysis to studies in which the
CRRT arm underwent only hemofiltration, only
hemodiafiltration, or either hemodiafiltration or
hemofiltration did not change our results: RRs
(mortality) were 0.97 (95% CI, 0.70 to 1.33),
1.06 (95% CI, 0.82 to 1.37), and 1.00 (95% CI,
0.82 to 1.22), respectively. Restricting analysis
to the three RCTs that enrolled patients in 1996
or later did not affect these results: RR (mortal-
ity) was 0.95 (95% CI, 0.76 to 1.19; P ⫽ 0.68).
Performing formal tests for interaction using
meta-regression did not show a significant rela-
tionship between these variables (type of CRRT
or publication year) and the effect of dialytic
modality on mortality. Finally, excluding the
RCT in which substantial baseline differences
were observed between treatment groups21 did
not change the pooled RR (mortality): 1.05 (95%
CI, 0.91 to 1.20; P ⫽ 0.53), but a test for
interaction was positive (P ⱕ 0.01), confirming
the subjective impression that results obtained in
this trial were significantly different from those
in the other included studies.
The Q statistic for the sensitivity analysis in
which all 18 studies were combined indicated
substantial heterogeneity (P ⫽ 0.0001). The ran-
dom-effect RR (mortality) for this model was
1.00 (95% CI, 0.88 to 1.14; P ⫽ 0.99). Including
only the 12 nonrandomized trials did not appre-
880
ciably change the RR (mortality): 1.00 (95% CI,
0.92 to 1.08; P ⫽ 0.94) or Q statistic (Fig 3).
Fail-safe N: mortality. We estimated how
large a sample size would be required for a single
new or undetected trial to make the pooled RR
(mortality) for RCTs significantly greater than 1.
We assumed that such a study would find a
substantial benefit of CRRT (RR for IHD of 1.2)
and have a mortality rate of 48% in the IHD
group, similar to that observed in the two most
recent RCTs. Using a fixed-effects model, the
required N was 1,250, producing a pooled RR
Fig 3. RR for death for
IHD: sensitivity analysis
(nonrandomized trials).
TONELLI, MANNS, AND FELLER-KOPMAN
Fig 2. RR for death for
IHD: primary analysis (ran-
domized trials).
(mortality) of 1.09 (95% CI, 1.00 to 1.20; P ⫽
0.049). To explore the effect of varying mortality
rate, we repeated this analysis assuming that
72% of patients in the IHD arm would die and
leaving the RR for mortality associated with IHD
at 1.2. The necessary N decreased to 720 in this
scenario: pooled RR (mortality) ⫽ 1.09 (95% CI,
1.00 to 1.18; P ⫽ 0.046). However, both these
estimates probably underestimate the required
sample size because the heterogeneity intro-
duced by such a positive trial suggests that a
random-effects model should be used.
DIALYTIC MODALITY IN ACUTE RENAL FAILURE
Only one RCT had an RR (mortality) greater
than 1 for IHD.33 We determined that the cur-
rently available literature would require the addi-
tion of six trials with identical N (113) and effect
sizes (RR ⫽ 1.13) to make the pooled RR (mor-
tality) significantly greater than 1.
Meta-analysis (renal recovery). Complete
data on renal recovery were available from four
RCTs. There was no heterogeneity of effect for
either renal death or dialysis dependence (Q ⫽
0.79; P ⫽ 0.85 and Q ⫽ 2.6; P ⫽ 0.46, respec-
tively). RR (renal death) from these studies was
1.02 (95% CI, 0.89 to 1.17; P ⫽ 0.78; Fig 4), and
RR (dialysis dependence) was 1.19 (95% CI,
0.62 to 2.27; P ⫽ 0.60; Fig 5).
Excluding the trial that was not designed to
collect renal outcomes40 did not change these
results. Conversely, excluding the RCT in which
substantial baseline differences were observed
between treatment groups21 increased the RR
(dialysis dependence) associated with IHD to
1.66 (95% CI, 0.78 to 3.52; P ⫽ 0.19), although
the RR (renal death) was essentially unaffected
(1.07; 95% CI, 0.90 to 1.28). Only one nonran-
domized trial reported data on renal recovery36;
thus, further sensitivity analysis was not per-
formed.
