The new england journal of medicine

original article

A Comparison of Albumin and Saline for Fluid

Resuscitation in the Intensive Care Unit
The SAFE Study Investigators*

abstract

background
It remains uncertain whether the choice of resuscitation fluid for patients in intensive The Saline versus Albumin Fluid Evalua-
care units (ICUs) affects survival. We conducted a multicenter, randomized, double-blind tion (SAFE) Study is a collaboration of the
Australian and New Zealand Intensive Care
trial to compare the effect of fluid resuscitation with albumin or saline on mortality in a Society Clinical Trials Group, the Australian
heterogeneous population of patients in the ICU. Red Cross Blood Service, and the George
Institute for International Health. The writ-
ing committee (Simon Finfer, M.B., B.S.,
methods Rinaldo Bellomo, M.B., B.S., M.D., Neil
We randomly assigned patients who had been admitted to the ICU to receive either Boyce, M.B., B.S., Ph.D., Julie French, R.N.,
4 percent albumin or normal saline for intravascular-fluid resuscitation during the next John Myburgh, M.B., B.Ch., Ph.D., and Ro-
byn Norton, Ph.D., M.P.H.) takes respon-
28 days. The primary outcome measure was death from any cause during the 28-day sibility for the content of this article. Address
period after randomization. reprint requests to Dr. Finfer at ANZICS
CTG, Level 3, 10 Ievers St., Carlton, VIC
3053, Australia, or at [email protected].
results
Of the 6997 patients who underwent randomization, 3497 were assigned to receive al- *The Saline versus Albumin Fluid Evalua-
bumin and 3500 to receive saline; the two groups had similar baseline characteristics. tion (SAFE) Study investigators are list-
ed in the Appendix.
There were 726 deaths in the albumin group, as compared with 729 deaths in the saline
group (relative risk of death, 0.99; 95 percent confidence interval, 0.91 to 1.09; P=0.87). N Engl J Med 2004;350:2247-56.
The proportion of patients with new single-organ and multiple-organ failure was sim- Copyright © 2004 Massachusetts Medical Society.

ilar in the two groups (P=0.85). There were no significant differences between the
groups in the mean (±SD) numbers of days spent in the ICU (6.5±6.6 in the albumin
group and 6.2±6.2 in the saline group, P=0.44), days spent in the hospital (15.3±9.6
and 15.6±9.6, respectively; P=0.30), days of mechanical ventilation (4.5±6.1 and
4.3±5.7, respectively; P=0.74), or days of renal-replacement therapy (0.5±2.3 and
0.4±2.0, respectively; P=0.41).

conclusions
In patients in the ICU, use of either 4 percent albumin or normal saline for fluid resus-
citation results in similar outcomes at 28 days.

n engl j med 350;22 www.nejm.org may 27, 2004 2247

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 The new england journal of medicine

sion and exclusion criteria are given in Table S1 of

t he administration of intrave-
nous fluids to maintain or increase intra-
vascular volume is a common intervention
in the intensive care unit (ICU), but there is uncer-
tainty whether the choice of fluid significantly influ-
the Supplementary Appendix (available with the full
text of this article at www.nejm.org). A detailed de-
scription of the study design has been published
elsewhere.8
ences patients’ outcomes.1-7 In particular, no ade- The study protocol was approved by the ethics
quately powered randomized, controlled trials have committees of the University of Sydney and of each
examined the effect of fluid choice on the survival participating institution. Written informed consent
of patients in the ICU. In the absence of such trials, was to be obtained from all competent patients; in
a number of meta-analyses have examined how the cases in which prior consent could not be obtained
choice of crystalloid or colloid solution and of al- from the patient because of critical illness or the use
bumin-containing or albumin-free fluid affects sur- of sedative or anesthetic drugs, consent could be de-
vival in critically ill patients and in patients who are layed, and a provision for delayed consent was ap-
less severely ill.1-3,7 A meta-analysis published by plied. In such cases, the patient or his or her surro-
the Cochrane Injuries Group Albumin Reviewers in- gate decision maker was informed of the study as
cluded 24 studies involving a total of 1419 patients soon as practicable, and consent was sought to con-
and suggested that the administration of albumin- tinue the study procedures and to access the partic-
containing fluids resulted in a 6 percent increase in ipant’s medical records for study-related data. The
the absolute risk of death when compared with the patients or their legal surrogates were informed of
administration of crystalloid solutions.1 However, their right to request that the study procedures be
a subsequent meta-analysis of 55 trials involving a discontinued and their right to refuse the study-
total of 3504 patients examined the effect of resus- related use of their medical records.
citation with albumin-containing fluid on the risk Eligible patients were randomly assigned to re-
of death in a general population of patients and did ceive either 4 percent albumin (Albumex, CSL) or
not find a significant increase in the risk of death.3 normal saline, with the random assignments strat-
The conflicting results of such meta-analyses ified according to institution and according to
have left many clinicians unsure about the effect of whether there was a diagnosis of trauma on admis-
albumin-containing fluids on survival in critically ill sion to the ICU. Randomization was carried out
patients. To address this uncertainty, we conducted centrally with the use of a minimization algorithm,
the Saline versus Albumin Fluid Evaluation (SAFE) and the service was accessed on the Internet through
Study in 16 ICUs in Australia and New Zealand. We a secure Web site. Study fluids were supplied in
tested the hypothesis that when 4 percent albumin identical 500-ml bottles, and blinding was ensured
is compared with 0.9 percent sodium chloride (nor- through the use of specially designed masking
mal saline) for intravascular-fluid resuscitation in cartons and specially designed and manufactured
patients in the ICU, there is no difference in the administration sets.8 The effectiveness of the blind-
28-day rate of death from any cause. ing was confirmed in a formal study before the trial
was initiated. The treating clinicians determined the
methods amount and rate of fluid administration according
to each patient’s clinical status and response to treat-
study design and treatment protocol ment. The allocated study treatment was to be used
Patients 18 years of age or older who had been ad- for all fluid resuscitation in the ICU until death or
mitted to the closed, multidisciplinary ICUs of 16 discharge or until 28 days after randomization. The
academic tertiary hospitals in Australia and New administration of intravenous fluids outside the ICU
Zealand between November 2001 and June 2003 was not controlled.
were assessed for eligibility for the study. Eligible In addition to the study fluid, patients received
patients were those whom the treating clinician maintenance fluids, specific replacement fluids, en-
judged to require fluid administration to maintain teral or parenteral nutrition, and blood products at
or increase intravascular volume, with this decision the discretion of the treating clinicians. The moni-
supported by the fulfillment of at least one objective toring of central venous pressure, pulmonary-artery
criterion. Patients admitted to the ICU after cardiac catheterization, and all other aspects of patient care
surgery, after liver transplantation, or for the treat- were performed at the discretion of the treating cli-
ment of burns were excluded. Details of the inclu- nicians.

