Search tips
Search criteria 


Logo of bmjThis ArticleThe BMJ
BMJ. 2007 December 15; 335(7632): 1248–1251.
Published online 2007 November 23. doi:  10.1136/bmj.39398.682500.25
PMCID: PMC2137061

Antithrombin III in critically ill patients: systematic review with meta-analysis and trial sequential analysis

Arash Afshari, senior resident ,1 Jørn Wetterslev, staff specialist,2 Jesper Brok, PhD student ,2 and Ann Møller, staff specialist 3


Objective To evaluate the benefits and harms of antithrombin III in critically ill patients.

Design Systematic review and meta-analysis of randomised trials.

Data sources CENTRAL, Medline, Embase, International Web of Science, LILACS, the Chinese Biomedical Literature Database, and CINHAL (to November 2006); hand search of reference lists, contact with authors and experts, and search of registers of ongoing trials.

Review methods Two reviewers independently selected parallel group randomised clinical trials comparing antithrombin with placebo or no intervention and extracted data related to study methods, interventions, outcomes, bias risk, and adverse events. Disagreements were resolved by discussion. Trials in any type of critically ill patients in intensive care were eligible. All trials, irrespective of blinding or language status, that compared any antithrombin III regimen with no intervention or placebo were included. Trials were considered to be at low risk of bias if they had adequate randomisation procedure, blinding, and used intention to treat analysis. Risk ratios with 95% confidence intervals were estimated with fixed and random effects models according to heterogeneity.

Main outcome measures Mortality, length of stay in intensive care or hospital, quality of life, severity of sepsis, respiratory failure, duration of mechanical ventilation, incidence of surgical intervention, intervention effect among various populations, and adverse events (such as bleeding).

Results 20 trials randomly assigning 3458 patients met inclusion criteria. Eight trials had low risk of bias. Compared with placebo or no intervention, antithrombin III did not reduce overall mortality (relative risk 0.96, 95% confidence interval 0.89 to 1.03). No subgroup analyses on risk of bias, populations of patients, or with and without adjuvant heparin yielded significant results. Antithrombin III increased the risk of bleeding events (1.52, 1.30 to 1.78). Heterogeneity was observed in only a few analyses.

Conclusion Antithrombin III cannot be recommended for critically ill patients based on the available evidence.


Mortality in critically ill patients is substantial and can reach as high as 30-50% in those with severe sepsis and septic shock.1 Each year in the United States around 750 000 patients become critically ill and are admitted to intensive care units.2 Critical illness results in uncontrolled inflammation and vascular damage even when the cause of the illness is not infection (sepsis)—for example, trauma, malignancy, complications of pregnancy, poisoning, allergic reactions, and liver failure.3 During critical illness, systemic activation of coagulation can occur, which can result in disseminated intravascular coagulation. This is characterised by simultaneous widespread microvascular thrombosis and profuse bleeding from various sites.4

Antithrombin III is a potent anticoagulant with anti-inflammatory properties. It is thought to have an inhibitory effect on the proinflammatory and procoagulant processes.5 6 The blood concentration of antithrombin III falls by 20-40% in critically ill (septic) patients and correlates with severity of disease.5 7 Hence, theoretically replenishing concentrations might benefit critically ill patients. Furthermore, its interaction with heparin, which is a standard treatment for patients with disseminated intravascular coagulation, seems important.w1

Four small meta-analyses assessed the use of antithrombin III in critically ill patients. Two assessed patients with severe sepsis, one included three randomised clinical trials,w2 and one included eight trials (both randomised and non-randomised).8 One recent meta-analysis examined the effect of antithrombin III on mortality among 364 patients with disseminated intravascular coagulation in three trials.w3 Finally, a Cochrane systematic review with two randomised controlled trials assessed use of antithrombin III for respiratory distress syndrome in preterm infants.9 These meta-analyses had selected populations. All found non-significant effects on mortality, with wide confidence intervals (that is, “absence of evidence”). Only one of these meta-analyses was a systematic review.9

Information from the Danish Medicinal Agency ( suggests that the extrapolated annual costs of antithrombin III in the US and the European Union are around £70m (€100m, $145m). The clinical benefit of supplementation in critically ill patients, however, is still controversial and its efficacy debated.