Fail-safe N: renal recovery. As with RR
(mortality), the addition of either one very large
or several intermediate RCTs would be required
to make RR (renal death) statistically significant
in favor of either therapy. However, the addition
Fig 4. RR for renal death
associated with IHD.
881
of one trial with 190 participants and an effect
size equivalent to that observed by Sandy et al37
would establish a significantly greater incidence
of dialysis dependence among survivors who
underwent IHD (RR, 1.52; 95% CI, 1.01 to 2.29;
P ⫽ 0.047). This shows that renal recovery is the
pooled outcome most likely to be affected by
data from future studies.
Adjustment for Differences in Illness Severity
APACHE II scores for both arms were avail-
able from 11 trials (Tables 1 and 2); 4 of these
trials were RCTs. We performed meta-regression
to adjust for differences in baseline APACHE II
scores, which tended to be higher in the CRRT
groups. This was performed as a form of sensitiv-
ity analysis that favored CRRT.
Using meta-analysis, pooled RR (mortality)
from the 11 studies was 1.02 (95% CI, 0.88 to
1.18; P ⫽ 0.78). Meta-regression yielded an
adjusted RR (mortality) of 1.18 (95% CI, 0.91 to
1.53; P ⫽ 0.20). When this analysis was repeated
on only the 4 RCTs for which APACHE II data
were available, pooled RR (mortality) by meta-
analysis was 0.88 (95% CI, 0.73 to 1.07; P ⫽
0.11), and meta-regression found an adjusted RR
(mortality) of 0.96 (95% CI, 0.81 to 1.14; P ⫽
0.66).
DISCUSSION
This meta-analysis of six RCTs included more
than 600 patients treated in four countries. We
882
found no significant difference in mortality rate
for critically ill patients with ARF treated with
IHD compared with those who underwent CRRT.
This result was unchanged by the inclusion of
nonrandomized trials or by controlling for base-
line differences in illness severity using meta-
regression and was consistent in sensitivity anal-
yses that stratified by CRRT technique and
treatment era.
Given the high mortality rates associated with
continuous therapies (illness acuity notwithstand-
ing), concern has been raised that CRRT may be
harmful, rather than beneficial.39 Our results show
no evidence of adverse outcomes associated with
CRRT, an important finding given that many
patients with ARF are treated with continuous
rather than intermittent therapies.
Patients assigned to CRRT appeared to be
more severely ill in most nonrandomized investi-
gations and in one RCT. These differences may
reflect physician bias in favor of continuous
therapies despite a lack of evidence to support
this belief. Unfortunately, APACHE II scores
were the best available marker of illness severity
in most studies, although limitations of this index
in ARF are well described.43 Nonetheless, adjust-
ing for baseline differences in APACHE II scores
did not significantly change our results.
For renal recovery, neither the incidence of
renal death nor the likelihood of dialysis depen-
dence among survivors was significantly differ-
TONELLI, MANNS, AND FELLER-KOPMAN
Fig 5. RR for dialysis de-
pendence (in survivors) as-
sociated with IHD.
ent between modalities. This component of the
analysis included 401 patients in four RCTs.
Renal recovery is important clinically because
it enables patients to discontinue dialysis therapy,
a treatment associated with significant impair-
ment in health-related quality of life.44-46 More-
over, chronic dialysis therapy is expensive, cost-
ing on average US $69,751/y.47 Because the
majority of this cost is from outpatient dialysis
treatments, greater rates of renal recovery might
save significant resources. In addition, even mild
chronic renal insufficiency is associated with
adverse patient outcomes and high health care
costs, suggesting that the relative incidence of
any sustained renal impairment is potentially
relevant.48
Although we did not find a significant associa-
tion between renal recovery and dialytic modal-
ity, the estimate for RR (dialysis dependence)
was the most likely of the three outcome mea-
sures to be affected by the inclusion of new
studies. Because available data are insufficient to
draw firm conclusions, it is important for future
studies to explore the impact of dialytic modality
on renal recovery from ARF in both the short and
long term.
