Corticosteroids for severe sepsis and septic shock: a systematic review and meta-analysisBMJ 2004; 329 doi: https://doi.org/10.1136/bmj.38181.482222.55 (Published 26 August 2004) Cite this as: BMJ 2004;329:480
- Djillali Annane, professor of intensive care medicine ()1,
- Eric Bellissant, professor of clinical pharmacology2,
- Pierre Edouard Bollaert, professor of intensive care medicine3,
- Josef Briegel, professor of anaesthesiology and intensive care medicine4,
- Didier Keh, attending physician5,
- Yizhak Kupfer, attending physician6
- 1 Critical Care Department, Université de Versailles Saint-Quentin en Yvelines, Assistance Publique-Hôpitaux de Paris, Hôpital Raymond Poincaré, Garches 92380, France
- 2 Clinical Investigation Centre, INSERM 0203, Université de Rennes 1, Centre Hospitalier Universitaire, Hôpital de Pontchaillou, Rennes 35033, France
- 3 Critical Care Department, Université de Nancy 1, Centre Hospitalier Universitaire, Hôpital Central, Nancy 54000, France
- 4 Institut für Anaesthesiology, Klinikum der Universität München, Munich, 81366, Germany
- 5 Intensive Care Unit, Charité-Campus Virchow Clinic, Berlin, 13353, Germany
- 6 Division of Pulmonary and Critical Care Medicine, Maimonides Medical Centre, Brooklyn, New York, NY 11219, USA
- Correspondence to: D Annane
- Accepted 16 June 2004
Objective To assess the effects of corticosteroids on mortality in patients with severe sepsis and septic shock.
Data sources Randomised and quasi-randomised trials of corticosteroids versus placebo (or supportive treatment alone) retrieved from the Cochrane infectious diseases group's trials register, the Cochrane central register of controlled trials, Medline, Embase, and LILACS.
Review method Two pairs of reviewers agreed on eligibility of trials. One reviewer entered data on to the computer and four reviewers checked them. We obtained some missing data from authors of trials and assessed methodological quality of trials.
Results 16/23 trials (n = 2063) were selected. Corticosteroids did not change 28 day mortality (15 trials, n = 2022; relative risk 0.92, 95% confidence interval 0.75 to 1.14) or hospital mortality (13 trials, n = 1418; 0.89, 0.71 to 1.11). There was significant heterogeneity. Subgroup analysis on long courses (≥ 5 days) with low dose (≤ 300 mg hydrocortisone or equivalent) corticosteroids showed no more heterogeneity. The relative risk for mortality was 0.80 at 28 days (five trials, n = 465; 0.67 to 0.95) and 0.83 at hospital discharge (five trials, n = 465, 0.71 to 0.97). Use of corticosteroids reduced mortality in intensive care units (four trials, n = 425, 0.83, 0.70 to 0.97), increased shock reversal at 7 days (four trials, n = 425; 1.60, 1.27 to 2.03) and 28 days (four trials, n = 425, 1.26, 1.04 to 1.52) without inducing side effects.
Conclusions For all trials, regardless of duration of treatment and dose, use of corticosteroids did not significantly affect mortality. With long courses of low doses of corticosteroids, however, mortality at 28 days and hospital morality was reduced.
Each year severe sepsis occurs in about three people per 1000 population and accounts for 2% of hospital stays.1 About 3% of such patients will develop septic shock,2 which itself accounts for 10% of stays in intensive care units.3 Overall, hospital mortality is 30% for severe sepsis and 50-60% for septic shock.1–3
Researchers have explored the biological mechanisms of septic shock for potential interventions. Corticosteroids have been tested because of their interactions with immune responses.4 Indeed, these hormones affect inflammation through their effects on white blood cells, cytokines, and nitric oxide production. However, cytokines may suppress the cortisol response to the adrenocorticotropin hormone, causing poor adrenal activity,5 and body tissues may become resistant to corticosteroids.6 The prevalence of adrenal insufficiency in septic shock is about 50%. For these reasons, it has been anticipated that corticosteroids could be beneficial in septic shock.
Initial studies with corticosteroids in sepsis and septic shock used short courses of high doses. They did not show any evidence of benefit, as shown by two meta-analyses of the randomised trials published during the period 1966-93.7 8 However, these reviews did not exclude a benefit of longer durations of treatment (≥ 5 days) and lower doses (≤ 300 mg hydrocortisone or equivalent a day), as observed in more recent trials.9–14 We systematically reanalysed the effects of corticosteroids in severe sepsis and septic shock, considering all currently available data.