Study selection

We searched the Cochrane central register of controlled trials (CENTRAL), Medline (1950 to November 2006), Embase (1980 to November 2006), International Web of Science (1945 to November 2006), LILACS (1984 to March 2005), the Chinese Biomedical Literature Database (to March 2005), and CINHAL (1982 to March 2005). We used the search terms: antithrombin, antithrombin-3, sepsis, and critically ill. Table 1 gives details of the search strategy.strategy. We limited searches to controlled or randomised controlled trials in humans. There were no language restrictions. We also hand searched the reference lists of included trials and relevant reviews for additional trials. We contacted the main authors of included trials and experts to ask for any missed, unreported, or ongoing trials. We searched for ongoing clinical trials and unpublished trials on,, and (Centre Watch Clinical Trials Listing Service) up to March 2005.

Table 1
Search strategy

Trials were included only if they were clearly randomised trials, reported in a full paper article, and provided data on mortality. The included patients were loosely defined as being critically ill—that is, sepsis, septic shock, disseminated intravascular coagulation, and other critical illnesses as proposed by the authors. There was no restriction of dose and duration of treatment with antithrombin III. We excluded trials assessing administration of antithrombin III for the reduction of cardiovascular events in invasive treatment of acute myocardial infarction.

Data abstraction

Two investigators (AA and JW) independently screened the titles and abstracts for relevant articles identified by the search strategy. They also abstracted trial data based on a predefined detailed data abstraction form. Disagreements were resolved by discussion. Authors of the included trials were approached for relevant additional information on outcome measures and risk of bias.

We identified the primary author’s name, year of publication, inclusion and exclusion criteria, number of randomised patients, type of participants (for example, trauma, septic, etc), dose and duration of antithrombin III, and duration of follow-up. We evaluated the validity and design characteristics of each trial and major potential sources of bias (such as adequate random sequence generation, adequate allocation concealment, adequate blinding, and use of intention to treat analysis). Trials were defined as low risk of bias if they fulfilled all of the above criteria.10 Our primary outcome was mortality. For this outcome we used the longest follow-up data from each trial. Secondary outcomes included length of stay in intensive care or hospital, quality of life, severity of sepsis, respiratory failure (mechanically assisted ventilation), duration of mechanical ventilation, incidence of surgical intervention, intervention effect among various populations and adverse events (such as bleeding).

Statistical analysis

We used random and fixed effect models for all meta-analyses. In the case of heterogeneity, we reported results from the random effects model. We analysed data by intention to treat and included all patients. We calculated the relative risk with 95% confidence intervals for dichotomous variables and the weighted mean difference for continuous outcomes. Heterogeneity was explored by Cochrane’s Q2 test and I2.11 I2 can be interpreted as the proportion of total variation observed between the trials attributable to differences between trials rather than sampling error (chance). I2 >75% is considered as a heterogeneous meta-analysis.

We carried out subgroup analyses in different populations (for example, trauma, newborns, etc), in patients with and without adjuvant heparin, in trials with short (less than one week) and long duration of treatment, in trials with short (less than the median follow-up) and long follow-up, and in trials with high and low risk of bias. If any results were significant, we estimated the difference between the estimates of the subgroups according to tests of interaction.12 P<0.05 indicates that the effects of treatment differ between the tested subgroups.

To assess publication bias and other types of bias we created funnel plots for mortality in which were plotted the log risk ratios against their standard errors as proposed by Egger et al.13

Estimation of risk ratio in our meta-analysis did not include trials with no events in both intervention groups. Because such trials might be important, however, we performed an exploratory analysis adding an imagined trial with one death in each study group and with a total number of participants being equal to the total number of participants in all the trials with no events.

