ANNALS OF SURGERY

REVIEW Vol. 237, No. 3, 319–334

© 2003 Lippincott Williams & Wilkins, Inc.

Hypoalbuminemia in Acute Illness: Is There a

Rationale for Intervention?
A Meta-Analysis of Cohort Studies and Controlled Trials
Jean-Louis Vincent, MD, PhD, FCCM,* Marc-Jacques Dubois, MD,* Roberta J. Navickis, PhD,† and Mahlon M. Wilkes, PhD†

From the *Department of Intensive Care, Université Libre de Bruxelles, Hôpital Erasme, Brussels, Belgium, and †Hygeia
Associates, Grass Valley, California, U.S.A.

Objective Results
To determine whether hypoalbuminemia is an independent Hypoalbuminemia was a potent, dose-dependent indepen-
risk factor for poor outcome in the acutely ill, and to assess dent predictor of poor outcome. Each 10-g/L decline in se-
the potential of exogenous albumin administration for improv- rum albumin concentration significantly raised the odds of
ing outcomes in hypoalbuminemic patients. mortality by 137%, morbidity by 89%, prolonged intensive
care unit and hospital stay respectively by 28% and 71%, and
increased resource utilization by 66%. The association be-
Summary Background Data tween hypoalbuminemia and poor outcome appeared to be
Hypoalbuminemia is associated with poor outcomes in independent of both nutritional status and inflammation. Anal-
acutely ill patients, but whether this association is causal has ysis of dose-dependency in controlled trials of albumin ther-
remained unclear. Trials investigating albumin therapy to cor- apy suggested that complication rates may be reduced when
rect hypoalbuminemia have proven inconclusive. the serum albumin level attained during albumin administra-
tion exceeds 30 g/L.

Methods Conclusions
A meta-analysis was conducted of 90 cohort studies with Hypoalbuminemia is strongly associated with poor clinical
291,433 total patients evaluating hypoalbuminemia as an out- outcomes. Further well-designed trials are needed to charac-
come predictor by multivariate analysis and, separately, of terize the effects of albumin therapy in hypoalbuminemic pa-
nine prospective controlled trials with 535 total patients on tients. In the interim, there is no compelling basis to withhold
correcting hypoalbuminemia. albumin therapy if it is judged clinically appropriate.

The normal serum concentration of albumin in healthy and hospital stay in acutely ill patients with hypoalbumin-
adults is approximately 35 to 50 g/L. Diminished circulating emia is well recognized.3,4 Because of its importance as an
level of albumin— hypoalbuminemia—is common in seri- outcome predictor, serum albumin level has been added as
ously ill patients. For instance, the reported frequency of one of the component parameters in the Acute Physiology
hypoalbuminemia, defined as a serum albumin concentra- and Chronic Health Evaluation (APACHE) III score.5 The
tion of less than 34 g/L, was 21% at the time of admission association between hypoalbuminemia and poor outcomes
in adult hospitalized patients.1 After admission, worsening has long motivated clinicians in administering exogenous
of existing hypoalbuminemia and development of de novo albumin to hypoalbuminemic patients,4 and hypoalbumine-
hypoalbuminemia are both frequently encountered.2 mia is a licensed indication for human albumin in the United
The increased likelihood of poor outcomes such as mor- States and other countries. However, the appropriateness of
tality, morbidity, and prolonged intensive care unit (ICU) this practice has been challenged on the basis of insufficient
evidence to support its efficacy.4,6
At the heart of the controversy is causality—whether
Correspondence: Jean-Louis Vincent, MD, PhD, FCCM, Head, Depart- hypoalbuminemia directly contributes to poor outcomes, in
ment of Intensive Care, Hôpital Erasme, Université Libre de Bruxelles,
which case albumin replacement therapy might bestow ben-
Route de Lennik 808, B-1070 Brussels, Belgium.
E-mail: jlvincen@ulb.ac.be efit, or merely serves as a marker for other “upstream”
Accepted for publication November 8, 2002. pathologic processes such as malnutrition or inflammation,
319
320 Vincent and Others Ann. Surg. ● March 2003

in which case exogenous albumin might be ineffective in were included if they were designed to evaluate the effects
altering the clinical course of the patient. Two types of of correcting existing hypoalbuminemia in the acutely ill.
currently available evidence might help elucidate the appro- Morbidity data must have been available. Trials of hy-
priateness of albumin replacement therapy for hypoalbu- poalbuminemia prevention in patients with normal serum
minemia: results from multivariate analysis of cohort study albumin levels at study entry were not included. A random-
data collected for the purpose of evaluating the relationship ized design was not required. The control group must have
between albumin level and outcome specifically in acutely received either crystalloid or no albumin. Trials comparing
ill patients, and findings of controlled trials evaluating the albumin with synthetic colloids, blood products, or plasma
effects of correcting hypoalbuminemia on morbidity. Such protein fraction as the control regimen were not included.
evidence has not previously been systematically reviewed.
Multivariate analysis is crucial to an epidemiologic in-
Search Techniques
quiry into causality, since it can directly address whether
other risk factors satisfactorily explain the hypoalbuminemia Relevant studies were sought without language restriction
effect and can provide more accurate estimates for the strength by computer searches of the MEDLINE and EMBASE
of the hypoalbuminemia effect and the dose-response relation- bibliographic databases, the Cochrane Controlled Trials
ship between endogenous serum albumin level and outcome. Register, and the Cochrane Medical Editors Trial Amnesty
Lack of detectable confounding, strength of association, and of unpublished trials. Additional Internet-resident resources
dose-response relationship, as well as plausible biologic mech- such as conference reports, abstracts, reference compila-
anisms, are the principal epidemiologic criteria for evaluating tions, and full-text journal articles were located using the
causality.7 Multivariate analysis has been reported on volumi- Altavista, Northernlight, Hotbot, Excite, and Google search
nous data generated in cohort studies examining the effects of engines. We searched the Journal of the American Medical
hypoalbuminemia on a variety of clinical outcomes in the Association, the New England Journal of Medicine, The
acutely ill. The preponderance of this evidence has been pub- Lancet, and the British Medical Journal by hand for the
lished only recently. period from January 1990 to May 2002 and Index Medicus
Two recent meta-analyses failed to detect a significant for the years 1940 through 1965. We also consulted the
effect of albumin administration on survival in randomized authors of published controlled trial reports related to albu-
controlled trials of hypoalbuminemic patients.8,9 However, min and the medical directors of albumin suppliers and
in these trials the numbers of patients enrolled were small. examined the reference citations from completed reviews
Furthermore, the reported mortality rate in the control group and protocols in the Cochrane Database of Systematic Re-
was only 10% to 12%. Consequently, mortality was a rel- views, other meta-analyses, review articles, and reports of
atively insensitive endpoint in this patient population. Com- controlled and uncontrolled studies involving albumin.
plications, on the other hand, are substantially more fre-
quent than deaths, and thus morbidity may serve as a more
Data Collection
sensitive indicator for the effects of administered albumin.
In addition, morbidity is a major concern of patients.10 Two investigators independently selected studies and ex-
Through systematic review, we have endeavored to iden- tracted data. Disparities in selection and extraction deci-
tify all cohort studies with multivariate analysis of serum sions were resolved through discussion. In controlled trials,
albumin level as an outcome predictor and all controlled complications were scored on an intention-to-treat basis.
trials on correction of hypoalbuminemia. We here report the
results of a meta-analysis encompassing both types of
Statistical Analysis
investigations.
Individual study results were expressed as odds ratios
(OR) with 95% confidence intervals (CI). To allow for
MATERIALS AND METHODS between-study statistical heterogeneity, OR estimates were
quantitatively combined under a random effects model. OR
Inclusion Criteria
values more than 1 signify an increased probability of poor
For inclusion, cohort studies must have focused on serum outcome.
albumin as an outcome predictor in the acutely ill and Most multivariate analyses from cohort studies were per-
entailed multivariate analysis by methods such as logistic formed by either logistic regression or Cox regression.
regression or Cox proportional hazards regression. No re- Exponentiated coefficients from these two types of analysis
strictions were placed on the types of acute illnesses or provide an OR and hazard ratio, respectively, for the out-
outcomes addressed. Thus, we adopted broad inclusion cri- come of interest. These two effect size metrics, though
teria. Views differ as to the optimal scope for meta-analy- distinct, have been shown both theoretically and empirically
ses; however, broad meta-analyses have been advocated on to be similar in magnitude under a range of conditions.12,13
the basis of their statistical power and generalizability.11 We use the term OR to denote both these effect size mea-
Published and unpublished prospective controlled trials sures. Logistic regression and Cox regression models can
Vol. 237 ● No. 3 Hypoalbuminemia 321