Studies and participants
We searched for randomised or quasi-randomised trials, with or without blinding, on severe sepsis and septic shock in children or adults.15 We included data from trials in sepsis, sepsis syndrome, or acute respiratory distress syndrome if separate data were available for septic shock.
We considered all studies reporting on intravenous treatment with any corticosteroid preparation (for example, cortisone, hydrocortisone, methylprednisolone, betamethasone, or dexamethasone). We defined length of treatment at full doses as long (≥ 5 days) or short (< 5 days) and classified daily doses of corticosteroids as low (≤ 300 mg of hydrocortisone or equivalent) or high (> 300 mg).
The control group received a standard treatment (that is, antibiotics, fluid replacement, inotropes or vasopressors, mechanical ventilation, renal replacement therapy), given either alone or with a placebo.
The primary outcome measure was all cause mortality at 28 days. Secondary outcome measures included mortality in the intensive care unit and in hospital, number of patients with reversal of shock (that is, stable haemodynamic status for at least 24 hours after patients are weaned from vasopressors) at 7 and 28 days, and number of patients with adverse events (for instance, gastroduodenal bleeding, superinfections, hyperglycaemia, and other adverse effects).
Search strategy for identification of studies
We attempted to identify all relevant studies regardless of language or publication status (published, unpublished, either in press or in progress). We searched the Cochrane infectious diseases group's trials register for relevant trials up to August 2003 using the search terms sepsis and septic shock as described in the Cochrane Library (issue 3, 2003). We searched the Cochrane central register of controlled trials (Cochrane Library, issue 3, 2003) using the search terms sepsis, septic shock, steroids, and corticosteroids; Medline (1966 to August 2003) using the search terms sepsis, septic shock, steroids, corticosteroids, adrenal cortex hormones, and glucocorticoids; Embase (1974 to August 2003) using the search terms sepsis, septic shock, steroids, and corticosteroids; and LILACS (to August 2003) using the search terms sepsis, steroids, and corticosteroids.16 We also checked the reference lists of resulting trials and, when possible, contacted authors to identify any additional published or unpublished data.
One reviewer (DA) checked all identified titles and abstracts, and three reviewers (PEB, JB, and DK) validated this check. Two pairs of reviewers (DA/PEB and JB/DK) examined all potential trials, selected eligible trials, and graded their methodological quality. Any disagreement within or between pairs was resolved by discussion within the four reviewers. We contacted authors for clarification when necessary.
Assessment of methodological quality
We documented the methodological quality of trials using a previously published score,8 and we graded generation of allocation sequence and allocation concealment as adequate, unclear, or inadequate.17 Methods for blinding were considered as double blind (neither patients nor care providers or assessors knew which treatment was given), single blind (either patients or care providers or assessors were aware of treatment), and open (all parties were aware of treatment). Loss to follow up was described as adequate (analysis included ≥ 90% of patients), unclear (not reported), and inadequate (analysis included < 90% of patients). Any disagreement within or between pairs was resolved by discussion within the four reviewers. We contacted authors for clarification when necessary.
One reviewer (DA) drew up a standard data extraction form and the other reviewers (PEB, JB, DK, and YK) validated it. Four reviewers (DA, PEB, JB, and DK) then independently extracted data and contacted authors of trials for missing data when possible. One reviewer (DA) entered data on to the computer, and four reviewers (PEB, JB, DK, and YK) checked them.
For each outcome measure, we computed 2×2 tables summarising, in each treatment group, the number of patients with the outcome and the total number of patients, and we organised the data so that a relative risk < 1.0 favoured corticosteroids (except for shock reversal at days 7 and 28, for which > 1.0 favoured corticosteroids). We performed intention to treat analyses. All statistical calculations used Review Manager 4.2. We calculated a weighted treatment effect (using fixed effects model) across trials. The results were expressed as relative risks with 95% confidence intervals. We considered using random effects model only in case of heterogeneity (that is, P ≤ 0.10 for χ2 test for heterogeneity).
Potential sources of heterogeneity were identified by sensitivity analyses on the basis of high quality trials and by subgroup analysis on the basis of long courses of low dose corticosteroids. This analysis allowed us to evaluate the strategy based on the high prevalence of adrenal insufficiency in septic shock and tested in trials performed after 1992.9–14 We sought evidence of publication bias using the funnel plot method.16
Description of studies
We identified 23 trials on corticosteroids in severe sepsis or septic shock.9–14 18–34 Of these, we excluded seven (table 1) 18 21 22 25 26 29 34 and included 16 trials (n = 2063) (table 2). For six trials (n = 524) we extracted data from both published and unpublished souces.9–12 14 28 For one trial, contact with authors did not provide any additional information.32 For the nine other trials the primary investigators could not be contacted.13 19 20 23 24 27 30 31 33 Tables 3 and 4 give details of the studies included.