In a single trial, interim analyses increase the risk of type I errors. To avoid this, monitoring boundaries can be applied to decide whether a single randomised trial could be terminated early because the P value was sufficiently small.14 Because there is no reason why the standards for a meta-analysis should be less rigorous than those for a single trial, analogous monitoring boundaries can be applied to meta-analyses; these are called trial sequential monitoring boundaries. Trial sequential analysis provides the necessary sample size for our meta-analysis and boundaries that determine whether the evidence in our meta-analysis is reliable and conclusive.15


Study search results

We identified 8778 references through electronic and hand searches (fig 1).1). After screening titles and abstracts, we excluded 8716 clearly irrelevant references and retrieved 62 references, all written in English, for further assessment. Twenty trials (described in 21 references with one duplicate publication) fulfilled our inclusion criteria (table 2).2). These trials were published from 1991 to 2003. We received additional data for six trials through correspondence with authors. We excluded 40 trials because they were reported as abstracts, were ongoing trials (data unavailable), had unclear trial design, were not randomised studies, or did not provide mortality data (table 3).3).

figure afsa481887.f1
Fig 1 Identification of trials for inclusion
Table 2
 Characteristics of 20 included randomised trials on use of antithrombin III in critically ill patients
Table 3
 Characteristics of excluded articles

Characteristics of included trials and assessment of risk of bias

The trials included 3458 patients, and the sample sizes varied from 25 to 2314. Patients had sepsis (n=13) or were from paediatric (n=3), obstetric (n=2), and trauma (n=2) specialties. The dose regimen of antithrombin III varied from a single bolus to 14 days of administration, aiming for above normal antithrombin III activity. Follow-up ranged from 7 to 90 days. All trials were carried out in high income countries.

Of the included trials, nine reported adequate allocation sequence generation; 12 adequate allocation concealment; 10 adequate double blinding; 14 performed intention to treat analysis, or provided sufficient data to allow us to do so, and the remaining trials excluded drop outs from the final analyses. Only eight trials reported adequate generation of allocation sequence, allocation concealment, blinding, and intention to treat analyses. These were classified as trials with low risk of bias (table 2).2).

Effects of antithrombin III

Combination of data from all 20 trials showed no significant effect of antithrombin III on mortality, with 667 (39.1%) deaths in the intervention group compared with 699 (39.9%) in the control group (pooled relative risk 0.96, 95% confidence interval 0.89 to 1.03). There was no heterogeneity between trials (I2=0%) (fig 2)2).

figure afsa481887.f2
Fig 2 Forest plot of mortality (subgroup analyses on risk of bias). Weight is relative contribution of each study to overall estimate of treatment effect on log scale assuming random effects model

Compared with no intervention or placebo, antithrombin III did not significantly influence the incidence of respiratory failure (0.93, 0.76 to 1.14), duration of mechanical ventilation (weighted mean difference 2.2 days, (−1.2 days to 5.6 days), need for surgical intervention (relative risk 1.04, 0.85 to 1.27), and length of stay in hospital (weighted mean difference −1.9 days, −11.4 days to 7.7 days) or in intensive care unit (0.0 days, −1.8 days to 1.8 days). One trial found no significant effect of antithrombin III on quality of life, and most scores used to assess severity of sepsis were also non-significant.6 Antithrombin III significantly increased the risk of bleeding events (1.52, 1.30 to 1.78, I2=0.3%) (fig 3).3).

