Adjunctive dexamethasone in bacterial meningitis: a meta-analysis of individual patient data

1The yearly incidence of bacterial meningitis is estimated to be 2·6–6·0 cases per 100 000 in Europe and might be ten times higher in less developed countries.,,, Experimental models have shown that outcome is related to the severity of the inflammatory process in the subarachnoid space, and treatment with corticosteroids results in a reduction of the inflammatory response and improved outcome.,, These findings have prompted several randomised controlled trials of corticosteroids for bacterial meningitis. Initial results suggested that the main beneficial effect of the corticosteroid dexamethasone was to reduce the risk of hearing loss in children with Haemophilus influenzae type b meningitis. Additional data extended the likely benefit to those with Streptococcus pneumoniae meningitis. In 2004, a meta-analysis of five randomised controlled trials showed that treatment with corticosteroids reduced both mortality and neurological sequelae in adults with bacterial meningitis, without detectable adverse effects. Subsequently, a Cochrane meta-analysis of data from 20 randomised controlled trials and involving 2750 people showed an overall mortality benefit and a reduction in neurological sequelae in patients treated with adjuvant corticosteroids. However, three large randomised controlled trials published after this analysis showed conflicting results.,, Adjunctive corticosteroids seem to benefit some patients with bacterial meningitis but not others, and how to select patients who are likely to benefit is unclear. Our aim was to address this question with a meta-analysis of data from five major trials for which individual patient data were available.

2Relevant trials were identified previously as part of a Cochrane review (figure 1). Individual patient data from five randomised, double-blind, placebo-controlled trials of dexamethasone for bacterial meningitis published since 2001 were included in the analysis;,,,, individual patient data could not be acquired from the older trials.,,,,,,,,,, The characteristics of the included studies are shown in table 1.

3The study from South America used a 2×2 design to randomly assign children with bacterial meningitis to dexamethasone plus glycerol, dexamethasone plus placebo, glycerol plus placebo, or placebo plus placebo. Data were available from children who were assigned dexamethasone plus placebo or placebo only but not from those who were given glycerol. During the study, the randomisation schedule was altered from a ratio of two dexamethasone per three placebo (randomisation schedule 1) to one dexamethasone per one placebo (randomisation schedule 2). Therefore, analyses from this study were stratified according to randomisation schedule. The study in Malawian adults used a 2×2 design to randomly assign patients to dexamethasone or placebo and to intravenous or intramuscular ceftriaxone. In all studies, patients were enrolled on the basis of clinically suspected bacterial meningitis and CSF criteria. All the studies used computer-generated randomisation to allocate patients to dexamethasone or placebo. Treatment concealment was adequate in all studies.

4The members of the study group met in October, 2006, and September, 2007, to discuss data sharing and the analysis plan, including the definitions of subgroups, which were specified before the data were collated, the final database created, and the analysis started. The principal investigators provided the raw data, which were checked by a statistician (PQT). Inconsistencies and outlying data were clarified with the principal investigators and resolved from their raw data before the analysis.

515 data fields for each patient were selected for the analyses. The dataset included prognostic factors for unfavourable outcome and potential modifiers of the treatment effect of dexamethasone, such as antibiotic treatment before admission, HIV infection, and malnutrition., Definitions were agreed during the two study-group meetings. Values for continuous variables were reassigned into categories. Exposure to antibiotics before randomisation was defined by administration of effective oral or intravenous antibiotics within 48 h before the first dose of study drug was received. Malnutrition was defined by individual investigators: patients who were not assessed were categorised according to the local prevalence of malnutrition. HIV tests were not done on every patient and an assessment was made of the likelihood of HIV infection based on local epidemiology. All untested Malawian adults were defined as likely to be HIV positive. No assumption was made for untested Malawian children. All other untested adults or children were defined as likely to be HIV negative. Impairment of consciousness was categorised by use of the Glasgow coma scale or the Blantyre coma score (table 2). The causative pathogen was defined by CSF microscopy, CSF or blood culture, PCR, or latex agglutination.

6The predetermined outcome measures were death at the time of first follow-up; death or severe neurological sequelae (including severe bilateral hearing loss) at 1 month follow-up; death or any neurological sequelae (including any degree of hearing loss) at first follow-up; and death or severe bilateral hearing loss at first follow-up. The number of studies that contributed to each outcome is shown by degrees of freedom (df=number of studies minus 1). Additionally, as part of a post-hoc exploratory analysis and to analyse every possible endpoint of interest, we analysed hearing loss of any degree among survivors. The severity of neurological sequelae in the adult studies was defined using the Glasgow outcome score or the modified Rankin scale., In the paediatric studies, severe neurological disability was defined as blindness, quadraparesis, hydrocephalus requiring a shunt, or severe psychomotor retardation. Hearing loss was categorised as moderate or severe according to definitions used in the individual studies.

