|Home | About | Journals | Submit | Contact Us | Français|
Tuberculous meningitis is a serious form of tuberculosis (TB) that affects the meninges that cover a person's brain and spinal cord. It is associated with high death rates and with disability in people who survive. Corticosteroids have been used as an adjunct to antituberculous drugs to treat people with tuberculous meningitis, but their role has been controversial.
To evaluate the effects of corticosteroids as an adjunct to antituberculous treatment on death and severe disability in people with tuberculous meningitis.
We searched the Cochrane Infectious Diseases Group Specialized Register up to the 18 March 2016; CENTRAL; MEDLINE; EMBASE; LILACS; and Current Controlled Trials. We also contacted researchers and organizations working in the field, and checked reference lists.
Randomized controlled trials that compared corticosteroid plus antituberculous treatment with antituberculous treatment alone in people with clinically diagnosed tuberculous meningitis and included death or disability as outcome measures.
We independently assessed search results and methodological quality, and extracted data from the included trials. We analysed the data using risk ratios (RR) with 95% confidence intervals (CIs) and used a fixed-effect model. We performed an intention-to-treat analysis, where we included all participants randomized to treatment in the denominator. This analysis assumes that all participants who were lost to follow-up have good outcomes. We carried out a sensitivity analysis to explore the impact of the missing data.
Nine trials that included 1337 participants (with 469 deaths) met the inclusion criteria.
At follow-up from three to 18 months, steroids reduce deaths by almost one quarter (RR 0.75, 95% CI 0.65 to 0.87; nine trials, 1337 participants, high quality evidence). Disabling neurological deficit is not common in survivors, and steroids may have little or no effect on this outcome (RR 0.92, 95% CI 0.71 to 1.20; eight trials, 1314 participants, low quality evidence). There was no difference between groups in the incidence of adverse events, which included gastrointestinal bleeding, invasive bacterial infections, hyperglycaemia, and liver dysfunction.
One trial followed up participants for five years. The effect on death was no longer apparent at this time-point (RR 0.93, 95% CI 0.78 to 1.12; one trial, 545 participants, moderate quality evidence); and there was no difference in disabling neurological deficit detected (RR 0.91, 95% CI 0.49 to 1.69; one trial, 545 participants, low quality evidence).
One trial included human immunodeficiency virus (HIV)-positive people. The stratified analysis by HIV status in this trial showed no heterogeneity, with point estimates for death (RR 0.90, 95% CI 0.67 to 1.20; one trial, 98 participants) and disability (RR 1.23, 95% CI 0.08 to 19.07; one trial, 98 participants) similar to HIV-negative participants in the same trial.
Corticosteroids reduce mortality from tuberculous meningitis, at least in the short term.
Corticosteroids may have no effect on the number of people who survive tuberculous meningitis with disabling neurological deficit, but this outcome is less common than death, and the CI for the relative effect includes possible harm. However, this small possible harm is unlikely to be quantitatively important when compared to the reduction in mortality.
The number of HIV-positive people included in the review is small, so we are not sure if the benefits in terms of reduced mortality are preserved in this group of patients.
What is tuberculous meningitis and how might corticosteroids work?
Tuberculous meningitis is a serious form of tuberculosis that affects the meninges that cover the brain and spinal cord, causing headache, coma and death. The clinical outcome is often poor even when people with tuberculous meningitis are treated with antituberculous drugs.
Corticosteroids are commonly used in addition to antituberculous drugs for treating people with the condition. These drugs help reduce inflammation of the surface of the brain and associated blood vessels, and are thought to decrease pressure inside the brain, and thus reduce the risk of death. Some clinicians are concerned that corticosteroids may improve survival, but result in more severely disabled survivors.
What the evidence shows
We examined the evidence published up to 18 March 2016 and included nine trials with 1337 people that evaluated either dexamethasone, methylprednisolone, or prednisolone given in addition to antituberculous drugs; one trial was of high quality, while the other trials had uncertainties over study quality due to incomplete reporting.
The analysis shows that corticosteroids reduce the risk of death by a quarter at two months to two years after treatment was started (high quality evidence). Corticosteroids make little or no difference to the number of people who survive TB meningitis with brain damage causing disability (low quality evidence); because this event is uncommon, even taking the most pessimistic estimate from the analysis of a slight increased risk with corticosteroids means this would not be quantitatively important when compared to the reduction in deaths.
One trial followed up participants for five years, by which time there was no difference in the effect on death between the two groups, although the reason for this change over time is unknown.
Only one trial evaluated the effects of corticosteroids in human immunodeficiency virus (HIV)-positive people but the number is small so we are not sure if the benefits in terms of fewer deaths are preserved in this group of patients.