figure afsa481887.f3
Fig 3 Forest plot for bleeding events

There was no significant effect on mortality or other outcome measures (P>0.05) in all analyses of different subgroup populations (sepsis, paediatrics, obstetrics, and trauma) showed. Additionally, subgroup analyses of trials with short and long duration of treatment, short or long follow-up, and high or low risk of bias showed no significant effect on the examined outcome measures. In the subgroup of patients who received antithrombin III without adjuvant heparin, antithrombin III was associated with a significant effect (0.87, 0.75 to 0.99) but with heterogeneity (I2=1.1, P=0.41). When we used a random effects model because of the observed heterogeneity, however, the significant effect was no longer present (0.87, 0.77 to 1.02). Antithrombin III showed no significant effect in patients with adjuvant heparin (0.99, 0.90 to 1.09) (fig 4)4)

figure afsa481887.f4
Fig 4 Forest plot of overall mortality based on heparin administration with fixed relative risk (Warren 2001w1 data split based on heparin administration)

We have also presented the forest plot for overall mortality with random (rather than fixed) relative risks because heterogeneity (I2) >0 (fig 5).5).

figure afsa481887.f5
Fig 5 Forest plot on overall mortality based on heparin administration, with random relative risk (Warren 2001w1 data split based on heparin administration)

The funnel plot analysis showed no evidence of bias (P=0.37) (fig 6).6). In three trials with 260 patients there were no deaths in either study group.w4-w6 Exploratory analysis in which we added an imagined trial with one death and 130 participants in each study group had no noticeable effect on the result.

figure afsa481887.f6
Fig 6 Funnel plot

Trial sequential analyses

To demonstrate or reject an a priori anticipated intervention effect of 10% relative risk reduction 3317 should be randomised. As more patients than that were included in our meta-analysis without the result becoming significant, a relative risk reduction of 10% or more on mortality is unlikely (fig 7)7).. To determine the required sample size we assumed a 48.5% mortality in the control group and a relative risk reduction of 5% (mortality and relative risk reduction among trials with low bias risk in our meta-analysis). Our calculations showed that we would need a sample of 14 294 (with 80% power and α 0.05) to detect a plausible treatment effect for antithrombin III on mortality, corresponding to a relative risk reduction of 5% (figs 7 and 8).8). We could, however, obtain solid evidence with fewer patients if eventually the cumulative meta-analysis’ z curve crosses the trial sequential monitoring boundary constructed for a required information size of 14 294 randomised patients. Currently 3458 patients have been randomised and no boundaries have been crossed, indicating that the cumulative evidence is inconclusive for a 5% relative risk reduction. Only 24.2% of the required number is actually available to reject a 5% relative risk reduction of mortality.

figure afsa481887.f7
Fig 7 Trial sequential analysis (cumulative meta-analysis) assessing effect of antithrombin III in critically ill patients. Cumulative z curve (red dashed line) does not cross sequential monitoring boundary constructed for sample size of 3317 patients ...
figure afsa481887.f8
Fig 8 Trial sequential analysis (cumulative meta-analysis) assessing effect of antithrombin III in critically ill patients with sample size adjusted for low bias heterogeneity. Cumulative z curve (red dashed line) does not cross trial sequential monitoring ...


Treatment of critically ill patients with antithrombin III does not significantly affect mortality and length of stay in hospital or in intensive care but is associated with a significantly increased risk of bleeding events. In patients who do not receive heparin there might be a benefit, though this should be explored in further randomised trials. In the overall mortality analysis, one low bias risk trialw1 dominates the analysis, contributing 80% of the available information. Not surprisingly the results of our meta-analysis are in accordance with the results from that trial.

Strength and limitations

We carried out a comprehensive search with no language limitations to identify randomised controlled trials. Two investigators independently abstracted data and obtained data from or confirmed data with all corresponding authors of included trials. This reduces the risk of publication, ascertainment, and information bias. We incorporated mortality data from 20 randomised controlled trials in our meta-analysis (adding 11 trials more than previously reported). We excluded six trials that reported mortality data in 285 patients because they were published only as abstracts or brief communications.w7-w12 Data from these trials showed 20% mortality in both groups, similar to that seen in the included trials (available from author) and there was no beneficial effect when including these studies in the overall mortality analysis based on bias risk (risk ratio 0.96, 95% confidence interval fixed 0.89 to 1.04).

