|Home | About | Journals | Submit | Contact Us | Français|
Contributors (listed alphabetically): Phil Alderson (UK Cochrane Centre) searched The Cochrane Controlled Trials Register for relevant trials, extracted the data from the trials, and commented on the paper. Frances Bunn (Institute of Child Health) searched the Cochrane Injuries Group Specialised Register for relevant trials, obtained copies of relevant papers, wrote to authors for further information on allocation concealment, and commented on the paper. Carol Lefebvre (UK Cochrane Centre) designed the search strategies for The Cochrane Controlled Trials Register and Embase, and searched these two databases for relevant trials. Leah Li (Institute of Child Health) did the funnel plot and the regression test of funnel plot asymmetry. Alain Li Wan Po (Centre for Evidence-Based Pharmacotherapy, University of Nottingham) helped to write the paper. Ian Roberts (Institute of Child Health) designed the protocol, extracted data from the trials, contacted authors for unpublished data, and wrote the paper. Gillian Schierhout proposed the study hypothesis and conducted preliminary searches of Medline, Embase, and BIDS Index to Scientific and Technical Proceedings.
Objective: To quantify effect on mortality of administering human albumin or plasma protein fraction during management of critically ill patients.
Design: Systematic review of randomised controlled trials comparing administration of albumin or plasma protein fraction with no administration or with administration of crystalloid solution in critically ill patients with hypovolaemia, burns, or hypoalbuminaemia.
Subjects: 30 randomised controlled trials including 1419 randomised patients.
Main outcome measure: Mortality from all causes at end of follow up for each trial.
Results: For each patient category the risk of death in the albumin treated group was higher than in the comparison group. For hypovolaemia the relative risk of death after albumin administration was 1.46 (95% confidence interval 0.97 to 2.22), for burns the relative risk was 2.40 (1.11 to 5.19), and for hypoalbuminaemia it was 1.69 (1.07 to 2.67). Pooled relative risk of death with albumin administration was 1.68 (1.26 to 2.23). Pooled difference in the risk of death with albumin was 6% (95% confidence interval 3% to 9%) with a fixed effects model. These data suggest that for every 17 critically ill patients treated with albumin there is one additional death.
Conclusions: There is no evidence that albumin administration reduces mortality in critically ill patients with hypovolaemia, burns, or hypoalbuminaemia and a strong suggestion that it may increase mortality. These data suggest that use of human albumin in critically ill patients should be urgently reviewed and that it should not be used outside the context of rigorously conducted, randomised controlled trials.
In patients with acute and chronic illness serum albumin concentration is inversely related to risk of death. A systematic review of cohort studies meeting specified criteria estimated that for each 2.5g/l decrement in serum albumin concentration the risk of death increases by between 24% and 56%.1 The association persists after adjustment for other known risk factors and pre-existing illness, and some commentators have suggested the possibility of the albumin molecule having a direct protective effect.1 Partly as a result of the association between serum albumin and mortality, human albumin solutions are now used in the management of a diverse range of medical and surgical problems. Licensed indications for human albumin solution are the emergency treatment of shock and other conditions in which restoration of blood volume is urgent, the acute management of burns, and clinical situations associated with hypoproteinaemia.2
Compared with other colloidal solutions and with crystalloid solutions, human albumin solutions are expensive.3 Volume for volume, human albumin solution is twice as expensive as hydroxyethyl starch and over 30 times more expensive than crystalloid solutions such as sodium chloride or Ringer’s lactate. Because of the high cost and limited availability of human albumin, it is imperative that its use should be restricted to the indications for which it has been shown to be effective. To quantify the effect on mortality of human albumin solution in the management of critically ill patients with hypovolaemia from injury or surgery, burns, and hypoproteinaemia, we conducted a systematic review of randomised controlled trials.
Our aim was to identify all relevant randomised controlled trials that were available for review by March 1998. A randomised controlled trial was defined as a trial in which the subjects followed were assigned prospectively to one of two (or more) interventions by random allocation or some quasi-random method of allocation. This definition was agreed at an international meeting held in Oxford in November 1992 in association with the official opening of the UK Cochrane Centre. We sought to identify all randomised controlled trials of administration of human albumin or plasma protein fraction (supplemental albumin or plasma protein fraction compared with no albumin or plasma protein fraction or with a crystalloid solution) in critically ill patients with hypovolaemia from trauma or surgery, with burns, or with hypoalbuminaemia. Studies that compared different levels of albumin supplementation were also included.