7All analyses were stratified according to study site (including two strata from the South American study) to account for any possible centre effect, including differences in mortality between centres. If appropriate, analyses were also stratified according to the baseline variable of interest. Combined odds ratios (ORs) and tests for heterogeneity were calculated using conventional Mantel-Haenszel statistics. We also used exploratory analyses with logistic regression. The main purpose of the analysis was to establish whether dexamethasone had a differential effect in different subgroups of patients; hence, heterogeneity between the subgroups (I2 values) with significance levels were calculated for each subgroup analysis. Tests for heterogeneity were calculated without allowing for multiple comparisons, to increase the sensitivity of detecting any evidence of between-subgroup heterogeneity. To maximise the power of finding significant heterogeneity, missing values were removed, except where indicated, from the subgroup analyses. A continuity correction was made for zero events. Significance tests, with the appropriate degrees of freedom, were calculated to test for possible heterogeneity between studies for each subgroup analysis.

8To calculate the combined ORs for death from studies included in the Cochrane reviews but not otherwise included in the present study, results available from the published literature were combined by use of conventional Mantel-Haenszel statistics. Calculation of combined ORs and 95% CIs, tests of heterogeneity between studies, and logistic regression analyses were done by use of STATA version 10.

9The study sponsors had no role in the study design, collection, analysis, and interpretation of the data, or the decision to submit the manuscript for publication. T E Peto had full access to all data in the study. All authors approved and were responsible for submission of the manuscript.

10The baseline characteristics were similar in placebo and dexamethasone groups within the five studies (table 2). 1019 (50·2%) patients received dexamethasone and 1010 (49·8%) patients received placebo. 833 (41·1%) patients were less than 15 years old, of whom 415 received dexamethasone and 418 received placebo. 1196 adults (aged ≥15 years) were included, of whom 604 (50·5%) received dexamethasone and 592 (49·5%) received placebo. The ages of five patients were unknown.

11HIV co-infection was confirmed in 549 (41·5%) of 1322 patients tested, of whom 391 (71·4%) were adults and 158 (28·8%) were children. An HIV test was not done in 707 (34·8%) patients but, on the basis of epidemiological risk, was judged likely to be positive in 31 untested adults from Malawi and negative in adults from Europe and children from South America. No assumption was made about 139 untested children from Malawi. In total, 286 confirmed or likely HIV-infected patients received dexamethasone and 294 received placebo.

12The diagnosis of bacterial meningitis was microbiologically confirmed in 1639 (80·8%) patients and was most frequently caused by S pneumoniae (759 cases), H influenzae (297 cases), and Neisseria meningitidis (239 cases). The most common causative bacteria per study were as follows: Europe, N meningitis (38%); Malawi (children), S pneumoniae (40%); Vietnam, Streptococcus suis (32%); Malawi (adults), S pneumoniae (59%); and South America, H influenzae (47%). Mortality in the placebo groups differed substantially between studies: 15% in Europe, 31% in Malawian children, 12% in Vietnam, 53% in Malawian adults, and 16% in South America.

13Dexamethasone was not associated with a significant reduction in death, death or severe neurological sequelae (including severe bilateral hearing loss), death or any neurological sequelae (including any hearing loss), or death or severe bilateral hearing loss, if all patients were included in the analysis (table 3). However, hearing loss (of any severity) in survivors was less common in the dexamethasone group (162 [24·1%] of 672 vs 195 [29·5%] of 660; OR 0·77 [95% CI 0·60–0·99], p=0·04).

14The subgroup analyses for all outcome measures are shown in Figure 2, Figure 3, and the webappendix. Duration of symptoms before treatment, severity of coma at start of treatment, whether dexamethasone was given before or after antibiotics, and HIV infection status did not significantly influence treatment response. Dexamethasone was more effective in patients aged older than 55 years in analyses of death (OR 0·41 [95% CI 0·20–0·84], p=0·01), death or severe neurological sequelae (OR 0·53 [0·30–0·84], p=0·03), and death or any neurological sequelae (OR 0·56 [0·31–1·00], p=0·05). However, there was no clear evidence of heterogeneity between the different age groups (death, χ2=6·9, 3 df, p=0·07, I2 54·5%; death or severe neurological sequelae, χ2=6·6, 3 df, p=0·09, I2=53·4%; death or any neurological sequelae, χ2=4·4, 3 df, p=0·23, I2=30·3%). Further exploratory analyses, using age as a continuous variable, did not show any consistent interaction between age and a treatment effect (data not shown). There was also no effect in a post-hoc analysis that restricted the study to patients treated with ceftriaxone (webappendix).

15The data were explored to identify evidence of heterogeneity between the studies. 23 subgroups were explored, each with five different endpoints. In patients with moderate CNS impairment on admission, there was some evidence of heterogeneity between three of the five endpoints. In the subgroup of patients with moderate CNS impairment on admission, there was evidence of benefit in death or severe neurological sequelae or bilateral hearing loss in the European study (OR 0·19 [95% CI 0·04–0·82], p=0·01), but also evidence of harm in the study of children in Malawi (OR 3·70 [1·36–10·08], p=0·006). However, no evidence of heterogeneity was observed in patients with either no or little CNS impairment or with severe CNS impairment. Overall, there was no evidence of any difference in outcome for any of the CNS subgroups in any of the five endpoints. The effect of HIV was explored by adjustment with logistic regression analysis and also by studying only patients with proven HIV status. However, HIV status did not have an effect on dexamethasone treatment outcome (webappendix). We further explored the relation between age, HIV status, and dexamethasone treatment effect (table 4). In HIV-negative adults, dexamethasone was associated with a reduction in death or severe neurological sequelae, including severe bilateral hearing loss (OR 0·68 [95% CI 0·48–0·95], p=0·02), death or any neurological sequelae, including any hearing loss (OR 0·67 [0·50–0·91], p=0·01), and death or severe bilateral hearing loss (OR 0·61 [0·42–0·89], p=0·01). However, this effect of dexamethasone was not present in HIV-negative children, or in HIV-positive children and adults.