accommodate both continuous variables such as serum al- surgery, 12 noncardiac surgery, and 37 renal dysfunction.
bumin level or body mass index and binary indicator vari- The total numbers of patients in the respective categories
ables such as the presence of chronic renal disease or were 27,730, 39,080, 65,828 and 158,795. The median
diabetes. The OR corresponding to the impact of hypoalbu- patient age across all included studies was 60 years (range
minemia on a particular endpoint is adjusted during the 10 – 89), and the median percentage of males in the study
estimation process to account for the effects of the other populations was 55% (range 26 –100%).
explanatory model variables. Forty-one included studies were prospective and 45 were
In some included cohort studies OR was reported on the retrospective. Four studies involved both prospective and
basis of broad serum albumin cutoffs (e.g., ⬍35 vs. ⱖ35 retrospective components. Five studies were multicenter
g/L). In such cases we used the provided OR value and investigations.
estimates of the median serum albumin values within the Multivariate analysis was conducted by logistic regres-
two cutoff ranges to calculate OR for each 10-g/L decre- sion in 40 studies, Cox regression in 36, and both in 1. OR
ment in serum albumin.14 The median values were wherever and CI could be extracted or derived from the reports of 67
possible estimated from within-study data on the distribu- studies. The remaining 23 studies either did not supply the
tion of serum albumin concentrations and otherwise from necessary data or employed a type of multivariate analysis
comparable studies in the same or similar clinical indica- that does not yield an OR estimate. Hence, data from these
tions. In a few cohort studies OR values for each of several 23 included studies could not be quantitatively combined.
serum albumin levels were supplied, and OR per 10-g/L The median number of covariates (i.e., variables in addi-
serum albumin decline was estimated by weighted least- tion to albumin evaluated for inclusion in the multivariate
squares regression.15 When P values were the only mea- models) per study was 11 (range 1–124). Mortality was an
sures of precision reported, the test-based CI was calculat- endpoint of 66 studies, morbidity of 27, hospital stay of 9,
ed.14 For cohort studies we employed the META computer ICU stay of 3, resource utilization (i.e., ventilatory support,
program16 to calculate pooled estimates of OR and CI and transfusion or hospitalization) of 9, treatment failure of 2,
evaluate heterogeneity. quality of life of 1, and overmedication of 1. A single
For prospective controlled studies, the primary outcome endpoint was subjected to multivariate analysis in 71 stud-
measure was the OR for occurrence of one or more com- ies, 2 endpoints in 14 studies, 3 in 3 studies, and 5 in 2
plications in individual patients. The METAN computer studies. The total number of multivariate analyses among all
program17 was used to calculate OR and CI for individual 90 studies was 118.
studies, assess heterogeneity, and derive pooled estimates of The strength of association between serum albumin and
OR and CI across studies. Because of pre-existing evidence outcome was reported for 26 final multivariate models. The
indicating a dose-response relationship between albumin median rank of albumin, from strongest to weakest predic-
level and outcome,18 the analysis plan also called a priori tor, versus all other covariates included in the final models
for an evaluation of outcome in relation to the mean peak was 2 (range 1–11). The median number of covariates in
serum albumin level attained during albumin therapy. This these final models, not counting albumin itself, was 5 (range
evaluation was performed by meta-analysis regression19 1–24). Thus, albumin was among the most powerful out-
using the METAREG program.20 come predictors.
Publication bias in controlled trials was assessed by the In eight analyses both univariate and multivariate OR
method of Egger et al21 using the METABIAS program.22 estimates were provided for the association between serum
The methodologic quality of controlled trials was appraised albumin and outcome. In three analyses the multivariate
on the basis of blinding, the presence of morbidity as a study exceeded the univariate estimate by an average of 38%,
endpoint, and between-group crossover by one or more while in the other five analyses the univariate estimate was
patients. In the case of randomized controlled trials, the larger by a mean of 24%. These observations confirm that
allocation concealment method was classified as adequate, adjustment for the effects of covariates can result in sub-
inadequate, or unclear.23 stantially altered effect size estimates.
Mortality
RESULTS OR and CI estimates for mortality associated with a
10-g/L fall in serum albumin were obtained for 53 studies
Cohort Studies
(Fig. 1). The pooled OR for these trials was 2.37 (CI
Ninety cohort studies fulfilling the inclusion criteria were 2.10 –2.68). Thus, the odds of death were increased by
identified (Table 1).1,2,18,24 –110 The total number of patients 137% with each 10-g/L decline in serum albumin, and the
in the 90 studies was 291,433 and the median number of effect was statistically significant. Similarly, based on pool-
patients per study was 281 (range 32–54,215). Forty-nine of ing within clinical indications, statistically significant in-
the studies, with 200,413 patients, representing 69% of the creases in mortality odds of 102%, 116%, 180%, and 148%
total patient population, were published since 1998. Thirty were observed for the hospitalization (OR 2.02; CI 1.52–
studies involved hospitalized patients in general, 11 cardiac 2.70), cardiac surgery (OR 2.16; CI 1.47–3.16), noncardiac
322 Vincent and Others Ann. Surg. ● March 2003

Table 1. CHARACTERISTICS OF COHORT STUDIES TRIALS DESIGNED TO EVALUATE

EFFECTS OF HYPOALBUMINEMIA ON OUTCOME

Study Patients Clinical Setting Design Analysis Covariates

Hospitalization
Harvey et al, 198125 282 Critically ill and/or malnourished patients RS DA 12
Agarwal et al, 198826 80 Consecutive elderly patients PS LR 9
Levkoff et al, 198827 1,756 Elderly hospitalized patients RS, PS RP 31
Trzepacz et al, 198828 108 Consecutive liver transplantation candidates PS DA 5
Bladé et al, 198929 180 Multiple myeloma RS CR 14
Sullivan et al, 199034 110 Consecutive admissions to Veterans Administration hospital PS DA 53
geriatric rehabilitation unit
Cherng et al, 199135 265 Multiple myeloma MC, RS CR 18
Levis et al, 199136 342 Chronic lymphocytic leukemia RS CR 18
Herrmann et al, 19921 15,511 Patients ⬎ 40 y of age RS LR 6
Burr et al, 199338 200 Patients admitted to rehabilitation center within 1 y of spinal PS ANOVA 12
cord lesion
Ferguson et al, 199339 81 Hospitalized elderly nursing home residents PS LR 2
McEllistrum et al, 199341 148 Consecutively discharged geriatric Veterans Administration RS MR 1
hospital patients
Sirott et al, 199344 284 Metastatic malignant melanoma RS WR 54
Toledo et al, 199345 185 Spontaneous bacterial peritonitis in cirrhotic patients treated RS LR 50
with cefotaxime
Espinosa et al, 199549 292 Advanced non-small cell lung cancer RS CR 8
Violi et al, 199551 165 Cirrhosis PS CR 8
McCluskey et al, 19962 348 Consecutive critically ill patients RS LR 1
Gariballa et al, 199865 201 Stroke PS CR 13
Incalzi et al, 199869 370 Patients ⬎ 70 y of age in acute-care university hospital PS LR 11
Marinella and Markert, 199871 144 Consecutive hospitalized patients 60 y of age or older PS ANOVA 1
Deschènes et al, 199977 140 Cirrhotic patients without initial infection PS LR 7
Axdorph et al, 200083 145 Patients ⬎ 15 y of age with Hodgkin’s disease RS CR 14
Dharmarajan et al, 200086 121 Clostridium difficile colitis RS LR 21
Dhiman et al, 200087 288 Consecutive patients with fulminant hepatic failure PS LR 7
Oki et al, 200091 1,594 Kawasaki disease PS LR 7
Leung et al, 200198 340 Patients referred to hospital diabetic foot clinic RS LR 5
Rozzini et al, 200199 840 Patients 肁 75 y consecutively admitted to acute care unit PS CR 5
Walter et al, 2001103 2,922 Hospitalized patients ⬎ 70 y with medical illnesses PS LR 15
Modawal et al, 2002106 145 Ventilator-dependent patients RS LR 10
Tincani et al, 2002109 143 Venous thromboembolism PR LR 5
Cardiac Surgery
Rich et al, 198930 92 Consecutive patients 肁 75 y undergoing cardiopulmonary RS MR 5
bypass
Magovern et al, 199656 2,802 Consecutive coronary artery bypass graft surgery patients RS, PS LR 124
Magovern et al, 199657 2,033 Isolated coronary artery bypass graft RS LR 13
Rady et al, 199761 1,461 Adult cardiovascular ICU patients PS LR 19
Rady et al, 199760 2,743 Adult cardiovascular ICU patients PS LR 3
Ryan et al, 199762 324 Adults in cardiovascular ICU 肁 14 d RS LR 18
Göl et al, 199866 9,352 Open heart surgery RS LR 19
Rady et al, 199875 1,157 Elderly cardiothoracic ICU patients PS LR 27
Engelman et al, 199978 5,168 Coronary artery bypass graft, valve operations or both PS LR 17
Rady and Ryan, 199981 11,330 Cardiothoracic ICU patients RS LR 22
Bashour et al, 200084 2,618 Coronary artery bypass graft and/or valve surgery PS CR, LR 9
Noncardiac Surgery
Buzby et al, 198024 261 General surgery, gastrointestinal surgery RS, PS DA 13
Altomare et al, 199031 70 Enterocutaneous fistulae RS LR 2
Lai et al, 199032 86 Consecutive patients undergoing emergency surgery for RS DA 22
severe acute cholangitis
Guijarro et al, 199654 675 Renal transplant recipients RS CR 6
Pacelli et al, 199658 604 Intra-abdominal infections RS LR 12

(continues)

RS, retrospective; PS, prospective; MC, multicenter; LR, logistic regression; CR, Cox regression; DA, discriminant analysis; MR, multiple regression; ANOVA, analysis of
variance; RP, recursive partitioning; WR, Weibull regression; ICU, intensive care unit; USRDS, United States Renal Data System; CAPD, continuous ambulatory peritoneal
dialysis; ESRD, end-stage renal disease.
Vol. 237 ● No. 3 Hypoalbuminemia 323
Table 1 (continued).