Trials were identified by computerised searches of the Cochrane Controlled Trials Register, Medline, Embase, and BIDS Index to Scientific and Technical Proceedings (search strategies are available from IR); by hand searching 29 international journals and the proceedings of several international meetings on fluid resuscitation; by checking the reference lists of all included trials; and by contacting the authors of identified trials and asking them about any other published or unpublished trials that may have been conducted. There were no language restrictions. To identify unpublished trials we searched the register of the Medical Editors’ Trial Amnesty,4 and contacted the Medical Directors of Bio Products Laboratory (Zenalb), Centeon (Albuminar), and Alpha Therapeutic UK (Albutein).
The outcome measure was mortality from all causes at the end of the follow up period scheduled for each trial. For all trials we collected data on the type of participants, details about the interventions, the quality of concealment of allocation, and mortality at the end of follow up. We rated quality of allocation concealment using the method proposed by Schulz et al.5 We sought mortality data in simple categorical form, and we did not extract data on time to death. If a report did not include the numbers of deaths in each group, we sought these data from the authors. Two reviewers independently extracted the data, and any disagreements were resolved by discussion.
We used the Mantel-Haenszel method to calculate relative risks, risk differences, and 95% confidence intervals for death for each trial on an intention to treat basis using RevMan (Review Manager) statistical software. When there are no events in one group the software adds 0.5 to each cell of the 2×2 table. We tested heterogeneity between trials using χ2 tests, with P0.05 indicating significant heterogeneity. As long as statistical heterogeneity did not exist, we used a fixed effects model to calculate summary relative risks and 95% confidence intervals.
To examine the extent to which the results of the meta-analyses may have been biased as a result of the selective inclusion of randomised trials with positive findings (publication and other selection bias), we prepared a funnel plot and used the regression approach to assessing funnel plot asymmetry proposed by Egger et al.6 We used the log odds ratio in the funnel plot because this is the measure that is used in the regression test of funnel plot asymmetry as described by Egger et al. Using simple unweighted linear regression, we regressed the standard normal deviate (defined as the log odds ratio divided by its standard error) against the estimate’s precision (defined as the inverse of the standard error). The larger the deviation of the intercept of the regression line from zero, the greater the asymmetry and the more likely it is that the meta-analysis will yield biased estimates of effect. As suggested by Egger et al, we considered P<0.1 to indicate significant asymmetry.
We identified a total of 32 randomised controlled trials that met the study’s inclusion criteria.7–38 The table shows details of these trials. Mortality data were available either from the published report or on contact with the authors in 30 of these trials. The two trials for which mortality data could not be obtained included a total of 42 randomised patients, comprising 3% of the total number of randomised patients in all trials meeting our inclusion criteria.22,31 One of the trials was an unpublished trial registered in the Medical Editors’ Trial Amnesty, and we obtained further details, including data on mortality, directly from the trialist. In six trials there were no deaths in either the intervention or comparison groups.8,12,23,25,26,35
The trial by Lucas et al was reported in five publications.21,39–42 An early report gave the mortality data for 52 randomised patients, 27 allocated to receive albumin and 25 allocated to receive no albumin.21 Subsequent publications indicated that recruitment to the trial continued until 94 patients were randomised. Mortality data for all the 94 patients were not published, nor were they available on contact with the author. Consequently, we present the outcome data for the 52 patients.
Of the 24 trials in which one or more deaths occurred in either the intervention or control groups, 13 included a method of allocation concealment that would be expected to reduce the risk of foreknowledge of treatment allocation (pharmacy controlled randomisation or serially numbered sealed opaque envelopes). In seven trials this was unclear, and in four trials concealment was inadequate (table).