16Gastrointestinal bleeding was reported in all studies: 13 (1·3%) of 1021 patients on dexamethasone and 19 (1·9%) of 1014 patients on placebo (p=0·14). Hyperglycaemia and infection by herpes simplex virus and varicella zoster virus were reported in some but not all studies.,, Hyperglycaemia was recorded by the trials in Malawian and European adults and was significantly associated with dexamethasone treatment (79 of 390 [20·3%] on dexamethasone vs 60 of 376 [16·0%] on placebo; p=0·02). Neither infection with herpes simplex virus (labial infection in all) nor infection with varicella zoster virus were significantly associated with dexamethasone treatment.

17Dexamethasone did not significantly affect mortality in a combined analysis with the data from other studies included in the Cochrane analysis (OR 0·88 [95% CI 0·73–1·04], p=0·14; figure 4).,,,,,,,,,,,,,,,, 349 (18·0%) of 1944 patients who received dexamethasone died, compared with 384 (19·8%) of 1939 patients who received placebo. There was no evidence of significant heterogeneity between the trials.

18The aim of this analysis was to establish whether any subgroups of patients with acute bacterial meningitis might benefit from adjunctive dexamethasone and thereby explain any differences between individual trial results. Extensive exploration of 15 prespecified subgroups did not show robust evidence that a particular subgroup would benefit. The apparent benefit in adults aged over 55 years might have occurred by chance. However, it is unclear whether it is more likely to have occurred by chance than the findings of no benefit in other subgroups.

19This analysis of 2029 patients from five trials showed that treatment with adjunctive dexamethasone did not significantly reduce mortality, neurological disability, or severe hearing loss in bacterial meningitis. Combination of these results with those from older published trials, for which the raw data were not obtainable, did not show any evidence that dexamethasone was significantly effective in reducing these outcomes overall. However, a post-hoc analysis on the incidence of deafness among survivors suggested that adjunctive dexamethasone treatment reduced the rate of hearing loss (OR 0·77 [95% CI 0·60–0·99; p=0·04), irrespective of whether patients had received antibiotics before dexamethasone treatment. The use of adjunctive dexamethasone treatment was not associated with an increased risk of adverse events.

20Factors previously considered relevant to the decision to start dexamethasone treatment in patients with suspected or proven bacterial meningitis could not explain differences in results between the five trials. These factors include duration of symptoms before treatment, severity of impaired consciousness at start of treatment, whether dexamethasone was given before or after antibiotics, and HIV infection status.,,,, Because the results of the prespecified analysis failed to show any significant heterogeneity, extensive post-hoc analyses were done with the inclusion of an additional deafness endpoint. Such analyses are usually considered unreliable, particularly if no statistical allowance is made for multiple comparisons, because of the high chance of a false-positive result. However, the extra analyses were undertaken to allow the identification of subgroups of interest for further possible study. These exploratory post-hoc analyses suggested a possible overall effect on deafness among survivors and on death and severe neurological sequelae in the subgroup of HIV-negative adults (OR 0·68 [95% CI 0·54–0·99], p=0·02). This apparent treatment effect ceased to be significant after adjustment for multiple comparisons.

21This meta-analysis is, as are all meta-analyses, limited by the possibility that more heterogeneity exists between the studies than has been identified. If such heterogeneity were to exist, combining the studies would be inappropriate. Formal tests for heterogeneity between studies and between subgroups failed to show any convincing evidence of heterogeneity. However, such tests are insensitive and could miss important effects. We have therefore explored the data exhaustively for relevant subgroups of patients that could reveal possible causes of heterogeneity, although little such evidence was found.

22On the basis of previous meta-analyses,, the administration of dexamethasone to children with H influenzae type b meningitis before the start of antibiotic therapy is thought to reduce the incidence of deafness. However, we found no evidence of a benefit of adjunctive dexamethasone in all children or in any subgroup of children with this infection.

23In summary, these data indicate that patients with bacterial meningitis neither benefit from nor are harmed by treatment with adjunctive dexamethasone. Despite an individual patient data meta-analysis of more than 2000 patients, we have been unable to determine conclusively whether a subgroup of patients might benefit. To establish with certainty whether dexamethasone has a place in the treatment of adult patients with bacterial meningitis, a large multinational randomised controlled trial would be necessary. This represents a formidable challenge and one that is not likely to be met for many years. In the meantime, we suggest the benefit of adjunctive dexamethasone for all or any subgroup of patients with bacterial meningitis remains unproven and there is little support for its routine use in the treatment of this disease.