Study Patients Clinical Setting Design Analysis Covariates

Noncardiac Surgery
Friedenberg et al, 199759 64 Percutaneous endoscopic gastrostomy PS LR 9
Hedström et al, 199867 437 Femoral neck fractures RS LR 7
Becker et al, 199976 232 Kidney-pancreas transplant RS CR 1
Gibbs et al, 199918 54,215 All noncardiac surgery MC, PS LR 61
Gerhards et al, 200088 112 Consecutive patients undergoing resection for hilar RS LR 8
cholangiocarcinoma
Nair et al, 200090 112 Consecutive patients undergoing resection for hilar RS LR 8
cholangiocarcinoma
Nair et al, 200090 56 Percutaneous endoscopic gastrostomy in elderly patients PS LR 10
with dementia
Scott et al, 2001100 9,016 Surgical patients receiving prophylactic antibiotic RS LR 29
Renal Dysfunction
Lowrie and Lew, 199033 19,746 Hemodialysis RS LR 17
USRDS, 199237 3,399 Hemodialysis RS CR 37
Goldwasser et al, 199340 184 New and long-standing hemodialysis patients RS CR 6
Owen et al, 199342 13,473 Adult hemodialysis patients RS LR 8
Rocco et al, 199343 45 CAPD RS LR 3
Collins et al, 199446 1,773 First-time hemodialysis patients RS CR 6
Marcus et al, 199447 89 Peritoneal dialysis RS, PS MR 3
Avram et al, 199548 250 Hemodialysis PS CR 13
Lowrie et al, 199550 17,926 Peritoneal dialysis and hemodialysis RS CR 36
Avram et al, 199652 169 CAPD PS CR 9
Foley et al, 199653 432 ESRD MC, PS CR 5
Iseki et al, 199655 1,982 Chronic dialysis RS CR 18
Bologa et al, 199863 90 Ambulatory adult hemodialysis patients PS CR 15
Chertow et al, 199864 256 Acute renal failure MC, PS CR 22
Ifudu et al, 199868 522 Hemodialysis patients 肁 20 y old PS CR 10
Leavey et al, 199870 3,607 Hemodialysis RS CR 14
Noh et al, 199872 106 CAPD PS CR 11
Owen et al, 199873 18,144 Hemodialysis RS LR 5
Owen and Lowrie, 199874 1,054 Hemodialysis PS LR 21
Obialo et al, 199979 100 Acute renal failure RS LR 16
Ohashi et al, 199980 91 CAPD RS CR 3
Zimmermann et al, 199982 280 Stable hemodialysis patients PS CR 17
Chung et al, 200085 213 CAPD PS CR 3
Moon et al, 200089 32 Hemodialysis RS LR 3
Sharma et al, 200092 41 CAPD PS CR 6
Tanna et al, 200093 432 Peritoneal dialysis or hemodialysis PS CR 14
Yeun et al, 200094 91 Hemodialysis PS CR 9
Cueto-Manzano et al, 200195 627 CAPD RS CR 6
Gulati et al, 200196 180 Children on continuous peritoneal dialysis RS LR 2
Kalantar-Zadeh et al, 200197 83 Maintenance hemodialysis patients PS CR 17
Tokars et al, 2001101 796 Hemodialysis MC, PS CR 12
Tveit et al, 2001102 33,479 Renal transplant recipients RS LR 46
Bakewell et al, 2002104 88 Peritoneal dialysis PS MR 17
Klassen et al, 2002105 37,069 Maintenance hemodialysis RS CR 15
Sezer et al, 2002107 68 Hemodialysis PS CR 12
Stefoni et al, 2002108 155 ESRD patients on hemodialysis PS CR 7
Wong et al, 2002110 1,723 Pediatric patients with ESRD RS CR 8

surgery (OR 2.80; CI 2.18 –3.58), and renal dysfunction analysis (Table 2). There were no substantial between-
(OR 2.48; CI 2.11–2.91) categories, respectively. stratum differences with respect to any of these variables,
For the four categories of indications, the point estimates however, and hypoalbuminemia was significantly predictive
of OR were similar in magnitude and the pooled CI over- of mortality in all strata. For instance, hypoalbuminemia
lapped extensively. There was nevertheless evidence of was significantly associated with mortality both among ret-
significant overall statistical heterogeneity (P ⬍ .005), and rospective and prospective cohort studies.
additional possible contributors to heterogeneity (study de- Thirteen included studies evaluated mortality but did not
sign, patient age, multivariate analytic method, number of provide OR and CI estimates.25,32,34,36,43– 45,51,59,72,86,88,90
covariates, and study size) were investigated by sensitivity In eight of these studies, serum albumin was a significant
324 Vincent and Others Ann. Surg. ● March 2003

Figure 1. Effect of serum albumin on mortality. ICU, intensive care unit; CAPD, chronic ambulatory
peritoneal dialysis; CRP, C-reactive protein.

independent predictor of mortality. Thus, the findings of To evaluate potential confounding by malnutrition, 15
these studies were qualitatively similar to those of the included cohort studies57,59,61,63,65,69,70,75,78,82,90,97,102,107,110
studies included in the meta-analysis shown in Figure 1. assessed body mass index as a covariate. For the 10 of these
Vol. 237 ● No. 3 Hypoalbuminemia 325

Table 2. SENSITIVITY ANALYSES OF COHORT STUDIES

Mortality Morbidity

Study Attributes OR CI Studies OR CI Studies

Retrospective 2.65 2.05–3.44 27 2.68 1.70–4.23 7

Prospective 2.05 1.82–2.32 26 1.72 1.42–2.09 11
Mean age ⬍ 60 y 2.63 2.00–3.46 24 1.69 1.33–2.14 8
Mean age 肁 60 y 2.24 1.92–2.62 29 1.94 1.57–2.41 10
Logistic regression 2.67 2.05–3.48 20 1.89 1.56–2.29 14
Cox regression 2.19 1.91–2.52 33 1.78 1.44–2.19 4
⬍ 10 covariates evaluated 2.53 2.03–3.15 25 2.40 1.70–3.36 9
肁 10 covariates evaluated 2.27 1.91–2.69 28 1.66 1.34–2.05 9
⬍ 1000 patients 2.34 1.92–2.86 34 2.11 1.55–2.88 9
肁 1000 patients 2.52 2.05–3.10 19 1.80 1.45–2.23 9
BMI evaluated as covariate 1.89 1.51–2.36 10 1.42 1.07–1.90 4
BMI not evaluated as covariate 2.43 2.12–2.78 43 1.98 1.69–2.33 14
CRP evaluated as covariate 2.77 1.66–4.62 6 — — —
CRP not evaluated as covariate 2.35 2.07–2.66 47 1.89 1.59–2.24 18

BMI, body mass index; CRP, C-reactive protein.

studies with a mortality endpoint and available OR and CI data statistically significant 89% increase in odds of complica-
(see Table 2), the pooled OR for mortality per 10-g/L serum tions corresponding to a 10-g/L reduction in serum albumin.
albumin decrement was 1.89 (CI 1.51–2.36). The associa- Significant increases in morbidity odds of 178%, 52%, 73%,
tion between hypoalbuminemia and mortality was also and 102% were documented respectively among the subsets
shown to be independent of other nutritional indices such of studies involving hospitalization (OR 2.78; CI 1.30 –
body weight, dry weight, body fat percentage, weight loss, 5.98), cardiac surgery (OR 1.52; CI 1.12–2.04), noncardiac
cachexia, midarm circumference, and biceps and triceps surgery (OR 1.73; CI 1.67–1.79), and renal dysfunction (OR
skinfold thicknesses.18,25,26,34,49,55,60,63,65,69,88,93,97,110 2.02; CI 1.48 –2.74).
In seven studies the effect of hypoalbuminemia on mor- There was significant statistical heterogeneity with re-
tality was assessed with the inflammatory marker C-reactive spect to the morbidity endpoint (P ⬍ .005). In sensitivity
protein (CRP) as a covariate.72,74,82,83,94,97,107 For six of analyses between-stratum pooled estimates of OR did not
these studies OR estimates were available, and the pooled
differ notably, and CI overlapped extensively (see Table 2).
OR was 2.77 (CI 1.66 – 4.62), indicating that hypoalbumin-
Significantly increased morbidity odds per 10-g/L serum
emia remained a significant mortality predictor with the
albumin decrement were apparent for all subsets of trials
effects of CRP taken into account (see Table 2). The sig-
with data represented in Table 2.
nificant association between hypoalbuminemia and mortal-
ity also persisted when other markers of inflammation were Morbidity was addressed in eight studies that could not
evaluated as covariates (white blood cell count, lymphocyte be included in the meta-analysis of Figure 2 due to unavail-
count, neutrophil count, interleukin-6, interleukin-10, tumor ability of OR and CI data.24,25,27,28,30,34,56,88 In seven of
necrosis factor-␣, ␤2-microglobulin, serum amyloid A, trans- these studies serum albumin was found to be a significant
ferrin, and fibrinogen).25,26,29,32,34,44,49,50,58,63,65,69,74,82,83,95,97 independent predictor of morbidity.
In four studies with an OR estimate provided for mor-
Morbidity bidity and inclusion of body mass index as a covariate (see
Table 2), hypoalbuminemia was significantly associated
Hypoalbuminemia was also an independent predictor of
with morbidity (OR 1.42; CI 1.07–1.90). Hypoalbuminemia
morbidity across all studies and within each of the four
categories of clinical indications (Fig. 2). Morbidity was also remained a significant independent morbidity predictor
reported as overall morbidity or as one or more types of in studies taking into account markers of nutritional status
individual complications, most frequently involving cardio- other than body mass index such as body weight, body surface
vascular morbidity, infection, or organ dysfunction. Thus, area, degree of weight loss, rate of weight loss, triceps skinfold,
morbidity did not constitute a single homogeneous outcome midarm muscle circumference, and cachexia.18,24,25,34,60,61,102
measure across studies, and the seriousness of particular While none of the included studies assessed morbidity with
complications evaluated was not necessarily similar. CRP as a covariate, in three studies hypoalbuminemia was
The pooled OR for morbidity among all 18 studies as- associated with significantly increased morbidity when
sessing this endpoint was 1.89 (CI 1.59 –2.24), indicating a other markers of inflammation were taken into account
326 Vincent and Others Ann. Surg. ● March 2003

Figure 2. Effect of serum albumin on morbidity. ESRD, end-stage renal disease.