Tuberculous meningitis is an inflammation of the meninges, which are membranes that envelope a person's brain and the spinal cord. It is caused by infection with one of several mycobacterial species that belong to the Mycobacterium tuberculosis complex, which are responsible for tuberculosis (TB) disease. Tuberculous meningitis is a severe form of TB and accounts for many deaths (Tandon 1988). It is a form of extrapulmonary TB (that is, TB that occurs outside the lungs). The World Health Organization (WHO) reported that 0.8 million of the 5.4 million new TB cases reported worldwide in 2013 were extrapulmonary cases (WHO 2014). There is an association between extrapulmonary TB and human immunodeficiency virus (HIV) infection, particularly in people with low CD4 cell counts (Naing 2013). It appears that the higher risk of TB infection in HIV-positive people means that tuberculous meningitis is also more common in this group (Berenguer 1992; Berger 1994).
People with tuberculous meningitis usually present with headache, fever, vomiting, altered conscious level, and sometimes convulsions. It is diagnosed clinically, with confirmation by microscopy and culture of cerebral spinal fluid (CSF) or a polymerase chain reaction (PCR) test. The low sensitivity of the diagnostic tests currently available presents a particular challenge for clinicians, especially when treating children and HIV-positive people. Early diagnosis and prompt treatment are the main determinants of a good outcome in people with tuberculous meningitis (Thwaites 2013).
The causes of death and disability in tuberculous meningitis are multifactorial. The main pathological mechanisms are persistent or progressive raised intracranial pressure with or without hydrocephalus, involvement of the optic nerves or optic chiasm leading to visual deficit, cranial neuropathies, arachnoiditis, and vasculitis of the cerebral blood vessels leading to stroke. Neurological disability related to antituberculous treatment may occur due to optic neuritis related to ethambutol or isoniazid, which sometimes causes permanent loss of vision, or isoniazid-related peripheral neuropathy.
Tuberculous meningitis can be classified according to its severity. The British Medical Research Council (MRC) staging system categorizes patients into three stages (MRC 1948): stage I (mild cases) for those without altered consciousness or focal neurological signs; stage II (moderately advanced cases) for those with altered consciousness who are not comatose and those with moderate neurological deficits (for example, single cranial nerve palsies, paraparesis, and hemiparesis); and stage III (severe cases) for comatose patients and those with multiple cranial nerve palsies, and hemiplegia or paraplegia, or both.
Without anti-tuberculous treatment, people with tuberculous meningitis die (Tandon 1988; Thwaites 2002). Streptomycin, one of the earliest antituberculous drugs to be introduced, reportedly reduced the case-fatality rate to 63% (Parsons 1988). Newer antituberculous drugs − isoniazid, rifampicin, pyrazinamide, and ethambutol − are associated with better survival, but mortality remains comparatively high. Reports of mortality rates vary from 20% to 32%, and permanent neurological deficits in an additional 5% to 40% of people who survive tuberculous meningitis (Ramchandran 1986; Alarcón 1990; Jacobs 1990; Jacobs 1992).
Indirect evidence from animal studies provides a biological basis for how corticosteroids could be effective (Feldman 1958). They may decrease inflammation, especially in the subarachnoid space; reduce cerebral and spinal cord oedema, and intracranial pressure (Feldman 1958; Parsons 1988); and reduce inflammation of small blood vessels, and damage due to blood flow slowing to the underlying brain tissue. However, corticosteroids could also cause harm by suppressing the person's immune system. They may suppress the symptoms of TB infection but promote an unchecked growth of the bacteria and an increased bacterial load, and reduce inflammation of the meninges, which will then reduce the ability of drugs to cross the blood-brain barrier and enter the subarachnoid space. Other adverse effects of corticosteroids include gastrointestinal haemorrhage, electrolyte imbalance, hyperglycaemia, hypertension, and increased risk of infections from other pathogens (D'Arcy-Hart 1950).
The use of adjunctive corticosteroids is not known to result in disability in tuberculous meningitis, especially when used for short periods of time as is the case in most clinical trials of this intervention. However, there is concern that although corticosteroids may save the lives of some people who have severe tuberculous meningitis, they may not necessarily improve their quality of life, as some people may survive but be left with a severe disability, rendering them bed-bound and highly dependent. In other words, if corticosteroids increase the survival rate but not disability-free survival, then corticosteroids might actually increase a person's suffering.
Several randomized controlled trials (RCTs) have been conducted on the effect of corticosteroids in managing people with tuberculous meningitis. The conclusions from these trials, seen individually, appear inconsistent. One trial, Thwaites 2004, showed that dexamethasone increases survival rate. However, it also raised two questions: do people who survive because of dexamethasone therapy tend to be left with severe disability, and are there differential effects among subgroups of people with different degrees of disease severity? The editorial that accompanied the trial, Quagliarello 2004, and several letters to the editor in response to this trial (Marras 2005; Seligman 2005) commented that the trial did not have sufficient statistical power to answer these questions. We have prepared a meta-analysis that synthesizes the results from all available RCTs to try and provide the necessary power to address these questions.
To evaluate the effects of corticosteroids as an adjunct to antituberculous treatment on death and severe disability in people with tuberculous meningitis.
Randomized controlled trials (RCTs).