In each of the patient categories the risk of death in the albumin treated group was higher than in the comparison group (fig (fig1).1). For hypovolaemia the relative risk of death after albumin administration was 1.46 (95% confidence interval 0.97 to 2.22), for burns the relative risk was 2.40 (1.11 to 5.19), and for hypoalbuminaemia the it was 1.69 (1.07 to 2.67). There was no significant heterogeneity either between or within the groups of trials, or overall (χ2=15.32, df=23, P>0.2). The pooled relative risk of death with albumin administration was 1.68 (1.26 to 2.23).
There was no significant heterogeneity in the risk difference for mortality (χ2=36.69, df=29, P>0.1). The pooled difference in the risk of death with albumin was 6% (95% confidence interval 3% to 9%).
Figure Figure22 shows a funnel plot for the 24 trials in which deaths occurred. There was no clear evidence of asymmetry, and the regression approach to funnel plot asymmetry yielded an intercept of −0.39 and P=0.33, indicating no statistical evidence of selection bias.
We repeated the analyses for the 13 trials with deaths in which allocation concealment was adequate.13,15,16,19,20,24,27–29,32,34,37,38 For hypovolaemia the relative risk of death with albumin administration was 1.39 (0.80 to 2.40), for burns the relative risk was 2.47 (0.69 to 8.79), and for hypoalbuminaemia it was 1.71 (0.92 to 3.18). There was no substantial heterogeneity between the trials in the various categories (χ2=4.42, df=12, P>0.2), and the pooled relative risk of death with albumin administration was 1.61 (1.09 to 2.38). Thus, restricting the analyses to the adequately concealed trials had almost no effect on the relative risks in each group or overall.
We found no evidence that albumin reduced mortality and a strong suggestion that it might increase the risk of death in patients with hypovolaemia, burns, or hypoproteinaemia. Overall, the risk of death in patients treated with albumin was 6% (95% confidence interval 3% to 9%) higher than in patients not given albumin.
Mortality was selected as the outcome measure in this systematic review for several reasons. In the context of critical illness, death or survival is a clinically relevant outcome that is of immediate importance to patients, and data on death are reported in nearly all studies. Furthermore, one might expect that mortality data would be less prone to measurement error or biased reporting than would data on pathophysiological outcomes. The use of a pathophysiological end point as a surrogate for an adverse outcome assumes a direct relationship between the two, an assumption that may sometimes be inappropriate. Finally, when trials collect data on a number of physiological end points, there is the potential for bias due to the selective publication of end points showing striking treatment effects. Because we obtained mortality data for all but two of the included trials, the likelihood of bias due to selective publication of trial outcomes is minimal. We examined mortality from all causes because the attribution of cause of death in critically ill patients, many of whom may have multiorgan failure, can be problematic and may be prone to bias. Length of follow up was not specified in many of the trials, but when these data were available, follow up was for the first week or until hospital discharge.
Although publication bias is a potent threat to the validity of systematic reviews, it is unlikely to have had an important impact in this study. There was no evidence of funnel plot asymmetry on visual inspection, and there was no statistical evidence of asymmetry from linear regression analysis.
In some of the trials included in this review allocation concealment was inadequate or unclear. As a result, it is possible that more severely ill patients were preferentially allocated to albumin treated groups, which could account for the increased mortality in these groups. Nevertheless, when we repeated the analyses for only those trials in which the method of allocation concealment would be expected to reduce the risk of foreknowledge of allocation, the point estimates were almost identical.
To what extent are the results of this review of 30 relatively small randomised trials of albumin administration generalisable to clinical practice? We believe that this is a matter for judgment by the responsible clinician faced with an individual patient.43 However, the advantage of an overview such as ours is that, since it includes many studies, the results are based on a wide range of patients. Because the results were consistent across the studies, they might reasonably be taken to apply to this wide variety of patients.43 Moreover, the evidence that we have brought together is, as far as we can ensure, the totality of the available randomised evidence for the use of albumin in hypovolaemia, burns, and hypoalbuminaemia, the indications for which albumin is currently licensed.