(white blood cell count, total lymphocyte count, and serum fusion,57 and hospitalization (i.e., admission or readmission
transferrin).24,25,27 to hospital).43,76,96,97 The economic impact of each type of
resource utilization was not quantified in the study reports
Length of Stay
and cannot be presumed to be similar in magnitude.
Hypoalbuminemia was a significant independent predic- Among the eight studies supplying OR data, hypoalbu-
tor of prolongation in both ICU and hospital stay. Length of minemia significantly increased resource utilization by 66%
ICU stay was a subject of three included studies.60,78,84 The per 10-g/L serum albumin decline (OR 1.66; CI 1.17–2.36).
pooled OR for ICU stay was 1.28 (CI 1.16 –1.40), indicating In the ninth study, no OR estimate was available, but
a significant 28% increase in odds for prolonged ICU stay hypoalbuminemia was significantly associated with in-
per 10-g/L decrement in serum albumin. creased resource utilization.
OR and CI estimates for prolonged hospital stay were
Other Endpoints
available from four included studies.1,39,60,78 The corre-
sponding pooled OR was 1.71 (CI 1.33–2.21), revealing a In two trials on the relation of hypoalbuminemia to treat-
significant hypoalbuminemia-related increase of 71% in ment failure,95,96 the pooled OR was 2.07 (CI 0.52– 8.30).
odds of prolonged hospital stay. In five additional included In one study hypoalbuminemia was predictive of signifi-
studies,30,38,41,47,71 length of hospital stay was evaluated, cantly poorer quality of life.104 The odds of overcoagulation
but OR and CI estimates were unavailable. Serum albumin were increased in hypoalbuminemic patients with venous
was a significant independent predictor of prolonged hos- thromboembolism (OR 3.60; CI 1.32–9.77).109
pital stay in all five studies.
Resource Utilization Controlled Trials
In nine studies resource utilization was evaluated with Nine prospective controlled trials with 535 total patients
respect to ventilatory support,60,78,81,106 postoperative trans- conformed to all inclusion criteria.111–119 The median num-
 Table 3. CHARACTERISTICS OF CONTROLLED TRIALS DESIGNED TO EVALUATE CORRECTION OF HYPOALBUMINEMIA
Vol. 237 ● No. 3

Trial Clinical Setting Test Regimen Endpoints Complication Types Reported

McMurray et al, 1948111 Low birth weight premature infants 1–2 injections per week of 3 cc Serum albumin; weight gain, hospital Gastroenteritis; jaundice
25 g/dL albumin per pound stay, morbidity, mortality
body weight vs no albumin
Smith et al, 1950112 Low birth weight premature infants 25% salt-poor albumin 0.5–0.75 Serum albumin; morbidity, mortality Pneumonia, diarrhea, edema, distension,
g per pound body weight vomiting
twice weekly
Ford et al, 1987113 Hypoalbuminemia (serum albumin ⬍ Salt poor albumin according to Tolerance to enteral feeding; serum Enteral feeding intolerance; sepsis; persistent
30 g/L) in pediatric patients formula ((3.5 g/dL ⫺ serum albumin diarrhea; aspiration or tube-related
selected for enteral tube feeding via albumin (g/dL)) ⫻ (weight complications
stomach or small bowel (kg) ⫻ 3) over 2–3 d in divided
doses vs no albumin
Brown et al, 1988114 Hypoalbuminemia (serum albumin ⬍ 12.5 g/L albumin, then 25–37.5 Hospital morbidity Septicemia; intra-abdominal abscess; fistula;
30 g/L) in adult patients requiring g/d albumin vs no albumin urinary tract infection; pneumonia; wound
TPN because of general surgery, infection; wound dehiscence; congestive heart
multiple trauma or medical failure; phlebitis; respiratory failure; pulmonary
conditions embolus; cerebrovascular accident; septic
shock
Foley et al, 1990115 Hypoalbuminemia (serum albumin ⬍ 25–50 g/d 25% albumin vs no Mortality; major complication rate Major arrhythmia; myocardial infarction; deep
25 g/L) in critically ill adult patients albumin vein thrombosis; shock; stroke; pneumonia;
referred for TPN pneumothorax; respiratory failure; cholecystitis;
bleeding; small-bowel obstruction; acute renal
failure; bacterial sepsis; viral sepsis; fungal
sepsis; intra-abdominal sepsis; mediastinitis;
Clostridium difficile enterocolitis; line sepsis;
diffuse peritonitis; urinary tract infection;
enterocutaneous fistula; decubitus ulcer;
wound infection
Kanarek et al, 1992116 Sick newborn infants with respiratory Albumin to maintain 30 g/L Serum albumin; severity of respiratory Bronchopulmonary dysplasia; necrotizing
distress, hypotension, serum albumin vs no added distress; mean arterial blood enterocolitis
hypoalbuminemia (serum albumin albumin pressure; weight gain; incidence of
⬍ 30 g/L) and a requirement for complications
TPN
Wojtysiak et al, 1992117 Hypoalbuminemia (serum albumin ⬍ Albumin 25 g/L as a continuous Serum albumin; COP; free water Bacteremia; pneumonia; urinary tract infection;
30 g/L) in adult patients requiring infusion for a 5 d study period clearance; electrolyte-free water intraabdominal sepsis; wound infection; vaginal
TPN resulting from multiple trauma, vs no albumin resorption; sodium excretion infection; cholangitis
general surgery, carcinoma or
medical conditions

(continues)
Hypoalbuminemia

TPN, total parenteral nutrition; COP, colloid osmotic pressure.

327
328 Vincent and Others Ann. Surg. ● March 2003

ber of patients per trial was 38 (range 24 –219). Seven of the

Pneumonia; sepsis; wound infection; urinary tract

gastrointestinal bleeding; renal failure; cardiac

trials were randomized.111,114 –119 None was unpublished.

respiratory failure; congestive heart failure;

infection; line infection; abdominal sepsis;
Characteristics of the included controlled trials are summa-

dehiscence; decubitus ulcer; liver failure;

arrest; cerebrovascular accident; enteral

Complication Types Reported
rized in Table 3.
Four trials involved pediatric patients.111–113,116 For the

myocardial infarction; arrhythmia;

adult studies the median patient age was 59 years (range
47–71). The median duration of follow-up for both adult
and pediatric trials was 26 days (range 5–150).

Bacteremia; pneumonia
Three trials involved some form of blinding.114,116,119

feeding intolerance
The method of allocation concealment was adequate in four
of six randomized trials,114,116,118,119 unclear in two,115,117
and inadequate in one.111 Morbidity was an endpoint of all
included trials except one.117 In four trials individual control-
group patients crossed over to albumin therapy.114,115,117,118
The median number of patients crossing over was 2.5 (range
Morbidity; mortality; serum albumin
2– 6).
Pooled Morbidity
As shown in Figure 3, the pooled OR for occurrence of
Endpoints

one or more complications in individual patients was 0.74

(CI 0.36 –1.49). Thus, morbidity was lower among albumin
Morbidity; mortality

recipients, but the effect was not statistically significant.