2248 n engl j med 350;22 www.nejm.org may 27 , 2004

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 albumin vs. saline for fluid resuscitation in the icu

baseline assessment and follow-up study and data management

data collection Two preplanned interim analyses were performed
Data collected at baseline included the Acute Phys- by an independent statistician after recruitment of
iology and Chronic Health Evaluation II score,9 as the first 2333 patients (33 percent of the planned
well as information pertaining to diagnostic crite- total) and the first 4666 patients (67 percent), and
ria for severe sepsis10 (Table S2 of the Supplemen- the results were reviewed by the independent data-
tary Appendix) and for the acute respiratory distress monitoring committee.
syndrome.11 Patients were identified as having trau- The George Institute for International Health at
matic brain injury at baseline if they had a history of the University of Sydney performed the data man-
trauma, a Glasgow Coma Score12 while not sedated agement, the site management, and the data analy-
of less than 14, and an abnormality consistent with sis, independently of the funding agencies. The
traumatic brain injury on a computed tomographic manuscript was prepared by the writing committee
scan of the head. The cardiovascular, respiratory, and was revised by the study investigators, who ap-
renal, hematologic, and hepatic components of proved the final manuscript.
the Sequential Organ-Failure Assessment (SOFA)
score,13 as described in Table S3 of the Supplemen- statistical analysis
tary Appendix, were recorded at the time of random- The trial was designed to enroll 7000 patients, there-
ization, daily for the next seven days, and then every by providing a power of 90 percent to detect a 3 per-
third day until discharge from the ICU or until day cent difference in absolute mortality rates between
28. After randomization, the heart rate, central the two groups from an estimated baseline mortal-
venous pressure, mean arterial blood pressure, vol- ity rate of 15 percent. The data were exported from
ume of study fluid administered, volume of non- the study database and analyzed with the use of
study fluid and blood products administered, net SPSS software (version 11.5). All analyses were per-
fluid balance (calculated as the total fluid input formed on an intention-to-treat basis. Where data
minus the total fluid output), use of mechanical were missing, we report the number of available ob-
ventilation, and use of renal-replacement therapy servations, and we make no assumptions about the
(intermittent or continuous hemodialysis, hemo- missing data.
filtration, or hemodiafiltration) were recorded dai- Proportions were compared by means of the
ly until discharge from the ICU or death or until chi-square test or Fisher’s exact test, and continu-
day 28. ous variables were compared by means of unpaired
t-tests. The results of comparisons of event rates in
outcome measures the two groups are presented as relative risks with
The primary outcome measure was death from any 95 percent confidence intervals. Survival times were
cause within 28 days after randomization. Second- compared by means of the log-rank test and are pre-
ary outcome measures were the survival time during sented as Kaplan–Meier curves without adjustment
the first 28 days, the proportion of patients who had for baseline covariates. Heterogeneity of treatment
one, two, three, four, or five new organ failures (de- effects among subgroups was assessed with the use
fined as a documented change in the cardiovascular, of the test for a common relative risk.14
respiratory, renal, hematologic, or hepatic compo-
nent of the SOFA score from 0, 1, or 2 at baseline to results
3 or 4 during the ICU stay, where higher scores in-
dicate increasingly severe organ dysfunction), the study patients
duration of mechanical ventilation, the duration of Seven thousand random assignments to a study
renal-replacement therapy, and the duration of the treatment were made (3499 to the albumin group
ICU and hospital stay. and 3501 to the saline group). Three patients mis-
Death from any cause within 28 days after ran- takenly underwent randomization twice within 28
domization was also examined in six predefined days; they were followed for 28 days beginning at
subgroups according to the presence or absence of the time of the first randomization, and for purpos-
trauma, the presence or absence of severe sepsis, es of data analysis were considered part of the group
and the presence or absence of the acute respiratory to which they were first assigned. Thus, the study
distress syndrome at baseline. population comprised 6997 patients, 3497 of whom