For methods, we followed published recom-
mendations for conducting and reporting system-
atic reviews.49,50 We included all available stud-
ies in our meta-analysis, including those published
only in the “grey literature” (unpublished or not
DIALYTIC MODALITY IN ACUTE RENAL FAILURE
peer reviewed). We have attempted to provide
information on other variables that are known or
suspected to influence outcome in ARF (Tables 1
and 2).51
Inclusion criteria, dialytic techniques, and pa-
tient populations varied among RCTs in our
meta-analysis. However, there was no statistical
evidence of heterogeneity in the RR for mortality
by dialytic modality. Nonetheless, we took the
conservative approach of using random-effects
models and performing multiple sensitivity anal-
yses to assess the impact of individual trials,
different CRRT techniques, baseline comorbid-
ity, and era in which patients were enrolled.
Results obtained were similar in all analyses,
supporting the conclusion that dialytic modality
used in ARF does not affect rates of death or
renal recovery.
A previous publication constitutes a complete
review of the literature available at that time, but
limited information was available on renal out-
comes.10 In addition, meta-analysis was not per-
formed, in part because of the low number of
randomized trials available to the investigators.
Although a recently published meta-analysis
attempted to address this topic, its primary out-
come was based on combined results of random-
ized and nonrandomized trials, and several appar-
ently relevant randomized investigations were
not included.52 In addition, most of its sensitivity
analyses were based on an unvalidated instru-
ment that was administered by the study’s inves-
tigators to assess data quality and illness severity.
Unfortunately, these analyses are unlikely to ad-
just appropriately for differences between stud-
ies. Finally, it is likely that including all available
investigations would have substantially changed
the results. Thus, we disagree with the investiga-
tors’ conclusions that available data suggest a
survival advantage associated with CRRT.
There are several potential limitations of our
study. Because the study populations included in
our meta-analysis were relatively unselected, our
conclusions may not apply to subsets of patients,
such as those with increased intracranial pressure
or severe hypotension, in whom CRRT might be
more efficacious.53,54
Our systematic review could be criticized on
the grounds that the majority of included studies
have not been published in peer-reviewed jour-
nals. Such a criticism would imply that unpub-
883
lished work may be of lesser quality or might be
less likely to find a difference between treat-
ments. We believe this assertion would be un-
founded. The most likely reason that many of
these studies have not been published is pre-
cisely because they failed to find a significant
difference between modalities, exemplifying the
well-known phenomenon of publication bias. In
addition, poor-quality literature is known to exag-
gerate effect sizes compared with more rigorous
studies55; it therefore is difficult to attribute that
we did not find a mortality difference between
treatments to the inclusion of unpublished trials.
Although limitations of meta-analysis have been
well described,56 we attempted to minimize bias
through rigorous methods. As with all meta-
analyses, despite the multiple sensitivity analyses
performed, it is conceivable that differences in
patient or study characteristics between studies
influenced our results. Although our study does not
eliminate the possibility that dialytic modality con-
fers a survival advantage, it shows that the magni-
tude of such a benefit (if it exists) is exceedingly
small in patients without severe hypotension. We
believe that consideration of unpublished studies
appropriately strengthens this position.
Because we found no difference in mortality
for patients with ARF who were treated with
IHD or CRRT, one could conclude that most
patients should be treated with the less expensive
form of treatment. Most reports suggest that
CRRT is at least twice as expensive as IHD,
increasing treatment costs by US $150 to $450
per day (depending on the form of CRRT consid-
ered).57,58 However, treatment of all patients with
IHD on this basis may be incorrect from both
clinical and economic standpoints because it
does not consider potential downstream out-
comes, such as those attributable to costs of
chronic dialysis therapy.
Future trials aimed at showing a mortality ben-
efit in unselected patients should be undertaken
with caution because our data suggest that the
required N for such a study would be very high.
Nonetheless, multicenter investigations of this is-
sue could be performed successfully, as with other
common interventions in critical illness, even those
laden with perceived ethical issues, such as blood
transfusion.59 The design of new studies evaluating
the impact of dialytic modality in ARF would need
to account for contamination (caused by treatment
884
crossover), illness severity (by stratification), and
variations in dialytic technique, dialysis membrane,
and dose (perhaps by a detailed protocol). Lessons
learned from the conduct of previous trials could be
used to develop inclusion criteria that would mini-
mize ethical concerns and produce a suitably homo-
geneous patient population. Finally, because the
mortality rate of critically ill individuals remains
elevated long after leaving the ICU, outcomes
should be assessed at hospital discharge, or perhaps
even later. Until information from such a trial is
available, a systematic review such as this may
constitute the best evidence possible.
In summary, we found no difference in mortal-
ity between CRRT or intermittent renal replace-
ment therapy for critically ill patients with ARF.