The pooled odds ratio was similar after exclusion of the two
nonrandomized trials (OR 0.81; CI 0.41–1.60).112,113 There
was no evidence of publication bias (P ⫽ .367). Among all
Table 3 (continued).

control group patients, 48% (125/262) experienced one or

more complications.
50 mL of 25% albumin every 8 h
to maintain serum albumin ⬎

Dose Dependency
25 g/d albumin vs 100 mL/d

Significant between-trial heterogeneity was apparent with

Test Regimen

30 g/L vs no albumin

respect to the OR for complications (P ⫽ .006), and possi-

ble sources of heterogeneity such as year of publication,
normal saline

duration of follow-up, and albumin dose as reflected by

attained serum albumin level during therapy were investi-
gated by meta-analysis regression. The OR for complica-
tions was not significantly related to either the year of
publication (P ⫽ .622) or duration of follow-up (P ⫽ .220).
bleeding, cancer, perforated viscus,

patients received nutritional support

in the form of a diet, TPN or enteral

Pooled baseline serum albumin concentration did not

albumin) in adult patients receiving
Hypoalbuminemia (serum albumin ⬍

patients to surgical intensive care

infection or bowel obstruction; all

unit due to vascular insufficiency,

Hypoalbuminemia (⬍ 25 g/L serum

differ significantly (P ⫽ .79) between the albumin (24.4 ⫾

30 g/L) upon admission of adult

1.4 g/L) and control group (24.5 ⫾ 1.4 g/L). Attained serum
pancreatitis, intra-abdominal
hip fracture, gastrointestinal
Clinical Setting

albumin level during therapy was reported in eight of nine

trials. As shown in Figure 4,120 the OR for complications
declined progressively as attained serum albumin level in-
creased (P ⫽ .002). In the five studies with an attained
albumin level of more than 30 g/L, complications were less
feedings

frequent in albumin recipients than control-group patients

TPN

and vice versa in the three studies with an attained albumin

level of less than 30 g/L. The between-study variance de-
termined by method-of-moments estimator was 0.0815 in
the meta-analysis regression as contrasted with 0.656 with
Golub et al, 1994118

Rubin et al, 1997119

no account taken of attained serum albumin level. Thus, the

dose effect could explain 88% of total between-study vari-
Trial

ance. The significant observed dose-response relationship

also persisted after exclusion of two early trials (P ⫽
.026).111,112 Meta-analysis regression failed to reveal any
significant relationship between morbidity and baseline se-
Vol. 237 ● No. 3 Hypoalbuminemia 329

Figure 3. Morbidity in controlled trials evaluating the correction of hypoalbuminemia.

rum albumin of either the albumin group (P ⫽ .22) or the come in the acutely ill. This association was striking both
control group (P ⫽ .33) or the weighted average of baseline for its consistency and pervasiveness. Unfavorable sequelae
serum albumin in both groups (P ⫽ .27). associated with lower serum albumin were evident in hospi-
talized patients generally and in populations undergoing
cardiac and noncardiac surgery or suffering from renal dys-
DISCUSSION function. The hypoalbuminemia effect manifested itself across
Our meta-analysis is the first to assemble comprehensive the full spectrum of clinical outcomes: mortality, morbidity,
evidence— based largely on recently reported data analyzed length of both ICU and hospital stay, and increased resource
by multivariate methods—that addresses the long-standing utilization. However, the evidence on length of stay, resource
debate over the clinical importance of hypoalbuminemia in utilization, and certain other endpoints was substantially more
the acutely ill. It also provides a novel quantitative model of limited than that on mortality and morbidity.
dose dependency that may explain the disparate results ob- Because of the strength of the association and low cost of
tained thus far in trials on the correction of hypoalbuminemia. serum albumin assays, monitoring albumin levels has been
We found hypoalbuminemia to be a powerful, reproduc- advocated as a prognostic tool to identify higher-risk pa-
ible, dose-dependent, independent risk factor for poor out- tients.1,18,78 As vividly demonstrated in the largest included

Figure 4. Meta-analysis regression assessing

the effect of attained serum albumin level on
morbidity in controlled trials to evaluate the cor-
rection of hypoalbuminemia. Data points are
scaled in proportion to precision. Incidence of
complications was reported on a per-patient
basis in six trials.111–115,118 The average num-
ber of complications per patient in these trials
was 1.1. For the other three studies complica-
tions per patient were estimated from total com-
plication data using 1.1 as the conversion fac-
tor. For analytic purposes the attained serum
albumin reported in two early trials111,112 was
adjusted downward by 13 g/L, corresponding
to the difference between the mean serum al-
bumin for healthy term infants using contempo-
rary assay methods (35 g/L)120 and the corre-
sponding mean using older methodology (48
g/L), as documented in one of the two early
trials.111
330 Vincent and Others Ann. Surg. ● March 2003

cohort study, involving 54,215 noncardiac surgery pa- come. A previous systematic review addressed the associ-
tients,18 both mortality and morbidity continuously in- ation between serum albumin and mortality in 10 cohort
creased as serum albumin progressively decreased over the studies with multivariate analysis involving 55,965 healthy
entire range of albumin levels between 22 and 46 g/L, and subjects and acutely ill patients.7 In that review, circulating
there was no evidence of any threshold above which albu- albumin concentration was inversely related to mortality
min no longer substantially affected these outcomes. In light risk in a progressive fashion over its entire range. The
of these findings, clinical risk stratification strategies em- association persisted after adjustment for other known risk
phasizing only severe hypoalbuminemia (e.g., serum albu- factors and pre-existing illness and exclusion of early mor-
min ⬍ 25 g/L) may insufficiently recognize the increased tality, and plausible biologic mechanisms underlying the
risk of patients with higher albumin levels (e.g., 25–35 g/L). effect were noted. The review prompted the conclusion that
Several limitations of our meta-analysis should be recog- albumin may exert a direct protective effect. Our review,
nized. Reliance on multivariate analysis of cohort study focusing exclusively on the acutely ill and addressing a
results was a principal feature of the meta-analysis, and the wider range of outcomes, also supports this conclusion.
included studies addressed a broad array of variables. Nev- Several mechanisms might help explain the apparent
ertheless, unidentified confounding variables may exist, and protective effects of serum albumin. Clearly, serum albumin
the possibility cannot be dismissed that the hypoalbumine- plays diverse, complex, and important roles in maintaining
mia effect is merely an epiphenomenon indicative of other physiologic homeostasis. At reduced albumin levels these
pathologic processes, as has often been argued. Further- homeostatic functions may be impaired, resulting in the
more, among the included cohort studies both the numbers development and/or progression of pathologic processes
and types of covariates differed widely. Additionally, quan- underlying poor outcome. The full spectrum of albumin
titative and qualitative differences were apparent among the biologic actions has as yet not been fully delineated and
complications assessed in both the included cohort studies forms the center of an active field of recent research. Nev-
and controlled trials. Because of these differences in the ertheless, a number of its actions are well established, such
number, type and seriousness of complications quantita- as the ability to maintain normal colloid osmotic pressure
tively combined, the morbidity endpoint evaluated in our (COP). Reduction in COP promotes edema formation.
meta-analysis may with some justification be regarded as While many patients may tolerate edema adequately, others
heterogeneous from a clinical point of view. may be adversely affected, particularly by pulmonary, myo-
Our data indicate that two important potential confound- cardial, or intestinal edema.
ing variables—malnutrition and inflammation— cannot The ability of serum albumin to sustain COP may only be
fully explain the hypoalbuminemia effect. The malnour- one of many possible protective effects of albumin. In a
ished state has long been recognized as a potential precip- study of 145 patients with prolonged critical illness, hy-
itating factor in the development of hypoalbuminemia. We poalbuminemia was associated with increased mortality,
found that the significant association between hypoalbumin- even though in these patients COP had been maintained by
emia and poor outcome persisted after adjustment for body administration of artificial colloid.121 Therefore, the hy-
mass index and other measures of nutritional status. poalbuminemia effect might be at least partly reliant on the
Growing attention has centered on the role of inflamma- additional properties of albumin such as its antioxidant and
tory processes in inducing hypoalbuminemia. By increasing free radical-scavenging activity, capacity to prevent apopto-
vascular permeability, inflammatory mediators may pro- sis, and affinity for binding lipids, drugs, toxic substances,
mote escape of albumin into the extravascular space. Such and other ligands.122–124 The antioxidant effects of albumin,
mediators may also reprioritize hepatic protein synthesis in for example, may translate into biologic protection of po-
favor of acute phase reactants at the expense of albumin tential clinical relevance. Albumin decreased both hydrogen
production. CRP, an acute phase protein produced by the peroxide formation and reperfusion injury in isolated rat
liver, is one marker of inflammation that has been proposed hearts.125 It also prevented ischemic and hypoxic damage in
to account for the association between hypoalbuminemia isolated perfused rat livers.126 Albumin inhibited peroxida-
and poor outcome.94 Nevertheless, we found the effects of tion of erythrocyte membrane lipids in chronic hemodialysis
hypoalbuminemia on outcome to be independent of CRP, as patients.127 Bilirubin bound to albumin substantially pro-
well as other markers of inflammation. Such observations longed the survival of human ventricular heart muscle cells
make plain that inflammation, at least as manifested by exposed to oxyradicals generated in situ.128
altered levels of currently identified inflammatory markers, Because of its high frequency in a broad range of patho-
may contribute to reduced serum albumin levels but never- logic conditions, hypoalbuminemia might plausibly be in-
theless cannot fully account for the association between terpreted as a normal compensatory mechanism not requir-
hypoalbuminemia and poor outcome. It is possible, how- ing intervention. For instance, albumin redistribution into
ever, that other as yet undetectable derangements in patient the interstitial space might provide protection from oxida-
inflammatory response might more completely explain the tive stress affecting extravascular tissues during disease
hypoalbuminemia effect. states. Nonetheless, hypoalbuminemia-related reductions in
Hypoalbuminemia may be causally related to poor out- COP, intravascular antioxidative reserve, binding activity,
Vol. 237 ● No. 3 Hypoalbuminemia 331