n engl j med 350;22 www.nejm.org may 27, 2004 2249

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 The new england journal of medicine

were assigned to receive albumin and 3500 of whom

Table 1. Baseline Characteristics of the Patients.*
were assigned to receive saline. The majority of the
Characteristic Albumin Group Saline Group patients (6628 [94.7 percent] — 3312 of those in
the albumin group [94.7 percent] and 3316 of those
Age — yr 58.6±19.1 58.5±18.7
in the saline group [94.7 percent]) were enrolled
Female sex — no. (%) 1424 (40.7) 1376 (39.3) with the use of the provision for delayed consent.
Reason for admission to ICU — no. (%) Delayed consent was obtained from the patient in
Surgical 1473 (43.0) 1465 (42.8)
Medical 1955 (57.0) 1958 (57.2)
2713 cases (38.8 percent) and from a surrogate de-
cision maker (a relative, a legally recognized surro-
Source of admission to ICU — no. (%)
gate, or an institutional ethics committee) in the re-
Emergency department 948 (27.7) 977 (28.5) maining cases. Prior consent was obtained from the
Hospital floor 614 (17.9) 573 (16.7) patient in 45 cases (0.6 percent) and from a surro-
Another ICU 63 (1.8) 66 (1.9) gate decision maker in 335 cases (4.8 percent).
Another hospital 323 (9.4) 341 (10.0) At baseline, the only statistically significant dif-
ference between the two groups was a higher mean
Operating room (emergency surgery) 801 (23.4) 780 (22.8)
(±SD) central venous pressure in the albumin group
Operating room (elective surgery) 662 (19.3) 678 (19.8)
(9.0±4.7 mm Hg, vs. 8.6±4.6 mm Hg in the saline
Same ICU (readmission) 17 (0.5) 8 (0.2) group; P=0.03). The baseline characteristics of the
Predefined subgroups — no. (%) 6997 patients are summarized in Table 1, and their
Trauma 597 (17.4) 590 (17.2) progress through the study is summarized in Fig-
Severe sepsis 603 (18.1) 615 (18.4)
Acute respiratory distress syndrome 61 (1.8) 66 (1.9) ure S1 of the Supplementary Appendix.
APACHE II score† 18.7±7.9 19.0±8.0
Study fluid was administered to all but 197 pa-
tients (2.8 percent), including 90 in the albumin
Physiological variables
group and 107 in the saline group. Resuscitation
Heart rate — beats/min 91.4±23.5 92.3±23.5 fluids in addition to the allocated study fluid were
Mean arterial pressure — mm Hg 77.8±16.4 78.2±16.3 administered to 309 patients in the albumin group
Central venous pressure — mm Hg 9.0±4.7 8.6±4.6‡ (8.8 percent) and 375 in the saline group (10.7 per-
Urine output — ml/hr 89.7±132.4 95.0±161.4 cent). The most common reason for the adminis-
tration of nonstudy resuscitation fluid was error (in
Serum albumin — g/liter 27.4±7.8 27.7±7.9
189 patients in the albumin group [5.4 percent] and
Organ failure— no. (%)§
190 in the saline group [5.4 percent]). Clinicians’
No failure 1962 (57.2) 1885 (55.1) preference for a specific nonstudy resuscitation flu-
1 organ 1075 (31.4) 1148 (33.5) id was the reason for its administration in 68 pa-
2 organs 335 (9.8) 329 (9.6) tients in the albumin group (1.9 percent) and 103
3 organs 50 (1.5) 57 (1.7) in the saline group (2.9 percent). At the completion
of the trial, information on vital status 28 days after
4 organs 5 (0.1) 4 (0.1)
randomization was unavailable for 67 patients (1.0
5 organs 1 (<0.1) 0
percent), including 26 in the albumin group and 41
Mechanical ventilation — no. (%) 2186 (63.8) 2217 (64.8) in the saline group. In 56 of these 67 cases, vital sta-
Renal-replacement therapy — no. (%) 45 (1.3) 41 (1.2) tus was missing because the patient or his or her le-
Albumin in previous 72 hr — no. (%) 127 (3.7) 135 (3.9) gal surrogate had withheld or withdrawn consent.

