Corticosteroids in acute traumatic brain injury: systematic review of randomised controlled trials

Inclusion criteria

We included studies in the review if they met the following criteria. Firstly, study participants had to have a clinically diagnosed acute traumatic brain injury of any severity. Secondly, the experimental intervention was corticosteroids (those steroids with predominantly glucocorticoid effects–namely, prednisolone, betamethasone, cortisone, dexamethasone, hydrocortisone, methylprednisolone, prednisone, and triamcinolone) administered in any dose by any route for any duration started within seven days of the injury. Thirdly, study participants were randomly assigned to treatment or control groups. Studies that used quasi-random methods of allocation, such as alternation, were excluded.

Identification of relevant trials

We searched Medline for 1966 to December 1995 using a combination of the March 1996 update of the optimally sensitive search strategy for trials from the Cochrane Collaboration (earlier version published as referenced9) with the MeSH headings “head injuries,” “intracranial pressure,” “brain edema,” and “brain-concussion” including all subheadings. The resulting citations were examined on screen to identify possibly relevant trials and those thus identified were retrieved in full and compared with the inclusion criteria. A search of Embase, years 1974-1996 (performed in March 1996) was done by using a similar approach to that for Medline.

We searched the Cochrane Library in August 1996 using each of the text terms “head,” “brain,” “dexamet*,” and “steroids.”10 We asked the Ottawa Stroke Trials Registry and the United Kingdom based Intensive Care National Audit and Research Centre to search their databases, which contain the results of hand searching many neurological, neurosurgical, intensive care, and emergency medicine journals. Several other journals were also hand searched for this review. All the journals from these sources are listed in an appendix, which is available on the internet at mailto:www.bmj.com.

The reference lists of all trials found were searched for additional trials. We attempted to contact all the trialists identified, asking them to identify any further published or unpublished trials. No language restrictions were used.

Data extraction and study appraisal

We each extracted the following information independently from each trial: strategy for concealment of allocation, number of randomised patients, duration of follow up, and number lost to follow up. The major outcome data sought were numbers of deaths and numbers of people disabled at the end of the study period. All but one study (see Faupel et al18) used the Glasgow outcome scale11 to assess neurological outcome; the categories for persistent vegetative state, severe disability, and moderate disability were combined into “disability” for this review. This enabled inclusion of the one trial that did not use the Glasgow outcome scale but used a similar ordinal categorisation of function. Where there was more than one steroid group in a trial (for example, low dose and high dose) those groups were combined. We also extracted data on side effects or complications when they were reported by using the authors' definitions of these.

As there is evidence that the quality of concealment of allocation particularly affects the results of studies, each of us scored this quality on the scale used by Schulz et al as shown below, assigning 1 to poorest quality and 3 to best quality12: 1=trials in which concealment was inadequate (such as alternation or reference to case record numbers or to dates of birth); 2=trials in which the authors either did not report an allocation concealment approach at all or reported an approach that did not fall into one of the other categories; and, 3=trials deemed to have taken adequate measures to conceal allocation (for instance, central randomisation; numbered or coded bottles or containers; drugs prepared by the pharmacy; serially numbered, opaque, sealed envelopes; or other description that contained elements convincing of concealment).

When the method used to conceal allocation was not clearly reported the author was contacted, if possible, for clarification. We then compared the scores allocated and resolved differences by discussion.

Statistical methods

Summary odds ratios were calculated in RevMan 3.0 software13 with the Mantel-Haenszel method. We tested for heterogeneity using a χ² test.

Summary odds ratios

Figures 1 and 2 present summary odds ratio charts for the two main outcomes for all trials in chronological order. For death the odds ratio was 0.91 (95% confidence interval 0.74 to 1.12), and when the categories of death and disability were summed for analysis it was 0.90 (0.72 to 1.11). The summary odds ratio was 0.92 (0.69 to 1.23) for infections and 1.05 (0.44 to 2.52) for gastrointestinal bleeds.

FIG 1

Summary odds ratio for death at end of study

• Download figure
• Open in new tab
• Download powerpoint

FIG 2

Summary odds ratio for death or disability at end of study

• Download figure
• Open in new tab
• Download powerpoint

Subgroup analysis–For the two main outcomes–death and death or disability–we performed a subgroup analysis by analysing the trials with only the highest quality of concealment of allocation. This was not done for the other two outcomes because of the small numbers of events involved. For death, the summary odds ratio was 1.04 (0.83 to 1.30) and for death and disability 0.97 (0.77 to 1.23).

Methodological issues

The inclusion of an Embase search identified one study not found on Medline.17 Contact with trialists enabled us to include data from two large unpublished studies20 21 but not from others.15 17 21

Numbers of events were small for infections and gastrointestinal bleeds, resulting in wide confidence intervals around the estimate of effect.

None of the tests for heterogeneity yielded significant results. When death was the outcome, however, the upper limit of the 95% confidence interval in the trial by Faupel et al18 did not overlap with the lower limit in the trial by Dearden et al.26 In Faupel's trial the outcome was assessed “at discharge,” yet overall 19% of the participants were classified as “unconscious stabilised.” The apparently short follow up period may account for the incongruous result.

Other sources of variation may include severity and pathology of the head injury, variations in corticosteroid regimens (for example, drug, dose, route), and temporal trends in the use of other interventions. The use of corticosteroids in spinal cord injury suggests that the timing of administration is important,32 so this is another possible source of variation.

When we excluded trials with less than the highest quality of concealment of allocation the differences between experimental and control groups was reduced for both death and death or disability. This is consistent with the evidence that inadequate concealment of allocation results in overestimates of the effect of treatment,12 but it could also be due to random variation. There were several trials in which the true quality of concealment was not known, which makes interpretation difficult.

Implications

Despite 25 years of randomised controlled trials of the use of corticosteroids in patients with head injury, their effects are still not clear. In this review the risk of death in those given corticosteroids was 1.8% less than in the control groups (95% confidence interval 5.7% less to 2.5% more) when we used the trials' average control death rate of 35.4% as the background rate.

The recent guidelines from the Brain Trauma Foundation on the management of severe head injury include a standard (a recommendation made with a “high degree of clinical certainty”) that “the use of glucocorticoids is not recommended.”33 These guidelines reviewed six of the randomised trials used in this systematic review and did not attempt a quantitive overview of them. Even with 14 trials, as in this systematic review, considerable uncertainty remains over the effects of corticosteroids.

Can this uncertainty be resolved? To do so would require large randomised trials. For example, a trial with 90% power to detect a 2.6% reduction in risk of death from 35.4% (the total control mortality in this review) to 32.8% at the 0.01 level of significance would require about 20 000 participants.34 Such large trials to detect effects of this size can be justified when the health problem is important and the treatment widely practicable.35 Corticosteroids for acute traumatic head injury meet these criteria. Without such a trial clinicians and patients and their families are being forced to make important decisions on the basis of inadequate evidence.