and other protective effects of albumin in the plasma com- lirious patients. Savings resulting from a reduced morbidity
partment are difficult to view as a beneficial adaptation. rate might exceed the acquisition cost of albumin.
If indeed serum albumin exerts a net protective effect, Further well-designed, adequately powered controlled
then exogenous albumin therapy might benefit hypoalbu- clinical trials are needed to resolve the question whether
minemic patients. Furthermore, even if albumin level did correcting hypoalbuminemia is beneficial. Pending the re-
primarily reflect other upstream pathologic processes, albu- sults of such future studies, there does appear to be a
min therapy might nevertheless be effective as a downstream coherent rationale for albumin replacement therapy and no
intervention to interrupt the chain of events culminating in poor compelling basis to withhold albumin if it is deemed clin-
outcome. Administration of purified albumin is clearly effec- ically appropriate in hypoalbuminemic patients.
tive in raising serum albumin levels.114 –116,118,129 Even in
sepsis, a condition marked by increased vascular permeability,
administered albumin can produce a prompt and sustained rise References
in serum albumin concentration.130
Available evidence from controlled trials is, however, 1. Herrmann FR, Safran C, Levkoff SE, et al. Serum albumin level on
inconclusive as to the potential benefits to be derived from admission as a predictor of death, length of stay, and readmission.
Arch Intern Med. 1992;152:125–130.
correcting hypoalbuminemia. In our meta-analysis, no sig- 2. McCluskey A, Thomas AN, Bowles BJ, et al. The prognostic value
nificant overall reduction in morbidity was demonstrable. of serial measurements of serum albumin concentration in patients
Nonetheless, the existing controlled trial evidence is lim- admitted to an intensive care unit. Anaesthesia. 1996;51:724 –727.
ited. Furthermore, based on our meta-analysis regression, 3. Reinhardt GF, Myscofski JW, Wilkens DB, et al. Incidence and
attained serum albumin level appears to be a major deter- mortality of hypoalbuminemic patients in hospitalized veterans.
minant of morbidity. Thus, the attained albumin level ac- JPEN J Parenter Enteral Nutr. 1980;4:357–359.
4. Margarson MP, Soni N. Serum albumin: touchstone or totem? An-
counted for most of the between-study variance, indicating
aesthesia. 1998;53:789 – 803.
that the differences in morbidity results reported among the 5. Knaus WA, Wagner DP, Draper EA, et al. The APACHE III prog-
included controlled trials were predominantly due to dose nostic system. Risk prediction of hospital mortality for critically ill
effects. This analysis suggests the hypothesis that adminis- hospitalized adults. Chest. 1991;100:1619 –1636.
tering sufficient exogenous albumin to achieve a serum 6. Vermeulen LC, Jr, Ratko TA, Erstad BL, et al. A paradigm for
albumin level of more than 30 g/L might lessen morbidity in consensus. The University Hospital Consortium guidelines for the
use of albumin, nonprotein colloid, and crystalloid solutions. Arch
hypoalbuminemic patients. The significant within-study
Intern Med. 1995;155:373–379.
findings for the trials in our meta-analysis regression are 7. Goldwasser P, Feldman J. Association of serum albumin and mor-
consistent with this hypothesis. tality risk. J Clin Epidemiol. 1997;50:693–703.
In the three studies with an attained serum albumin level 8. Wilkes MM, Navickis RJ. Patient survival after human albumin
of less than 30 g/L, no within-study evidence was obtained administration: a meta-analysis of randomized, controlled trials. Ann
of clinical benefit due to albumin therapy.112,115,118 Con- Intern Med. 2001;135:149 –164.
9. Alderson P, Bunn F, Lefebvre C, et al. Human albumin solution for
versely, such benefit was demonstrable in the four studies
resuscitation and volume expansion in critically ill patients. Co-
with a morbidity endpoint and an attained albumin level of chrane Database Syst Rev. 2002;1:CD001208.
more than 30 g/L. Specifically, in these studies with higher 10. Fried TR, Bradley EH, Towle VR, et al. Understanding the treatment
attained albumin levels, the albumin group experienced preferences of seriously ill patients. N Engl J Med. 2002;346:1061–
an acceleration of time to regain birthweight and fewer ill- 1066.
nesses;111 improved nutritional status and tolerance to enteral 11. Gøtzsche PC. Why we need a broad perspective on meta-analysis. It
may be crucially important for patients. Br Med J. 2000;321:585–
feeding;113 fewer total hospital complications and lower fre-
586.
quencies of septicemia and pneumonia;114 and greater relief of 12. Abbott RD. Logistic regression in survival analysis. Am J Epidemiol.
hypotension and shorter time to regain birthweight.116 1985;121:465– 471.
A morbidity benefit due to albumin therapy, if demon- 13. Ingram DD, Kleinman JC. Empirical comparisons of proportional
strable, might improve the overall costs of care. Significant hazards and logistic regression models. Stat Med. 1989;8:525–538.
edema may limit mobilization and delay convalescence. In 14. Greenland S. Quantitative methods in the review of epidemiologic
one included cohort study of 9,352 cardiac surgery pa- literature. Epidemiol Rev. 1987;9:1–30.
15. Greenland S, Longnecker MP. Methods for trend estimation from
tients,66 hypoalbuminemia was predictive of postoperative summarized dose-response data, with applications to meta-analysis.
infection, and ICU stay was increased over fivefold in Am J Epidemiol. 1992;135:1301–1309.
patients developing infections (13.4 ⫾ 20.1 days vs. 2.5 ⫾ 16. Sharp S, Sterne J. Meta-analysis. Stata Tech Bull. 1997;STB-38:100 –
3.8 days, P ⬍ .001). In another included cohort study of 106.
elderly hospitalized patients, hypoalbuminemia was associ- 17. Bradburn MJ, Deeks JJ, Altman DG. metan—an alternative meta-
ated with increased risk for a discharge diagnosis of delir- analysis command. Stata Tech Bull. 1998;STB-44:86 –100.
18. Gibbs J, Cull W, Henderson W, et al. Preoperative serum albumin
ium.27 Patients with this diagnosis incurred average total
level as a predictor of operative mortality and morbidity: results from
hospital charges of $17,377 versus $11,946 for those with- the National VA Surgical Risk Study. Arch Surg. 1999;134:36 – 42.
out the diagnosis. Mean hospital stay was 23.7 days for 19. Jüni P, Witschi A, Bloch R, et al. The hazards of scoring the quality
patients with delirium compared with 13.6 days for nonde- of clinical trials for meta-analysis. JAMA 1999; 282:1054 –1060.
332 Vincent and Others Ann. Surg. ● March 2003