* Plus–minus values are means ±SD. Percentages were calculated according to fluids administered and treatment effects
the number of patients for whom data were available: for sex, 3497 in the albu- On each of the first three study days, the patients
min group and 3500 in the saline group; for severe sepsis, 3339 in the albumin
group and 3338 in the saline group; and for all the other variables, 3428 in the who had been randomly assigned to receive albumin
albumin group and 3423 in the saline group. Because of rounding, not all per- received significantly less study fluid than did those
centages total 100. ICU denotes intensive care unit, and APACHE II Acute assigned to saline, resulting in a significantly great-
Physiology and Chronic Health Evaluation II.
† Higher scores on APACHE II indicate more severe illness. er net positive fluid balance in the saline group on
‡P=0.03 for the comparison with the value in the albumin group (without cor- each of those days (Table 2). The ratios of the volume
rection for multiple-hypothesis testing). of albumin to the volume of saline administered
§ Organ failure was defined as a Sequential Organ-Failure Assessment score13
of 3 or 4 for any individual organ system. during the first four days were as follows: 1:1.3 on
day 1, 1:1.6 on day 2, 1:1.3 on day 3, and 1:1.2 on

2250 n engl j med 350;22 www.nejm.org may 27 , 2004

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 albumin vs. saline for fluid resuscitation in the icu

Table 2. Fluids Administered and Physiological Effects of Treatment.*

Variable Albumin Group Saline Group P Value†

No. of Patients Value No. of Patients Value
Study fluid (ml)
Day 1 3410 1183.9±973.6 3418 1565.3±1536.1 <0.001
Day 2 3059 602.7±892.7 3068 954.0±1484.4 <0.001
Day 3 2210 268.0±554.5 2202 348.3±753.5 0.03
Day 4 1686 192.3±427.0 1664 228.6±642.6 0.57
Nonstudy fluid (ml)
Day 1 3392 1459.4±1183.2 3405 1505.6±1254.3 0.30
Day 2 3051 2615.9±1372.5 3057 2707.3±1435.7 0.009
Day 3 2199 2618.5±1346.5 2191 2660.9±1319.3 0.15
Day 4 1680 2691.5±1228.7 1656 2707.7±1255.4 0.36
Packed red cells (ml)
Day 1 3411 97.8±360.7 3415 71.7±296.8 <0.001
Day 2 3066 106.5±321.4 3074 61.1±235.2 <0.001
Day 3 2217 59.8±225.5 2210 49.5±190.8 0.30
Day 4 1692 43.6±167.5 1668 46.0±189.0 0.77
Net positive fluid balance (ml)
Day 1 3363 1543.6±1619.7 3382 1990.5±2061.7 <0.001
Day 2 3044 1015.3±1826.9 3052 1505.1±2215.9 <0.001
Day 3 2190 422.1±1633.3 2182 553.0±1732.3 0.007
Day 4 1671 137.2±1491.0 1649 155.7±1650.6 0.70
Mean arterial pressure (mm Hg)
Day 1 3406 81.4±14.4 3408 80.9±14.5 0.14
Day 2 3068 84.4±15.1 3075 84.2±15.7 0.49
Day 3 2215 87.2±15.3 2209 86.9±16.1 0.62
Day 4 1688 88.3±15.9 1666 88.4±16.3 0.87
Heart rate (beats/min)
Day 1 3398 88.0±20.2 3406 89.7±20.8 <0.001
Day 3 3071 88.5±19.5 3075 89.5±19.2 0.06
Day 3 2216 88.8±19.1 2213 89.7±18.8 0.10
Day 4 1691 89.5±18.9 1668 89.9±18.5 0.52
Central venous pressure (mm Hg)
Day 1 2204 11.2±4.8 2270 10.0±4.5 <0.001
Day 2 2095 11.6±4.9 2135 10.4±4.3 <0.001
Day 3 1531 11.4±4.8 1589 10.7±4.4 <0.001
Day 4 1221 11.1±4.8 1230 10.5±4.4 <0.001
Serum albumin (g/liter)
Day 1 2081 28.7±7.0 2061 24.7±6.5 <0.001
Day 2 2708 30.8±6.4 2703 24.5±5.9 <0.001
Day 3 1921 30.0±6.4 1905 23.6±5.6 <0.001
Day 4 1498 29.0±6.2 1478 23.1±5.5 <0.001

* Plus–minus values are means ±SD.

† P values are for the comparison between the two means for each variable at each time point.

day 4. The overall ratio of the volume of albumin to to receive saline. On day 2, patients in the saline
the volume of saline administered during the first group received a greater volume of nonstudy fluids
four days was approximately 1:1.4. Patients in the than did those in the albumin group (Table 2). After
two groups received similar volumes of other fluids day 4, there were no differences between the two
during the first four days, except on days 1 and 2, groups in the volume of study fluids administered.
when the patients in the albumin group received a There were no significant differences between the
greater volume of packed red cells than did those in groups in the mean arterial pressure measured at
the saline group; on average, during the first four the end of each of the first four days of the study. The
days, patients assigned to receive albumin received patients assigned to receive albumin had a lower
71.0 ml more packed red cells than those assigned heart rate at the end of the first day than those as-

n engl j med 350;22 www.nejm.org may 27, 2004 2251

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 The new england journal of medicine