Is there a plausible mechanism by which human albumin might increase mortality? Albumin is used in hypovolaemia and hypoalbuminaemia because it is believed to be effective in replacing volume and supporting colloid oncotic pressure.44 However, albumin is also believed to have anticoagulant properties, inhibiting platelet aggregation and enhancing the inhibition of factor Xa by antithrombin III.44 Such anticoagulant activity might be detrimental in critically ill patients, particularly those with haemorrhagic hypovolaemia. Furthermore, albumin has been shown to distribute across the capillary membrane, a process that is accelerated in critically ill patients.45 It has been suggested that increased leakage of albumin into the extravascular spaces might reduce the oncotic pressure difference across the capillary wall, making oedema more likely.45
Because this review was based on relatively small trials in which there were only a small number of deaths the results must be interpreted with caution. Nevertheless, we believe that a reasonable conclusion from these results is that the use of human albumin in the management of critically ill patients should be reviewed. A strong argument could be made that human albumin should not be used outside the context of a properly concealed and otherwise rigorously conducted randomised controlled trial with mortality as the end point. Until such data become available, there is also a case for a review of the licensed indications for albumin use.
This review will also be published in the Cochrane Library, where it will be regularly updated to take account of new data and comments on this version.
Editorial by Offringa and Letters p277
We thank the Intensive Care National Audit and Research Centre in London for help with identifying trials for this review and for their extensive hand searching. We thank AJ Woittiez for providing unpublished trial data from the trial that was registered in the Medical Editors’ Trial Amnesty. We thank Elizabeth Bryant, information officer at Centeon, and Martin O’Fobve, at Bio Products, for searching their databases for albumin trials. We thank Anne Greenough for re-examining individual patient records in order to provide data on mortality. We thank Iain Chalmers, Jos Kleijnen, Richard Peto, Dave Signorini, and David Yates for their comments on the manuscript.
Funding: The infrastructure of the Cochrane Injuries Group is supported by the NHS Research and Development Programme.
Conflict of interest: None.
Starling’s principle is often represented as the leakage of fluids from the arterial end of capillaries, where the hydrostatic pressure is greater than the oncotic pressure (derived from the plasma proteins), and the reabsorption of fluid into the venous end, where the oncotic pressure exceeds the hydrostatic pressure. A small excess of fluid in the interstitial space—when filtration from the capillaries is greater than reabsorption—is dealt with by lymphatic drainage from the interstitial space. The rationale for giving albumin solutions rather than crystalloid solutions in cases of hypovolaemic shock is that fluid reabsorption from the interstitial space is enhanced, and fluid therefore remains in the vascular system for longer.
But in recent years the assumed reabsorption of fluid at the venous end of capillaries has been challenged. There is now good evidence to show that, except in the gut and the renal circulation, there is no sustained reabsorption of fluid at the venous end of capillaries. Instead, there is a small constant level of filtration from the capillaries, restrained by the osmotic pressure of the plasma proteins. In some rare circumstances—for example, in hypovolaemic shock—there is a transient reabsorption of fluid, but this lasts for only a few minutes and it amounts to an “internal transfusion” of about 500ml of fluid over 15 minutes.
The production of life threatening pulmonary oedema begins when the loss of protein and fluid from the blood vessels exceeds the volume of fluid that can be drained from the interstitial space by the lymphatics. In some disease states or when tissue is damaged, as in severe burns, the capillary walls become very much more permeable under the influence of direct cellular damage and from inflammatory mediators. The filtration of fluids, together with proteins, out into the interstitial space is greatly increased and cannot be matched by lymphatic drainage. The filtration rate may be further increased by a fall in the hydrostatic pressure in the interstitial space as a result of tissue damage, so that even more fluid is sucked out of the capillaries.
Conventionally, colloids such as albumin are administered to these patients in an attempt to maintain their intravascular volume, but because of the increased permeability of the vessels, the albumin solution becomes much less effective in maintaining plasma volume than in healthy individuals who have normal vessel permeability. Thus the rationale for administering albumin solutions becomes much less clear. In disease states such as the nephrotic syndrome, for example, there is new evidence to show that protein is lost not only from the renal circulation owing to greater permeability of the renal vessels, but also from the rest of the systemic circulation. This being the case, it is difficult to see how the administration of albumin could ever replace the deficit without causing further problems.
Abi Berger—Science editor, BMJ