20. Sharp S. Meta-analysis regression. Stata Tech Bull. 1998;STB-42: 44. Sirott MN, Bajorin DF, Wong GY, et al. Prognostic factors in patients
148 –155. with metastatic malignant melanoma. A multivariate analysis. Can-
21. Egger M, Davey Smith G, Schneider M, et al. Bias in meta-analysis cer. 1993;72:3091–3098.
detected by a simple, graphical test. Br Med J. 1997;315:629 – 634. 45. Toledo C, Salmerón JM, Rimola A, et al. Spontaneous bacterial
22. Steichen TJ. Tests for publication bias in meta-analysis. Stata Tech peritonitis in cirrhosis: predictive factors of infection resolution and
Bull. 1998;STB-41:125–133. survival in patients treated with cefotaxime. Hepatology. 1993;17:
23. Schulz KF, Chalmers I, Hayes RJ, et al. Empirical evidence of bias. 251–257.
Dimensions of methodological quality associated with estimates of 46. Collins AJ, Ma JZ, Umen A, et al. Urea index and other predictors of
treatment effects in controlled trials. JAMA. 1995;273:408 – 412. hemodialysis patient survival. Am J Kidney Dis. 1994;23:272–282.
24. Buzby GP, Mullen JL, Matthews DC, et al. Prognostic nutritional 47. Marcus RG, Chaing E, Dimaano F, et al. Serum albumin: associ-
index in gastrointestinal surgery. Am J Surg. 1980;139:160 –167. ations and significance in peritoneal dialysis. Adv Perit Dial.
25. Harvey KB, Moldawer LL, Bistrian BR, et al. Biological measures 1994;10:94 –98.
for the formulation of a hospital prognostic index. Am J Clin Nutr. 48. Avram MM, Mittman N, Bonomini L, et al. Markers for survival in
1981;34:2013–2022. dialysis: a seven-year prospective study. Am J Kidney Dis. 1995;
26. Agarwal N, Acevedo F, Leighton LS, et al. Predictive ability of 26:209 –219.
various nutritional variables for mortality in elderly people. Am J Clin 49. Espinosa E, Feliu J, Zamora P, et al. Serum albumin and other
Nutr 1988;48:1173–1178. prognostic factors related to response and survival in patients with
27. Levkoff SE, Safran C, Cleary PD, et al. Identification of factors advanced non-small cell lung cancer. Lung Cancer. 1995;12:67–76.
associated with the diagnosis of delirium in elderly hospitalized 50. Lowrie EG, Huang WH, Lew NL. Death risk predictors among
patients. J Am Geriatr Soc. 1988;36:1099 –1104. peritoneal dialysis and hemodialysis patients: a preliminary compar-
28. Trzepacz PT, Brenner RP, Coffman G, et al. Delirium in liver ison. Am J Kidney Dis. 1995;26:220 –228.
transplantation candidates: discriminant analysis of multiple test vari- 51. Violi F, Ferro D, Basili S, et al. Prognostic value of clotting and
ables. Biol Psychiatry. 1988;24:3–14. fibrinolytic systems in a follow-up of 165 liver cirrhotic patients.
29. Bladé J, Rozman C, Cervantes F, et al. A new prognostic system for CALC Group. Hepatology. 1995;22:96 –100.
multiple myeloma based on easily available parameters. Br J Haema- 52. Avram MM, Fein PA, Bonomini L, et al. Predictors of survival in
tol. 1989;72:507–511. continuous ambulatory peritoneal dialysis patients: a five-year pro-
spective study. Perit Dial Int. 1996;16:S190 –194.
30. Rich MW, Keller AJ, Schechtman KB, et al. Increased complications
53. Foley RN, Parfrey PS, Harnett JD, et al. Hypoalbuminemia, cardiac
and prolonged hospital stay in elderly cardiac surgical patients with
morbidity, and mortality in end-stage renal disease. J Am Soc Neph-
low serum albumin. Am J Cardiol. 1989;63:714 –718.
rol. 1996;7:728 –736.
31. Altomare DF, Serio G, Pannarale OC, et al. Prediction of mortality by
54. Guijarro C, Massy ZA, Wiederkehr MR, et al. Serum albumin and
logistic regression analysis in patients with postoperative enterocu-
mortality after renal transplantation. Am J Kidney Dis. 1996;27:117–
taneous fistulae. Br J Surg. 1990;77:450 – 453.
123.
32. Lai EC, Tam PC, Paterson IA, et al. Emergency surgery for severe
55. Iseki K, Uehara H, Nishime K, et al. Impact of the initial levels of
acute cholangitis. The high-risk patients. Ann Surg. 1990;211:55–59.
laboratory variables on survival in chronic dialysis patients. Am J
33. Lowrie EG, Lew NL. Death risk in hemodialysis patients: the pre-
Kidney Dis. 1996;28:541–548.
dictive value of commonly measured variables and an evaluation of
56. Magovern JA, Sakert T, Magovern GJ, et al. A model that predicts
death rate differences between facilities. Am J Kidney Dis. 1990;15:
morbidity and mortality after coronary artery bypass graft surgery.
458 – 482.
J Am Coll Cardiol. 1996;28:1147–1153.
34. Sullivan DH, Patch GA, Walls RC, et al. Impact of nutrition status on 57. Magovern JA, Sakert T, Benckart DH, et al. A model for predicting
morbidity and mortality in a select population of geriatric rehabilita- transfusion after coronary artery bypass grafting. Ann Thorac Surg.
tion patients. Am J Clin Nutr. 1990;51:749 –758. 1996;61:27–32.
35. Cherng NC, Asal NR, Kuebler JP, et al. Prognostic factors in multiple 58. Pacelli F, Doglietto GB, Alfieri S, et al. Prognosis in intra-abdominal
myeloma. Cancer. 1991; 67:3150 –3156. infections. Multivariate analysis on 604 patients. Arch Surg. 1996;
36. Levis A, Ficara F, Marmont F, et al. Prognostic significance of serum 131:641– 645.
albumin in chronic lymphocytic leukemia. Haematologica. 1991;76: 59. Friedenberg F, Jensen G, Gujral N, et al. Serum albumin is predictive
113–119. of 30-day survival after percutaneous endoscopic gastrostomy. JPEN
37. United States Renal Data System. Comorbid conditions and correla- J Parenter Enteral Nutr. 1997;21:72–74.
tions with mortality risk among 3,399 incident hemodialysis patients. 60. Rady MY, Ryan T, Starr NJ. Clinical characteristics of preoperative
Am J Kidney Dis. 1992;20:32–38. hypoalbuminemia predict outcome of cardiovascular surgery. JPEN J
38. Burr RG, Clift-Peace L, Nuseibeh I. Haemoglobin and albumin as Parenter Enteral Nutr. 1997;21:81–90.
predictors of length of stay of spinal injured patients in a rehabilita- 61. Rady MY, Ryan T, Starr NJ. Early onset of acute pulmonary dys-
tion centre. Paraplegia. 1993;31:473– 478. function after cardiovascular surgery: risk factors and clinical out-
39. Ferguson RP, O’Connor P, Crabtree B, et al. Serum albumin and come. Crit Care Med 1997;25:1831–1839.
prealbumin as predictors of clinical outcomes of hospitalized elderly 62. Ryan TA, Rady MY, Bashour CA, et al. Predictors of outcome in
nursing home residents. J Am Geriatr Soc. 1993;41:545–549. cardiac surgical patients with prolonged intensive care stay. Chest
40. Goldwasser P, Mittman N, Antignani A, et al. Predictors of mortality 1997;112:1035–1042.
in hemodialysis patients. J Am Soc Nephrol. 1993;3:1613–1622. 63. Bologa RM, Levine DM, Parker TS, et al. Interleukin-6 predicts
41. McEllistrum MC, Collins JC, Powers JS. Admission serum albumin hypoalbuminemia, hypocholesterolemia, and mortality in hemodial-
level as a predictor of outcome among geriatric patients. South Med ysis patients. Am J Kidney Dis. 1998;32:107–114.
J. 1993;86:1360 –1361. 64. Chertow GM, Lazarus JM, Paganini EP, et al. Predictors of mortality
42. Owen WF, Jr, Lew NL, Liu Y, et al. The urea reduction ratio and and the provision of dialysis in patients with acute tubular necrosis.
serum albumin concentration as predictors of mortality in patients The Auriculin Anaritide Acute Renal Failure Study Group. J Am Soc
undergoing hemodialysis. N Engl J Med. 1993;329:1001–1006. Nephrol 1998;9:692– 698.
43. Rocco MV, Jordan JR, Burkart JM. The efficacy number as a pre- 65. Gariballa SE, Parker SG, Taub N, et al. Influence of nutritional status
dictor of morbidity and mortality in peritoneal dialysis patients. J Am on clinical outcome after acute stroke. Am J Clin Nutr. 1998;68:275–
Soc Nephrol. 1993;4:1184 –1191. 281.
Vol. 237 ● No. 3 Hypoalbuminemia 333