signed to receive saline. Central venous pressure compared with those assigned to receive saline was
was significantly higher in the albumin group than 0.99 (95 percent confidence interval, 0.91 to 1.09;
in the saline group at all time points during the first P=0.87). At 28 days, 111 patients in the albumin
four days, and the serum albumin concentration was group (3.2 percent) and 87 patients in the saline
higher in the albumin group throughout the study group (2.5 percent) remained in the ICU (relative
period (Table 2). risk, 1.27; P=0.09); 793 (22.8 percent) and 848
(24.5 percent), respectively, remained in the hospi-
outcomes tal (relative risk, 0.93; 95 percent confidence inter-
Within 28 days after randomization, 726 of 3473 pa- val, 0.86 to 1.01; P=0.10) (Table 3). There was no
tients in the albumin group (20.9 percent) and 729 significant difference in survival times between the
of 3460 patients in the saline group (21.1 percent) two groups (Fig. 1).
had died. For the albumin group as compared with The number of patients who had new single-
the saline group, the absolute difference in mortal- organ or multiple-organ failure, assessed according
ity was –0.2 percent (95 percent confidence inter- to their SOFA scores, was similar in the two groups
val, –2.1 to +1.8 percent). The relative risk of death (P=0.85 by Fisher’s exact test) (Table 3). During the
among patients assigned to receive albumin as 28-day study period the mean length of stay in the

Table 3. Primary and Secondary Outcomes.*

Relative Risk Absolute Difference

Outcome Albumin Group Saline Group (95% CI) (95% CI) P Value
Status at 28 days — no./total no. (%)
Dead 726/3473 (20.9) 729/3460 (21.1) 0.99 (0.91 to 1.09) 0.87
Alive in ICU 111/3473 (3.2) 87/3460 (2.5) 1.27 (0.96 to 1.68) 0.09
Alive in hospital† 793/3473 (22.8) 848/3460 (24.5) 0.93 (0.86 to 1.01) 0.10
Length of stay in ICU — days 6.5±6.6 6.2±6.2 0.24 (–0.06 to 0.54) 0.44
Length of stay in hospital — days† 15.3±9.6 15.6±9.6 –0.24 (–0.70 to 0.21) 0.30
Duration of mechanical ventilation — 4.5±6.1 4.3±5.7 0.19 (–0.08 to 0.47) 0.74
days
Duration of renal-replacement therapy 0.48±2.28 0.39±2.0 0.09 (–0.0 to 0.19) 0.41
— days
New organ failure — no. (%)‡ 0.85§
No failure 1397 (52.7) 1424 (53.3)
1 organ 795 (30.0) 796 (29.8)
2 organs 369 (13.9) 361 (13.5)
3 organs 68 (2.6) 75 (2.8)
4 organs 18 (0.7) 17 (0.6)
5 organs 2 (0.1) 0
Death within 28 days according to sub-
group — no./total no. (%)
Patients with trauma 81/596 (13.6) 59/590 (10.0) 1.36 (0.99 to 1.86) 0.06
Patients with severe sepsis 185/603 (30.7) 217/615 (35.3) 0.87 (0.74 to 1.02) 0.09
Patients with acute respiratory dis- 24/61 (39.3) 28/66 (42.4) 0.93 (0.61 to 1.41) 0.72
tress syndrome

* Plus–minus values are means ±SD. CI denotes confidence interval, and ICU intensive care unit.
† The data include the numbers of patients in the ICU or the length of stay in the ICU.
‡ Data were available for 2649 patients in the albumin group and 2673 patients in the saline group. New organ failure was defined as a Sequential
Organ-Failure Assessment score13 of 0, 1, or 2 in any individual organ system at baseline, followed by an increase in the score to 3 or 4 in the
same system.
§ The P value pertains to the comparison between the albumin and saline groups in the numbers of patients who had no new organ failure or
new failure of one, two, three, four, or five organs.

2252 n engl j med 350;22 www.nejm.org may 27 , 2004

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 albumin vs. saline for fluid resuscitation in the icu

ICU was 6.5±6.6 days in the albumin group and without this syndrome was 1.00 (P=0.74 by the test
6.2±6.2 days in the saline group (P=0.44). The mean for a common relative risk).
length of stay in the hospital was 15.3±9.6 days and
15.6±9.6 days, respectively (P=0.30). The numbers discussion
of days of mechanical ventilation and days of renal-
replacement therapy were similar in the two groups In this randomized trial, we found that the use of
(Table 3). 4 percent albumin or normal saline for intravascu-
lar volume resuscitation in a heterogeneous popu-
subgroup analyses lation of patients in the ICU resulted in equivalent
During the 28-day study period, the relative risk of rates of death from any cause during the 28-day
death among patients with trauma in the albumin study period. Requirements for mechanical venti-
group as compared with such patients in the saline lation and renal-replacement therapy, time spent in
group was 1.36; the corresponding relative risk of the ICU and in the hospital during the 28-day study
death among patients without trauma was 0.96 period, and the time until death (among the patients
(P=0.04 by the test for a common relative risk). This who died) were also equivalent. The proportion of
difference in the relative risk of death was due to the patients in the two groups in whom new single-
greater number of patients with trauma and an as- organ or multiple-organ failure developed were sim-
sociated brain injury who died after random assign- ilar. Our findings do not support the results of the
ment to albumin as opposed to saline: 59 of 241 Cochrane Injuries Group Albumin Reviewers’ meta-
such patients in the albumin group died (24.5 per- analysis, which suggested that the use of albumin
cent), as compared with 38 of 251 such patients in was associated with an increased mortality rate
the saline group (15.1 percent) (relative risk, 1.62; among critically ill patients.1
95 percent confidence interval, 1.12 to 2.34; P= Our study was conducted as a double-blind, ran-
0.009). Among patients who had trauma without domized trial. Albumin and saline are not consid-
brain injury, there was no difference between the ered equipotent intravascular volume expanders, but
groups in terms of mortality: 22 such patients in their relative potencies have not previously been
the albumin group (6.2 percent) and 21 in the sa- examined in an adequately powered, blinded trial.
line group (6.2 percent) died (relative risk, 1.00; 95 In our study, patients who were resuscitated with al-
percent confidence interval, 0.56 to 1.79; P=1.00). bumin received less fluid than those who were re-
Among all the patients who had trauma (596 in the suscitated with saline. During the first four days,
albumin group and 590 in the saline group), there
were 81 (13.6 percent) deaths in the albumin group
and 59 (10.0 percent) in the saline group (relative 1.0
risk, 1.36; 95 percent confidence interval, 0.99 to Albumin
Saline
1.86; P=0.06) (Fig. 2 and Table 3).
In a subgroup analysis of patients with severe 0.9
Probability of Survival