Although we did not find a difference in the
frequency of renal recovery between treatments,
this outcome has been less extensively studied.
However, even a slight difference in renal recov-
ery between therapies might significantly influ-
ence the overall cost-effectiveness of each dia-
lytic modality. As such, future investigations
should collect detailed information on long-term
costs and the relative likelihood of renal recov-
ery associated with dialysis modality.
REFERENCES
1. Bellomo R, Mehta R: Acute renal replacement in the
intensive care unit: Now and tomorrow. N Horiz 3:760-767,
1995
2. Lameire N, Van Biesen W, Vanholder R, Colardijn F:
The place of intermittent hemodialysis in the treatment of
acute renal failure in the ICU patient. Kidney Int Suppl
66:S110-S119, 1998
3. Brivet FG, Kleinknecht DJ, Loirat P, Landais PJ: Acute
renal failure in intensive care units—Causes, outcome, and
prognostic factors of hospital mortality; A prospective, mul-
ticenter study. Crit Care Med 24:192-198, 1996
4. van Bommel EF, Leunissen KM, Weimar W: Continu-
ous renal replacement therapy for critically ill patients: An
update. J Intensive Care Med 9:265-280, 1994
5. Silvester W: Outcome studies of continuous renal
replacement therapy in the intensive care unit. Kidney Int
Suppl 66:S138-S141, 1998
6. Mehta RL: Continuous renal replacement therapies in
the acute renal failure setting: Current concepts. Adv Ren
Replace Ther 4:81-92, 1997
7. Ronco C, Brendolan A, Bellomo R: Continuous versus
intermittent renal replacement therapy in the treatment of
acute renal failure. Nephrol Dial Transplant 13:S79-S85,
1998 (suppl 6)
8. Kierdorf H, Sieberth HG: Continuous renal replace-
ment therapies versus intermittent hemodialysis in acute
TONELLI, MANNS, AND FELLER-KOPMAN
renal failure: What do we know? Am J Kidney Dis 28:S90-
S96, 1996 (suppl 3)
9. Misset B, Timsit JF, Chevret S, Renaud B, Tamion F,
Carlet J: A randomized cross-over comparison of the hemo-
dynamic response to intermittent hemodialysis and continu-
ous hemofiltration in ICU patients with acute renal failure.
Intensive Care Med 22:742-746, 1996
10. Jakob SM, Frey FJ, Uehlinger DE: Does continuous
renal replacement therapy favourably influence the outcome of
the patients? Nephrol Dial Transplant 11:1250-1255, 1996
11. Lameire N, Van Biesen W, Vanholder R: Dialysing
the patient with acute renal failure in the ICU: The emper-
or’s clothes? Nephrol Dial Transplant 14:2570-2573, 1999
12. Murray P, Hall J: Renal replacement therapy for acute
renal failure. Am J Respir Crit Care Med 162:777-781, 2000
13. Cole L, Bellomo R, Silvester W, Reeves JH: A
prospective, multicenter study of the epidemiology, manage-
ment, and outcome of severe acute renal failure in a “closed”
ICU system. Am J Respir Crit Care Med 162:191-196, 2000
14. Imperiale TF, McCullough AJ: Do corticosteroids
reduce mortality from alcoholic hepatitis? A meta-analysis
of the randomized trials. Ann Intern Med 113:299-307, 1990
15. Juni P, Altman DG, Egger M: Assessing the quality of
randomized controlled trials, in Egger M, Davey Smith G,
Altman DG (eds): Systematic Reviews in Health Care:
Meta-Analysis in Context (ed 2). London, UK, BMJ, 2001,
pp 87-108
16. Hedges LV, Ingram O: Statistical Methods for Meta-
Analysis. Boston, MA, Academic, 1985, pp 122-124
17. Egger M, Davey SG, Schneider M, Minder C: Bias in
meta-analysis detected by a simple, graphical test. BMJ
315:629-634, 1997
18. DerSimonian R, Laird N: Meta-analysis in clinical
trials. Control Clin Trials 7:177-188, 1986
19. Egger M, Schneider M, Smith GD: Meta-analysis:
Spurious precision? Meta-analysis of observational studies.