Furthermore, we evaluated the strength of the available evidence through trial sequential analysis, a method adapted from interim monitoring boundaries in single randomised controlled trials applied to cumulative meta-analysis.15 This analysis supports our findings and shows that if antithrombin III has any effect on mortality, it is small.

The weaknesses of our conclusions are closely linked with the weaknesses in the individual trials. Only eight of the 20 included trials reported adequate randomisation, double blinding, and intention to treat analysis. These aspects may be essential to minimise the risk of bias. The impact of bias, however, seems to vary across interventions and diseases.10 We found no significant association between risk of bias and trial estimates of intervention effects in subgroup analyses. Furthermore, the funnel plot did not indicate publication bias and other bias. These findings, combined with the absence of heterogeneity observed in most meta-analyses, support the robustness of our results but do not exclude the possibility of bias.

Although there was minimal heterogeneity among trial results on mortality, we are aware that we pooled heterogeneous trials regarding patients, setting, and treatment regimen. However, all included conditions were associated with low concentrations of antithrombin III, can result in disseminated intravascular coagulation, and have similar inflammatory pathways. Therefore, we think that there is a good biological reason to perform a broad meta-analysis, which increases the generalisability and usefulness considerably. Further, a broad meta-analysis increases power, reduces the risk of erroneous conclusions, and facilitates exploratory analyses, which can generate hypotheses for future research (such as adjuvant heparin).16

Effects of antithrombin III in subgroups

The included patients were heterogeneous and the dose and duration of treatment varied. The analysed outcomes, however, had minimal heterogeneity indicating robust results. Furthermore, all subgroup analyses but one had non-significant results, analogous to the overall analyses on mortality (data available from author). Thus, antithrombin III in a particular dose and duration or in a particular group among the included types of patients does not seem to reduce mortality.

In one subgroup that did not receive adjuvant heparin, antithrombin III seemed to reduce mortality significantly (0.87, 0.75 to 0.99) with a fixed effect model. When we used a random effects model, however, as defined in our protocol, the effect was no longer significant (0.89, 0.77 to 1.02). This might or might not support the previously generated hypothesis that antithrombin III is beneficial in patients who do not receive adjuvant heparin.w1 This subgroup analysis, however, has an important weakness. We split one trialw1 into two “separate trials” with and without concomitant use of heparin. Subsequently we pooled these results in subgroups with the other trials. This procedure probably violates the randomisation, which may cast considerable doubt on the isolated significant result in the fixed effect model. If this trial was not split, the subgroup without adjuvant heparin becomes insignificant. Accordingly, antithrombin III cannot be recommended for patients without adjuvant heparin, but it may be relevant to explore this further in future trials. The negative interaction of antithrombin III and heparin has been recognised on a molecular level,7 and a randomised trial evaluating effectiveness of antithrombin III without heparin was already proposed after the KyberSept study.w13

Nevertheless, if antithrombin III is used in clinical practice adjuvant heparin should be avoided as the potentially harmful interactions are unclear.2 4

Trial sequential analysis

The results of trial sequential analysis rule out an effect of antithrombin III on mortality larger than 10% relative risk reduction because the number of randomised patients in our meta-analysis exceeds the required sample size for detecting an intervention effect of this size. Trial sequential analysis, however, also revealed absence of evidence of an intervention effect of a relative risk reduction of 5%, as suggested by the low bias risk trials. We included 3458 critically ill patients in the cumulative meta-analysis, and neither the trial sequential monitoring boundary nor the traditional z=1.96 (P<0.05) was crossed. The number of patients randomised is much smaller than the sample size of 14 294 we calculated was required, which is the number of patients needed to reliably reject an effect of antithrombin III on mortality based on the 5% relative risk reduction and a control mortality of about 50%. Our use of trial sequential monitoring boundaries applied to cumulative meta-analysis, adapted from interim monitoring boundaries of a single trial, showed that the current evidence for a clinically relevant effect of antithrombin III is still inconclusive.