sepsis, the relative risk of death during the 28-day

study period among those randomly assigned to re- 0.8
ceive albumin as opposed to saline was 0.87, as
compared with a corresponding relative risk of 1.05
among patients without severe sepsis (P=0.06 by 0.7
the test for a common relative risk). Of the 603 pa-
tients with severe sepsis who had been assigned to
0.6
receive albumin, 185 (30.7 percent) died, and of the
615 patients with severe sepsis who had been as- 0.0
signed to receive saline, 217 (35.3 percent) died (rel- 0 4 8 12 16 20 24 28
ative risk, 0.87; 95 percent confidence interval, 0.74 Days
to 1.02; P=0.09) (Table 3). In a subgroup analysis of
patients with the acute respiratory distress syn- Figure 1. Kaplan–Meier Estimates of the Probability of Survival.
drome, the relative risk of death among those as- P=0.96 for the comparison between patients assigned to receive albumin and
signed to receive albumin as opposed to saline was those assigned to receive saline.
0.93; the corresponding relative risk among patients

n engl j med 350;22 www.nejm.org may 27, 2004 2253

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 The new england journal of medicine

Albumin Saline
Patients Group Group Relative Risk (95% CI)
no. of deaths/total no.
Overall 726/3473 729/3460 0.99 (0.91–1.09)
Trauma
Yes 81/596 59/590 1.36 (0.99–1.86)
No 641/2831 666/2830 0.96 (0.88–1.06)
Severe sepsis
Yes 185/603 217/615 0.87 (0.74–1.02)
No 518/2734 492/2720 1.05 (0.94–1.17)
ARDS
Yes 24/61 28/66 0.93 (0.61–1.41)
No 697/3365 697/3354 1.00 (0.91–1.09)

0.5 1.0 2.0

Albumin Saline
Better Better

Figure 2. Relative Risk of Death from Any Cause among All the Patients and among the Patients in the Six Predefined
Subgroups.
The size of each symbol indicates the relative number of events in the given group. The horizontal bars represent the
confidence intervals (CI). ARDS denotes the acute respiratory distress syndrome.

the ratio of albumin administered to saline admin- excess of 3.0 units of packed red cells. This was as-
istered was approximately 1:1.4. However, there sociated with an increase in in-hospital mortality of
was no significant difference in mean arterial pres- 5.9 percentage points.15 During the first four days
sure between the groups, and the differences in of our study, the excess volume of fluid transfused
central venous pressure and heart rate were small. in the albumin group averaged 71.0 ml per patient
Thus, we believe that the patients in the two groups (less than one quarter of a unit). Accordingly, we do
were resuscitated to similar and acceptable end not believe this small excess in transfused volume
points. influenced the results.
Our study was also a large, pragmatic study in a Given that our study had insufficient power to
population of patients subject to a large number of detect small but important differences in mortality
concurrent interventions. We did not collect infor- among the predefined subgroups, the results pro-
mation on all concurrent interventions performed in vide only limited evidence that the treatment effects
the ICU. However, randomization was stratified ac- varied among these subgroups. The finding that pa-
cording to participating institution, so that each in- tients with trauma might benefit more from resus-
stitution treated equal numbers of patients assigned citation with saline than patients without trauma
to saline or to albumin. As a result, we do not believe appears to be consistent with the results of a meta-
that an imbalance in concurrent interventions could analysis by Choi et al., who suggested that colloid
have influenced the results. resuscitation was associated with increased mortal-
Patients who were assigned to albumin received ity in patients with trauma.2 In our study, however,
a significantly greater volume of packed red cells the increased relative risk of death among patients
during the first two days of the study. The reasons with trauma as compared with those without trau-
for this difference remain speculative but may in- ma resulted from a small excess number of deaths
clude greater hemodilution with albumin than with among patients who had trauma with brain injury,
saline or increased blood loss with albumin due to whereas the meta-analysis by Choi et al. did not in-
transient alterations in coagulation. In a study of clude studies in patients with brain injury.2
transfusion requirements in critically ill patients In our study, the difference in mortality between
conducted by Hébert and colleagues, a liberal trans- the albumin and saline groups among patients with
fusion policy resulted in the administration of an trauma involving brain injury should be interpreted