All cause mortality at 28 days
We extracted data for all cause mortality at 28 days from 15 trials (n = 2022) (fig 1). There were 351/1033 (34%) deaths in the treated group compared with 329/989 (33%) in the control group. There was significant heterogeneity in the results (χ2 = 33.09, P = 0.003). The relative risk of dying at 28 days was 0.92 (95% confidence interval 0.75 to 1.14, P = 0.46; random effects model).
The subgroup analysis on five trials (n = 465) with long courses of low dose corticosteroids no longer showed heterogeneity across the trials, and the all cause mortality at 28 days was lower (0.80, 0.67 to 0.95, P = 0.01). In contrast, the subgroup analysis on eight trials (n = 1115) with short courses of high dose corticosteroids did not show any difference (0.97, 0.72 to 1.31, P = 0.84; random effects model). Subgroup analyses based on high quality trials had a relative risk near 1.0 and failed to explain heterogeneity (data not shown).
Mortality in intensive care unit
We extracted data for mortality in intensive care units from four trials (n = 425), all of which investigated the effects of long courses of low dose corticosteroids (fig 2). There were 108/216 (50%) deaths in the intensive care unit in the treated group compared with 127/209 (61%) in the control group (0.83, 0.70 to 0.97, P = 0.02).
Mortality in hospital
We extracted data for hospital mortality from 13 trials (n = 1418) (fig 3). There were 279/730 (38%) hospital deaths in the treated group compared with 271/688 (39%) in the control group. There was significant heterogeneity in the results (χ2 = 27.68, P = 0.006). The relative risk of dying in hospital was 0.89 (0.71 to 1.11, P = 0.30; random effects model).
The subgroup analysis on five trials (n = 465) with long courses of low dose corticosteroids no longer showed heterogeneity across the trials and showed reduced mortality in hospital (0.83, 0.71 to 0.97, P = 0.02). In contrast, the subgroup analysis on six trials (n = 511) with short courses of high dose corticosteroids did not show any difference in hospital mortality (0.89, 0.57 to 1.37, P = 0.59; random effects model). Subgroup analyses based on high quality trials had a relative risk near 1.0 and failed to explain heterogeneity (data not shown).
Shock reversal at day 7
We extracted data for shock reversal at day 7 from six trials (n = 728) (fig 4). There were 218/389 (56%) shock reversals at day 7 in the treated group compared with 154/339 (45%) in the control group (1.43, 1.01 to 2.01, P = 0.04; random effects model). There was significant heterogeneity in the results (χ2 = 20.38, P = 0.001).
The subgroup analysis on four trials (n = 425) with long courses of low dose corticosteroids no longer showed heterogeneity across the trials, and showed increased rate of shock reversals at 7 days (108/216 (50%) v 65/209 (31%); 1.60, 1.27 to 2.03, P < 0.0001).
Shock reversal at day 28
We extracted data for shock reversal at day 28 from four trials (n = 425) (fig 4). There were 117/216 (54%) shock reversals at day 28 in the treated group compared with 90/209 (43%) in the control group (1.26, 1.04 to 1.52, P = 0.02).
There was no evidence that corticosteroids increased the risk of gastroduodenal bleeding (10 trials, n = 1321; 1.16, 0.82 to 1.65, P = 0.40), superinfections (12 trials, n = 1705; 0.93, 0.73 to 1.18, P = 0.54), or hyperglycaemia (6 trials, n = 608; 1.22, 0.84 to 1.78, P = 0.30) (fig 5). Only one trial reported the definition for hyperglycaemia,32 the others reporting only the number of patients with hyperglycaemia. Another trial reported a significant rise in serum sodium concentration (> 155 mmol/l) in 6/20 (30%) patients in the treated group and in 1/20 (5%) patients in the placebo group.10
When we considered all the trials included in this systematic review, regardless of duration of treatment and dose, we found no evidence of a beneficial effect of corticosteroids on all cause mortality at 28 days and mortality in hospital from severe sepsis and septic shock. However, for both outcomes, the results showed strong heterogeneity that was not explained by the quality of the trials. For both outcomes, sorting the trials by year of publication showed that before 1992 almost all trials showed a relative risk of dying > 1.0, whereas after 1992 all trials had a relative risk of dying < 1.0. This date coincides with the consensus definition for sepsis15 and with the observation that septic shock is often complicated by adrenal insufficiency.35 The trials conducted before 1992 probably included patients with a heterogeneous risk of death, while the trials designed after 1992 focused on a more homogeneous population of patients with septic shock. More recent trials also used long courses of low dose corticosteroids, with the aim of treating adrenal insufficiency5 35 or because of cortisol tissue resistance.6 The preferred drug was hydrocortisone, with doses of 200-300 mg used to reproduce the cortisol concentrations achieved at maximum exercise in healthy people.9 10 14 Treatment lasted about a week, corresponding roughly to the mean time that patients with septic shock take vasopressors. Sensitivity analyses of these trials showed significant reduction in all cause mortality at 28 days and mortality in hospital. Long courses of low dose corticosteroids also reduced mortality in intensive care units.