BMJ 316:140-144, 1998
20. Midgley JP, Matthew AG, Greenwood CM, Logan
AG: Effect of reduced dietary sodium on blood pressure: A
meta-analysis of randomized controlled trials. JAMA 275:
1590-1597, 1996
21. Mehta RL, McDonald BR, Gabbai FB, et al: A
randomized clinical trial of continuous vs intermittent dialy-
sis for acute renal failure. Kidney Int 60:1154-1163, 2001
22. Conger J: Dialysis and related therapies. Semin Neph-
rol 18:533-540, 1998
23. Wartenberg D: Residential magnetic fields and child-
hood leukemia: A meta-analysis. Am J Public Health 88:1787-
1794, 1998
24. Ji D, Gong D, Xie H, Xu B, Liu Y, Li L: A retrospec-
tive study of continuous renal replacement therapy versus
intermittent hemodialysis in severe acute renal failure. Chin
Med J (Engl) 114:1157-1161, 2001
25. Uehlinger DE, Jakob SM, Eichelberger M, et al: A
randomized, controlled single center study for the compari-
son of continuous renal replacement therapy with intermit-
tent hemodialysis in critically ill patients with acute renal
failure. J Am Soc Nephrol 12:278A, 2001 (abstr)
26. Paganini EP: Slow continuous hemofiltration and
slow continuous ultrafiltration. ASAIO Trans 34:63-66, 1988
27. Mauritz W, Sporn P, Schindler I, Zadrobilek E, Roth
DIALYTIC MODALITY IN ACUTE RENAL FAILURE
E, Appel W: [Acute renal failure in abdominal infection.
Comparison of hemodialysis and continuous arteriovenous
hemofiltration]. Anasthesiol Intensivmed Notfallmed Schmer-
zther 21:212-217, 1986
28. Bastien O, Saroul C, Hercule C, George M, Estanove
S: Continuous venovenous hemodialysis after cardiac sur-
gery. Contrib Nephrol 93:76-78, 1991
29. Kruczynski K, Irvine BK, Toffelmire EB, Morton AR:
A comparison of continuous arteriovenous hemofiltration and
intermittent hemodialysis in acute renal failure patients in the
intensive care unit. ASAIO J 39:M778-M781, 1993
30. McDonald BR, Mehta RL: Decreased mortality in
patients with acute renal failure undergoing continuous
arteriovenous hemodialysis. Contrib Nephrol 93:51-56, 1991
31. Rialp G, Roglan A, Betbese AJ, et al: Prognostic
indexes and mortality in critically ill patients with acute
renal failure treated with different dialytic techniques. Ren
Fail 18:667-675, 1996
32. Kierdorf H: Continuous versus intermittent treat-
ment: Clinical results in acute renal failure. Contrib Nephrol
93:1-12, 1991
33. Simpson HK, Allison ME: Dialysis and acute renal
failure: Can mortality be improved? Nephrol Dial Trans-
plant 8:946A, 1993 (abstr)
34. Neveu H, Kleinknecht D, Brivet F, Loirat P, Landais
P: Prognostic factors in acute renal failure due to sepsis.
Results of a prospective multicentre study. Nephrol Dial
Transplant 11:293-299, 1996
35. Kierdorf H: Einflub der kontinuierlichen Hamofiltra-
tion auf Proteinkatabolismus, Mediatorsubstanzen und Prog-
nose des akuten Nierenversagens. Medical Faculty, Techni-
cal University, Aachen, Germany, 1994
36. van-Bommel E, Bouvy ND, So KL, et al: Acute
dialytic support for the critically ill: Intermittent hemodialy-
sis versus continuous arteriovenous hemodiafiltration. Am J
Nephrol 15:192-200, 1995
37. Sandy D, Moreno L, Lee J, Paganini EP: A random-
ized stratified, dose equivalent comparison of continuous
veno-venous hemodialysis (CVVHD) vs intermittent hemo-
dialysis (IHD) support in ICU acute renal failure patients
(ARF). J Am Soc Nephrol 9:225A, 1998 (abstr)
38. Bellomo R, Farmer M, Bhonagiri S, et al: Changing
acute renal failure treatment from intermittent hemodialysis
to continuous hemofiltration: Impact on azotemic control.