2254 n engl j med 350;22 www.nejm.org may 27 , 2004

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 albumin vs. saline for fluid resuscitation in the icu

with caution. Patients with traumatic brain injury suscitation in a heterogeneous population of pa-
constituted only 7 percent of the study population, tients in the ICU. Whether either albumin or saline
and the excess number of deaths in the albumin confers benefit in more highly selected populations
group was only 21. In large studies, such subgroup of critically ill patients requires further study. Ac-
differences frequently occur by chance.16 In addi- cording to the current state of knowledge, factors
tion, the rate of death from any cause over a 28-day that may influence the choice of resuscitation fluid
period is not considered the most appropriate out- for a critically ill patient include the individual clini-
come measurement with which to assess treatment cian’s preference, the tolerability of the treatment,
effects in patients with brain injury. Assessment of its safety, and its cost.
mortality and functional neurologic status at least Supported by the Auckland District Health Board, Middlemore
Hospital, and the Health Research Council of New Zealand — all in
six months after injury is recommended.17 In con- New Zealand; and by the Australian Commonwealth Department of
trast with our findings in patients with trauma, the Health and Aged Care, CSL (Melbourne, Victoria), the Australian
comparison of the relative risk of death among pa- National Health and Medical Research Council, the Health Depart-
ment of Western Australia, the New South Wales Health Depart-
tients with severe sepsis and those without severe ment, the Northern Territory Health Services, the Queensland
sepsis provides limited evidence of a treatment effect Health Services Department, Royal Hobart Hospital (Tasmania), the
that favors albumin in patients with severe sepsis. South Australian Department of Human Services, and the Victorian
Department of Human Services — all in Australia.
It should be noted that such differences between Dr. A. Davies and Dr. D. Stephens report owning shares in CSL.
subgroups frequently occur by chance and that only We are indebted to the nursing and medical staff of the intensive
specifically designed and appropriately powered care units and the blood banks of the participating institutions; to
the staff of the Australian Red Cross Blood Service and the New
studies can determine whether any such treatment Zealand Blood Service, whose enthusiasm and hard work made the
effects are real. SAFE Study possible; to the staff of Tuta Healthcare, who manufac-
In conclusion, our study provides evidence that tured and supplied the fluid-administration sets; and to the manag-
ers and staff of CSL, who manufactured, blinded, packed, and sup-
albumin and saline should be considered clinically plied the saline and albumin used in the study.
equivalent treatments for intravascular volume re-

ap p e n d i x
The Saline versus Albumin Fluid Evaluation (SAFE) Study investigators are as follows: Writing Committee — S. Finfer (Chair), R. Bellomo, N.
Boyce, J. French, J. Myburgh, and R. Norton. Management Committee — R. Norton (Chair), J. French (Senior Project Manager), R. Bellomo, S.
Finfer, J. Myburgh, G. Doig, M. Hayek, and S. O’Donnell. Steering Committee — S. Finfer (Chair), A. Bell, R. Bellomo, N. Boyce, D. Blythe, J.
Cade, M. Chapman, L. Cole, J. Cooper, A. Davies, C. French, J. French, C. Joyce, C. McArthur, S. MacMahon, J. Myburgh, B. Neal, R. Norton,
J. Presneill, P. Saul, I. Seppelt, D. Stephens, A. Turner, A. Williams, and C. Woolfe. External Safety and Data Monitoring Committee — R. Peto
(Chair), P. Sandercock, C. Sprung, and D. Young. Statistical Analysis (George Institute for International Health, University of Sydney, Sydney,
N.S.W., Australia) — S.K. Lo, S. Sivarajasingham, L. Francis, M. Woodward. Site investigators (all in Australia unless otherwise specified) —
Alfred Hospital, Melbourne: J. Charlton, J. Cooper, A. Davies, C. Harry, L. Higgins, K. Moulden, and S. Vallance. Auckland Hospital, Auck-
land, New Zealand: J. Chadderton, L. Newby, and C. McArthur. Austin and Repatriation Medical Centre, Melbourne: S. Bates, R. Bellomo,
D. Goldsmith, and A. Voss. Australian Red Cross Blood Service, Melbourne: N. Boyce. Fremantle Hospital, Fremantle: D. Blythe and A. Pal-
ermo. George Institute for International Health, University of Sydney, Sydney: L. Francis, J. French, M. Hayek, K. Jayne, S. MacMahon, M.
Merai, B. Neal, R. Norton, S. Pandey, S. O’Donnell, M. Schmidt, S. Sivarajasingham, and M. Woodward. John Hunter Hospital, Newcastle:
R. Carroll, B. McFadyen, and P. Saul. Middlemore Hospital, Auckland, New Zealand: J. Clarke, J. Powell, A. Williams, and J. Tai. Nepean
Hospital, Penrith: L. Cole, I. Hynesova, I. Seppelt, and L. Weisbrodt. Princess Alexandra Hospital, Brisbane, Queensland: L. Bradley, C.
Joyce, T. Kelly, A. Limpus, and R. Moore. Royal Adelaide Hospital, Adelaide: M. Chapman, S. Creed, S. Kaplan, J. Rivett. Royal Darwin Hos-
pital, Darwin: D. Stephens and J. Thomas. Royal Hobart Hospital, Hobart: A. Bell, K. Marsden, and A. Turner. Royal Melbourne Hospital,
Melbourne: C. Boyce, J. Cade, B. Howe, J. Presneill, and M. Robertson. Royal North Shore Hospital, Sydney: G. Doig, S. Finfer, A. O’Con-
nor, J. Potter, and N. Ramakrishnan. Royal Prince Alfred Hospital, Sydney: C. Powell, D. Rajbhandari, and C. Woolfe. St. George Hospital,
Sydney: K. Girling, M. Hodgetts, A. Jovanovska, and J. Myburgh. Western Hospital, Melbourne: C. French and L. Little.