Improvement in survival with corticosteroids may result from reduced duration of shock (as shown by the higher proportions of shock reversal at days 7 and 28), severity of inflammation,14 and number of organ dysfunctions.10 14 The benefits we have shown are in line with findings from studies on animals, isolated vascular smooth muscles, and inflammatory cells and on healthy volunteers challenged with endotoxin.4 Indeed, studies consistently showed that corticosteroids improved vessels' contractility and haemodynamics; prevented inflammatory cells' recruitment, proliferation, and release of pro-inflammatory mediators; and improved survival from all types of animal models of sepsis.4 Finally, there was no evidence of increased rates of gastroduodenal bleeding, superinfections, or hyperglycaemia associated with the use of corticosteroids.
There were differences between the six trials conducted after 1992. One trial included both severe sepsis and septic shock,13 whereas the others included only septic shock. One trial allowed concomitant therapies with anti-thrombin III or intravenous polyclonal immunoglobulins,10 and the others did not. The time on shock before randomisation was also different: one trial included only early septic shock,12 two included late septic shock,9 11 and two included both early and late septic shock.10 14
Strengths of study
Because of our comprehensive search strategy, omission of important trials seems unlikely. We included 16 trials, but the outcomes foreseen for this review were not available in one crossover trial.14 This trial showed short term improved haemodynamic and immune outcomes with a low dose of hydrocortisone. We considered it acceptable in a meta-analysis to pool the results from the 15 remaining trials. We converted outcome measures corresponding to censored data into dichotomous variables—that is, proportion of patients with an event after one and four weeks or in the intensive care unit or at hospital discharge.
What is already known
Short courses of high dose corticosteroids do not affect mortality from severe sepsis and septic shock
Long courses of low dose corticosteroids improve systemic haemodynamics and reduce the time on vasopressor treatment
What this paper adds
Long courses of low dose corticosteroids reduce mortality at 28 days, in intensive care units, and in hospital
Long courses of low dose corticosteroids do not significantly alter the risk of gastroduodenal bleeding, superinfections, or hyperglycaemia
We cannot provide definite recommendations for the selection of patients who might most benefit from corticosteroid. Separate data for adrenal insufficiency were available in only two studies.9 12 However, different definitions for adrenal insufficiency were used. In the first trial, too few patients had adrenal insufficiency to draw any conclusion.9 In the second trial, a benefit from corticosteroids was shown only in patients with a cortisol increase after adrenocorticotropin hormone ≤ 248 nmol/l.12 The weight of this trial in the meta-analysis was about 70%. Until there is further research on optimising diagnostic testing of adrenal insufficiency in patients with septic shock, corticosteroids should be given only to patients with a random cortisol concentration ≤ 414 nmol/l (that is, absolute adrenal insufficiency) or a cortisol response to adrenocorticotropin hormone ≤ 248 nmol/l (that is, relative adrenal insufficiency).36
In conclusion, hydrocortisone (or equivalent) should be given to patients with septic shock immediately after they undergo an adrenocorticotropin hormone test, at a dose of 200-300 mg, and should be continued for 5-11 days, only when absolute or relative adrenal insufficiency is present.
A longer version of this review has been published in the Cochrane Library.37
Contributors All authors were involved in study concept, design, and critical revision for important intellectual content. DA, PEB, JB, DK, and YK were responsible for acquisition of data. DA, PEB, JB, and DK analysed and interpreted the data. DA and EB drafted the review. DA is guarantor.
Funding Hôpital Raymond Poincaré, Garches, France. UK Department for International Development.
Competing interests Authors of this review have been involved in randomised controlled trials of low dose hydrocortisone that are included in this review.
Ethical approval Not required.