66. Göl MK, Karahan M, Ulus AT, et al. Bloodstream, respiratory, and 88. Gerhards MF, van Gulik TM, de Wit LT, et al. Evaluation of
deep surgical wound infections after open heart surgery. J Card Surg. morbidity and mortality after resection for hilar cholangiocarcino-
1998;13:252–259. ma—a single-center experience. Surgery. 2000;127:395– 404.
67. Hedström M, Gröndal L, Örtquist A, et al. Serum albumin and deep 89. Moon KH, Song IS, Yang WS, et al. Hypoalbuminemia as a risk
infection in femoral neck fractures. A study of 437 cases followed for factor for progressive left-ventricular hypertrophy in hemodialysis
one year. Int Orthop. 1998;22:182–184. patients. Am J Nephrol. 2000;20:396 – 401.
68. Ifudu O, Paul HR, Homel P, et al. Predictive value of functional 90. Nair S, Hertan H, Pitchumoni CS. Hypoalbuminemia is a poor
status for mortality in patients on maintenance hemodialysis. Am J predictor of survival after percutaneous endoscopic gastrostomy in
Nephrol. 1998;18:109 –116. elderly patients with dementia. Am J Gastroenterol. 2000;95:133–
69. Incalzi R, Capparella O, Gemma A, et al. Inadequate caloric intake: 136.
a risk factor for mortality of geriatric patients in the acute-care 91. Oki I, Tanihara S, Ojima T, et al. A multicenter collaborative study
hospital. Age Ageing. 1998;27:303–310. on the risk factors of cardiac sequelae due to Kawasaki disease: a
70. Leavey SF, Strawderman RL, Jones CA, et al. Simple nutritional one-year follow-up study. Acta Paediatr. 2000;89:1435–1438.
indicators as independent predictors of mortality in hemodialysis 92. Sharma AP, Gupta A, Sharma RK, et al. Does serum albumin at start
patients. Am J Kidney Dis. 1998;31:997–1006. of continuous ambulatory peritoneal dialysis (CAPD) or its drop
71. Marinella MA, Markert RJ. Admission serum albumin level and during CAPD determine patient outcome? Adv Perit Dial. 2000;16:
length of hospitalization in elderly patients. South Med J. 1998;91: 119 –122.
851– 854. 93. Tanna MM, Vonesh EF, Korbet SM. Patient survival among incident
72. Noh H, Lee SW, Kang SW, et al. Serum C-reactive protein: a peritoneal dialysis and hemodialysis patients in an urban setting.
predictor of mortality in continuous ambulatory peritoneal dialysis Am J Kidney Dis. 2000;36:1175–1182.
patients. Perit Dial Int. 1998;18:387–394. 94. Yeun JY, Levine RA, Mantadilok V, et al. C-reactive protein predicts
73. Owen WF Jr, Chertow GM, Lazarus JM, et al. Dose of hemodialysis all-cause and cardiovascular mortality in hemodialysis patients. Am J
and survival: differences by race and sex. JAMA 1998;280:1764 – Kidney Dis. 2000;35:469 – 476.
1768. 95. Cueto-Manzano AM, Quintana-Piña E, Correa-Rotter R. Long-term
74. Owen WF, Lowrie EG. C-reactive protein as an outcome predictor CAPD survival and analysis of mortality risk factors: 12-year expe-
for maintenance hemodialysis patients. Kidney Int. 1998;54:627– 636. rience of a single Mexican center. Perit Dial Int. 2001;21:148 –153.
75. Rady MY, Ryan T, Starr NJ. Perioperative determinants of morbidity 96. Gulati S, Stephens D, Balfe JA, et al. Children with hypoalbumine-
and mortality in elderly patients undergoing cardiac surgery. Crit mia on continuous peritoneal dialysis are at risk for technique failure.
Care Med. 1998;26:225–235. Kidney Int. 2001;59:2361–2367.
76. Becker BN, Becker YT, Heisey DM, et al. The impact of hypoalbu- 97. Kalantar-Zadeh K, Kopple JD, Block G, et al. A malnutrition-inflam-
minemia in kidney-pancreas transplant recipients. Transplantation. mation score is correlated with morbidity and mortality in mainte-
1999;68:72–75. nance hemodialysis patients. Am J Kidney Dis. 2001;38:1251–1263.
77. Deschènes M, Villeneuve JP. Risk factors for the development of 98. Leung HB, Ho YC, Carnett J, et al. Diabetic foot ulcers in the Hong
bacterial infections in hospitalized patients with cirrhosis. Am J Kong Chinese population: retrospective study. Hong Kong Med J.
Gastroenterol. 1999;94:2193–2197. 2001;7:350 –355.
78. Engelman DT, Adams DH, Byrne JG, et al. Impact of body mass 99. Rozzini R, Sabatini T, Trabucchi M. Prediction of 6-month mortality
index and albumin on morbidity and mortality after cardiac surgery. among older hospitalized adults. JAMA. 2001;286:1315–1316.
J Thorac Cardiovasc Surg. 1999;118:866 – 873. 100. Scott JD, Forrest A, Feuerstein S, et al. Factors associated with
79. Obialo CI, Okonofua EC, Nzerue MC, et al. Role of hypoalbumin- postoperative infection. Infect Control Hosp Epidemiol. 2001;22:
emia and hypocholesterolemia as copredictors of mortality in acute 347–351.
renal failure. Kidney Int. 1999;56:1058 –1063. 101. Tokars JI, Light P, Anderson J, et al. A prospective study of vascular
80. Ohashi H, Oda H, Ohno M, et al. Predictors of survival in continuous access infections at seven outpatient hemodialysis centers. Am J
ambulatory peritoneal dialysis patients: the importance of left ven- Kidney Dis. 2001;37:1232–1240.
tricular hypertrophy and diabetic nephropathy. Adv Perit Dial. 1999; 102. Tveit DP, Hypolite I, Bucci J, et al. Risk factors for hospitalizations
15:87–90. resulting from pulmonary embolism after renal transplantation in the
81. Rady MY, Ryan T. Perioperative predictors of extubation failure and United States. J Nephrol. 2001;14:361–368.
the effect on clinical outcome after cardiac surgery. Crit Care Med. 103. Walter LC, Brand RJ, Counsell SR, et al. Development and validation
1999;27:340 –347. of a prognostic index for 1-year mortality in older adults after
82. Zimmermann J, Herrlinger S, Pruy A, et al. Inflammation enhances hospitalization. JAMA. 2001;285:2987–2994.
cardiovascular risk and mortality in hemodialysis patients. Kidney 104. Bakewell AB, Higgins RM, Edmunds ME. Quality of life in perito-
Int. 1999;55:648 – 658. neal dialysis patients: Decline over time and association with clinical
83. Axdorph U, Sjoberg J, Grimfors G, et al. Biological markers may add outcomes. Kidney Int. 2002;61:239 –248.
to prediction of outcome achieved by the International Prognostic 105. Klassen PS, Lowrie EG, Reddan DN, et al. Association between
Score in Hodgkin’s disease. Ann Oncol. 2000;11:1405–1411. pulse pressure and mortality in patients undergoing maintenance
84. Bashour CA, Yared JP, Ryan TA, et al. Long-term survival and hemodialysis. JAMA 2002;287:1548 –1555.
functional capacity in cardiac surgery patients after prolonged inten- 106. Modawal A, Candadai NP, Mandell KM, et al. Weaning success
sive care. Crit Care Med. 2000;28:3847–3853. among ventilator-dependent patients in a rehabilitation facility. Arch
85. Chung SH, Chu WS, Lee HA, et al. Peritoneal transport character- Phys Med Rehabil. 2002;83:154 –157.
istics, comorbid diseases and survival in CAPD patients. Perit Dial 107. Sezer S, Özdemir FN, Arat Z, et al. The association of interdialytic
Int. 2000;20:541–547. weight gain with nutritional parameters and mortality risk in hemo-
86. Dharmarajan T, Sipalay M, Shyamsundar R, et al. Co-morbidity, not dialysis patients. Ren Fail. 2002;24:37– 48.
age predicts adverse outcome in Clostridium difficile colitis. World J 108. Stefoni S, Cianciolo G, Donati G, et al. Low TGF-␤1 serum levels are
Gastroenterol. 2000;6:198 –201. a risk factor for atherosclerosis disease in ESRD patients. Kidney Int.
87. Dhiman RK, Makharia GK, Jain S, et al. Ascites and spontaneous 2002;61:324 –335.
bacterial peritonitis in fulminant hepatic failure. Am J Gastroenterol. 109. Tincani E, Mazzali F, Morini L. Hypoalbuminemia as a risk factor for
2000;95:233–238. over-anticoagulation. Am J Med. 2002;112:247–248.
334 Vincent and Others Ann. Surg. ● March 2003

110. Wong CS, Hingorani S, Gillen DL, et al. Hypoalbuminemia and risk 120. Krauer B, Dayer P, Anner R. Changes in serum albumin and ␣1-acid
of death in pediatric patients with end-stage renal disease. Kidney Int. glycoprotein concentrations during pregnancy: an analysis of fetal-
2002;61:630 – 637. maternal pairs. Br J Obstet Gynaecol. 1984;91:875– 881.
111. McMurray L-G, Roe JH, Sweet LK. Plasma protein studies on 121. Blunt MC, Nicholson JP, Park GR. Serum albumin and colloid
normal newborn and premature infants. I. Plasma protein values for osmotic pressure in survivors and nonsurvivors of prolonged critical
normal full term and normal premature infants; II. Use of concen- illness. Anaesthesia. 1998;53:755–761.
trated normal human serum albumin in treatment of premature in- 122. Emerson TE, Jr. Unique features of albumin: a brief review. Crit
fants. Am J Dis Child. 1948;75:265–278. Care Med. 1989;17:690 – 694.
112. Smith CA, Phillips KG, Roth RO. Effects and fate of human serum 123. Zoellner H, Hou JY, Lovery M, et al. Inhibition of microvascular
albumin administered intravenously and orally to premature infants. endothelial apoptosis in tissue explants by serum albumin. Microvasc
J Clin Invest. 1950;29:218 –226. Res. 1999;57:162–173.
113. Ford EG, Jennings LM, Andrassy RJ. Serum albumin (oncotic pres- 124. Mimoz O, Soreda S, Padoin C, et al. Ceftriaxone pharmacokinetics
sure) correlates with enteral feeding tolerance in the pediatric surgical during iatrogenic hydroxyethyl starch-induced hypoalbuminemia: a
model to explore the effects of decreased protein binding capacity on
patient. J Pediatr Surg. 1987;22:597–599.
highly bound drugs. Anesthesiology. 2000;93:735–743.
114. Brown RO, Bradley JE, Bekemeyer WB, et al. Effect of albumin
125. Brown JM, Beehler CJ, Berger EM, et al. Albumin decreases hydro-
supplementation during parenteral nutrition on hospital morbidity.
gen peroxide and reperfusion injury in isolated rat hearts. Inflamma-
Crit Care Med. 1988; 16:1177–1182.
tion. 1989;13:583–589.
115. Foley EF, Borlase BC, Dzik WH, et al. Albumin supplementation in
126. Strubelt O, Younes M, Li Y. Protection by albumin against isch-
the critically ill. A prospective, randomized trial. Arch Surg. 1990;
aemia- and hypoxia-induced hepatic injury. Pharmacol Toxicol.
125:739 –742. 1994;75:280 –284.
116. Kanarek KS, Williams PR, Blair C. Concurrent administration of 127. Soejima A, Matsuzawa N, Miyake N, et al. Hypoalbuminemia accel-
albumin with total parenteral nutrition in sick newborn infants. JPEN erates erythrocyte membrane lipid peroxidation in chronic hemodi-
J Parenter Enteral Nutr. 1992;16:49 –53. alysis patients. Clin Nephrol. 1999;51:92–97.
117. Wojtysiak SL, Brown RO, Roberson D, et al. Effect of hypoalbu- 128. Wu TW, Wu J, Li RK, et al. Albumin-bound bilirubins protect human
minemia and parenteral nutrition on free water excretion and elec- ventricular myocytes against oxyradical damage. Biochem Cell Biol.
trolyte-free water resorption. Crit Care Med. 1992;20:164 –169. 1991;69:683– 688.
118. Golub R, Sorrento JJ Jr, Cantu R Jr, et al. Efficacy of albumin 129. Greenough A, Emery E, Hird MF, et al. Randomised controlled trial
supplementation in the surgical intensive care unit: a prospective, of albumin infusion in ill preterm infants. Eur J Pediatr. 1993;152:
randomized study. Crit Care Med. 1994;22:613– 619. 157–159.
119. Rubin H, Carlson S, DeMeo M, et al. Randomized, double-blind 130. Quinlan GJ, Margarson MP, Mumby S, et al. Administration of
study of intravenous human albumin in hypoalbuminemic patients re- albumin to patients with sepsis syndrome: a possible beneficial role in
ceiving total parenteral nutrition. Crit Care Med. 1997;25:249–252. plasma thiol repletion. Clin Sci (Colch). 1998;95:459 – 465.