In our systematic review, we have shown that antithrombin III seems ineffective in any population of critically ill patients regarding mortality and it even increases the risk of bleeding events. Its use in critically ill patients cannot be recommended based on the available evidence nor in patients without adjuvant heparin, but it may be relevant to explore this further in future trials.

What is already known on this topic

  • Antithrombin III is a costly intervention that is widely used for critically ill patients
  • Four minor meta-analyses of randomised and non-randomised trials in selected populations have previously shown inconclusive evidence on mortality

What this study adds

  • Trial sequential analysis showed that there is evidence of absence of beneficial effects (10% mortality reduction or more) of antithrombin III in critically ill patients
  • Antithrombin III increases the risk of bleeding

Supplementary Material

[extra: Web references]


We thank J Albert, C Waydhas, S Opral, T Kobayashi, and F Fourrier for providing valuable additional data.

Contributors: AA and AM conceived and designed the study. AA and JW acquired analysed and interpreted data. All authors drafted and revised the review. AA and JW wrote to authors of papers for additional information. AA is guarantor.

Funding: None.

Competing interests: None declared.

Ethical approval: Not required.

Provenance and peer review: Not commissioned; externally peer reviewed.


1. Bernard GR, Vincent JL, Laterre PF, LaRosa SP, Dhainaut JF, Lopez-Rodriguez A, et al, Recombinant human protein C worldwide evaluation in severe sepsis (PROWESS) study group. Efficacy and safety of recombinant human activated protein C for severe sepsis. N Engl J Med 2001;344:699-709. [PubMed]
2. Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med 2001;29:1303-10. [PubMed]
3. Periti P. Current treatment of sepsis and endotoxaemia. Expert Opin Pharmacother 2000;1:1203-17. [PubMed]
4. Levi M, de Jonge E, van der Poll T. New treatment strategies for disseminated intravascular coagulation based on current understanding of the pathophysiology. Ann Med 2004;36:41-9.
5. Opal SM, Kessler CM, Roemisch J, Knaub S. Antithrombin, heparin, and heparan sulfate. Crit Care Med 2002;30(5 suppl):S325-31. [PubMed]
6. Rublee D, Opal SM, Schramm W, Keinecke HO, Knaub S. Quality of life effects of antithrombin III in sepsis survivors: results from the KyberSept trial. Crit Care 2002;6:349-56. [PMC free article] [PubMed]
7. Wiedermann CJ, Römisch J. The anti-inflammatory actions of antithrombin—a review. Acta Med Austriaca 2002;29:89-92. [PubMed]
8. Freeman BD, Zehnbauer BA, Buchman TG. A meta-analysis of controlled trials of anticoagulant therapies in patients with sepsis. Shock 2003;20:5-9. [PubMed]
9. Bassler D, Millar D, Schmidt B. Antithrombin for respiratory distress syndrome in preterm infants. Cochrane Database Syst Rev 2006;(4):CD005383.
10. Gluud LL. Bias in clinical intervention research. Am J Epidemiol 2006;163:493-501. [PubMed]
11. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002;21:1539-58. [PubMed]
12. Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ 2003;326:219. [PMC free article] [PubMed]
13. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629-34. [PMC free article] [PubMed]
14. Lan KKG, DeMets DL. Discrete sequential boundaries for clinical trials. Biometrika 1983;70:659-63.
15. Wetterslev J, Thorlund K, Brok J, Gluud C. Trial sequential analyses may establish when firm evidence is reached in cumulative meta-analyses. J Clin Epidemiol doi: 10.1016/j.jclinepi.2007.03.013.
16. Gotzsche PC. Why we need a broad perspective on meta-analysis. It may be crucially important for patients. BMJ 2000;321:585-6. [PMC free article] [PubMed]

Articles from The BMJ are provided here courtesy of BMJ Publishing Group