references
1. Cochrane Injuries Group Albumin Re- vival after human albumin administration: ume replacement regimens: what did we
viewers. Human albumin administration a meta-analysis of randomized, controlled learn from the past three years? Anesth
in critically ill patients: systematic review of trials. Ann Intern Med 2001;135:149-64. Analg 2003;97:1595-604.
randomised controlled trials. BMJ 1998; 4. Cook D, Guyatt G. Colloid use for fluid 7. Bunn F, Alderson P, Hawkins V. Colloid
317:235-40. resuscitation: evidence and spin. Ann Intern solutions for fluid resuscitation. Cochrane
2. Choi PT, Yip G, Quinonez LG, Cook DJ. Med 2001;135:205-8. Database Syst Rev 2003;1:CD001319.
Crystalloids vs. colloids in fluid resuscita- 5. Whether albumin therapy improves or 8. ANZICS Clinical Trials Group, Insti-
tion: a systematic review. Crit Care Med worsens survival of critically ill patients is tute for International Health SAFE Study
1999;27:200-10. not known. Ann Intern Med 2001;135:S-25. Investigators. The Saline vs. Albumin Fluid
3. Wilkes MM, Navickis RJ. Patient sur- 6. Boldt J. New light on intravascular vol- Evaluation (SAFE) Study (ISRCTN76588266):

n engl j med 350;22 www.nejm.org may 27, 2004 2255

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.
 albumin vs. saline for fluid resuscitation in the icu

design and conduct of a multi-centre, blind- The American-European Consensus Confer- sign and data analysis. Boca Raton, Fla.:
ed randomised controlled trial of intravenous ence on ARDS: definitions, mechanisms, rel- Chapman & Hall/CRC Press, 1999.
fluid resuscitation in critically ill patients. evant outcomes, and clinical trial coordina- 15. Hébert PC, Wells G, Blajchman MA, et
London: BMJ Publishing, 2004. (Accessed tion. Am J Respir Crit Care Med 1994;149: al. A multicenter, randomized, controlled
May 3, 2004, at http://bmj.com/cgi/content/ 818-24. clinical trial of transfusion requirements in
full/326/7389/559/DC1.) 12. Teasdale G, Jennett B. Assessment of critical care. N Engl J Med 1999;340:409-17.
9. Knaus WA, Draper EA, Wagner DP, Zim- coma and impaired consciousness: a practi- [Erratum, 1999;340:1056.]
merman JE. APACHE II: a severity of disease cal scale. Lancet 1974;2:81-4. 16. Collins R, MacMahon S. Reliable as-
classification system. Crit Care Med 1985; 13. Vincent JL, de Mendonca A, Cantraine F, sessment of the effects of treatment on mor-
13:818-29. et al. Use of the SOFA score to assess the in- tality and major morbidity. I. Clinical trials.
10. Bone RC, Balk RA, Cerra FB, et al. Defini- cidence of organ dysfunction/failure in in- Lancet 2001;357:373-80.
tions for sepsis and organ failure and guide- tensive care units: results of a multicenter, 17. Narayan RK, Michel ME, Ansell B, et al.
lines for the use of innovative therapies in prospective study. Crit Care Med 1998;26: Clinical trials in head injury. J Neurotrauma
sepsis. Chest 1992;101:1644-55. 1793-800. 2002;19:503-57.
11. Bernard GR, Artigas A, Brigham KL, et al. 14. Woodward M. Epidemiology: study de- Copyright © 2004 Massachusetts Medical Society.

posting presentations at medical meetings on the internet

Posting an audio recording of an oral presentation at a medical meeting on the
Internet, with selected slides from the presentation, will not be considered prior
publication. This will allow students and physicians who are unable to attend the
meeting to hear the presentation and view the slides. If there are any questions
about this policy, authors should feel free to call the Journal’ s Editorial Offices.

2256 n engl j med 350;22 www.nejm.org may 27 , 2004

Downloaded from www.nejm.org at ARIZONA HEALTH INFORMATION NETWORK on November 5, 2009 .

Copyright © 2004 Massachusetts Medical Society. All rights reserved.