Int J Artif Organs 22:145-150, 1999
39. Swartz RD, Messana JM, Orzol S, Port FK: Compar-
ing continuous hemofiltration with hemodialysis in patients
with severe acute renal failure. Am J Kidney Dis 34:424-
432, 1999
40. John S, Griesbach D, Baumgartel M, Weihprecht H,
Schmieder RE, Geiger H: Effects of continuous haemofiltra-
tion vs intermittent haemodialysis on systemic haemodynam-
ics and splanchnic regional perfusion in septic shock pa-
tients: A prospective, randomized clinical trial. Nephrol Dial
Transplant 16:320-327, 2001
41. Simpson HK, Allison ME, Telfer AB: Improving the
prognosis in acute renal and respiratory failure. Ren Fail
10:45-54, 1987
42. Maxvold NJ, Smoyer WE, Gardner JJ, Bunchman
TE: Management of acute renal failure in the pediatric
885
patient: Hemofiltration versus hemodialysis. Am J Kidney
Dis 30:S84-S88, 1997 (suppl 4)
43. Paganini EP, Halstenberg WK, Goormastic M: Risk
modeling in acute renal failure requiring dialysis: The intro-
duction of a new model. Clin Nephrol 46:206-211, 1996
44. Gokal R: Quality of life in patients undergoing renal
replacement therapy. Kidney Int Suppl 40:S23-S27, 1993
45. de Wit GA, Ramsteijn PG, de Charro FT: Economic
evaluation of end stage renal disease treatment. Health
Policy 44:215-232, 1998
46. Churchill DN, Torrance GW, Taylor DW, et al: Mea-
surement of quality of life in end-stage renal disease: The
time trade-off approach. Clin Invest Med 10:14-20, 1987
47. Manns BJ, Taub KJ, Donaldson C: Economic evalua-
tion and end-stage renal disease: From basics to bedside.
Am J Kidney Dis 36:12-28, 2000
48. Culleton BF, Larson MG, Wilson PW, Evans JC,
Parfrey PS, Levy D: Cardiovascular disease and mortality in
a community-based cohort with mild renal insufficiency.
Kidney Int 56:2214-2219, 1999
49. Moher D, Cook DJ, Eastwood S, Olkin I, Rennie D,
Stroup DF: Improving the quality of reports of meta-
analyses of randomised controlled trials: The QUOROM
statement. Quality of Reporting of Meta-Analyses. Lancet
354:1896-1900, 1999
50. Centre for Reviews and Dissemination: Undertaking
Systematic Reviews of Research on Effectiveness. CRD
Guidelines for Those Carrying Out or Commissioning Re-
views. CRD report no. 4 (ed 2). York, NHS, Centre for
Reviews and Dissemination, University of York, 2001, pp
I.0.1-III.8.16
51. Karsou SA, Jaber BL, Pereira BJ: Impact of intermit-
tent hemodialysis variables on clinical outcomes in acute
renal failure. Am J Kidney Dis 35:980-991, 2000
52. Kellum JA, Angus DC, Johnson JP, et al: Continuous
vs intermittent renal replacement therapy: A meta-analysis.
Intensive Care Med 28:29-37, 2002
53. Davenport A: Renal replacement therapy in the patient
with acute brain injury. Am J Kidney Dis 37:457-466, 2001
54. DuBose TD, Warnock DG, Mehta RL, et al: Acute
renal failure in the 21st century: Recommendations for
management and outcomes assessment. Am J Kidney Dis
29:793-799, 1997
55. McAuley L, Pham B, Tugwell P, Moher D: Does the
inclusion of grey literature influence estimates of interven-
tion effectiveness reported in meta-analyses? Lancet 356:
1228-1231, 2000
56. LeLorier J, Gregoire G, Benhaddad A, Lapierre J,
Derderian F: Discrepancies between meta-analyses and sub-
sequent large randomized, controlled trials. N Engl J Med
337:536-542, 1997
57. Hoyt DB: CRRT in the area of cost containment: Is it
justified? Am J Kidney Dis 30:S102-S104, 1997 (suppl 4)
58. Moreno L, Heyka RJ, Paganini EP: Continuous renal
replacement therapy: Cost considerations and reimburse-
ment. Semin Dial 9:209-214, 1996
59. Hebert PC, Wells G, Blajchman MA, et al: A multi-
center, randomized, controlled clinical trial of transfusion
requirements in critical care. Transfusion Requirements in
Critical Care Investigators, Canadian Critical Care Trials
Group. N Engl J Med 340:409-417, 1999