People of any age with clinically diagnosed tuberculous meningitis.
Corticosteroid (hydrocortisone, prednisolone, methylprednisolone, or dexamethasone) given orally, intramuscularly, or intravenously plus antituberculous treatment.
Antituberculous treatment (same as intervention) with or without placebo.
Adverse events as reported by the authors, including upper gastrointestinal bleeding, invasive bacterial or fungal infections, and hyperglycaemia.
We attempted to identify all relevant trials regardless of language or publication status (published, unpublished, in press, and in progress).
We searched the following databases using the search terms and strategy described in Appendix 1: Cochrane Infectious Diseases Group Specialized Register (18 March 2016); Cochrane Central Register of Controlled Trials (CENTRAL), published in the Cochrane Library, up to Issue 2, February 2016; MEDLINE (1966 to 18 March 2016); EMBASE (1974 to 18 March 2016); and LILACS (1982 to 18 March 2016). We also searched Current Controlled Trials (www.controlled-trials.com; accessed 18 March 2016) using 'tuberculosis' and 'meningitis' as search terms.
We contacted the following organizations and individuals working in the field: delegates at the Vth Annual Conference of Indian Academy of Neurology, Madras, India, 1997; delegates at the XIIIth Global Joint Meeting of the International Clinical Epidemiology Network and Field Epidemiology Training Program, Victoria Falls, Zimbabwe, 1994; and members of the INDEX-TB Guidelines technical advisory committee, New Delhi, India, 2015.
For selection of studies and data extraction, we independently conducted each step, and examined agreement between the review authors. We resolved any disagreements through discussion.
We independently screened the search results and retrieved the full-text articles of all potentially relevant trials. We examined each trial report to ensure that we included multiple publications from the same trial only once. We contacted trial authors for clarification if a trial's eligibility was unclear. We resolved any disagreements through discussion and listed the excluded studies and the reasons for their exclusion.
One of the review authors, KP, conducted one of the included trials (Prasad 2006), which was started at the same time as Prasad 2000 (the first edition of this Cochrane Review). As of March 2016, this trial had not been published, but the unpublished data is included in this review. KP is also a co-author on Kumarvelu 1994. For both of these studies, HR performed the description of studies, 'Risk of bias' assessments, data extraction, and interpretation in consultation with the CIDG Co-ordinating Editor, Paul Garner.
We independently extracted data on participant characteristics, diagnostic criteria, disease severity, HIV status, antituberculous drug regimen, corticosteroid regimen, and outcome measures using a pre-piloted data extraction form. We resolved disagreements through discussion and contacted the corresponding trial author in the case of unclear or missing data. We contacted the authors of Lardizabal 1998 to determine the number of deaths in participants with stage II and III disease, and also the authors of Thwaites 2004 to determine the number of deaths in the five-year follow-up study (Török 2011).
For dichotomous outcomes, we recorded the number of participants that experienced the event and the number of participants randomized to each treatment group, and used them in the analysis. We also recorded number of participants analysed in each treatment arm, and used the discrepancy between the figures to calculate the number of participants lost to follow-up. These figures allowed us to perform a worst-case scenario analysis to investigate the effect of missing data.
We independently assessed methodological quality using the Cochrane 'Risk of bias' tool and reported the results in a 'Risk of bias' table (Higgins 2011). Regarding generation of allocation sequence and allocation concealment, we classified each of these as either adequate, inadequate, or unclear according to Jüni 2001. We reported who was blinded in each trial, and assessed the risk of bias associated with blinding separately for the two primary outcomes. If at least 90% of participants were followed up to the trial's completion we classified inclusion of all randomized participants as adequate; otherwise we classified inclusion as inadequate. We attempted to contact the trial authors if this information was not specified or if it was unclear. We resolved any disagreements by discussion between the review authors.
We used relative risk as the measure of treatment effect for analysis.
There were no cluster RCTs.
The primary analysis is an intention-to-treat analysis where all participants randomized to treatment are included in the denominator. This analysis assumes that all losses to follow-up have good outcomes. We carried out a sensitivity analysis to explore the impact of the missing data on the summary effect estimate for death.
We assessed heterogeneity by visually inspecting the forest plots to determine closeness of point estimates with each other and overlap of confidence intervals (CIs). We used the Chi² test with a P value of 0.10 to indicate statistical significance, and the I² statistic to assess heterogeneity with a value of 50% taken to indicate statistical heterogeneity. We planned to investigate heterogeneity through the following subgroup analyses: drug resistance (susceptible versus resistant M. tuberculosis); severity of illness (MRC stages I, II, and III); and HIV status (seropositive versus seronegative).
We conducted visual inspection of the funnel plot of the trials for any obvious asymmetry that could be evidence of publication bias.
We analysed the data using Review Manager (RevMan) (RevMan 2014). In view of the absence of significant heterogeneity we decided to perform a meta-analysis. We used risk ratios (RR) with 95% CIs and the fixed-effect model. We summarized the adverse event data in tables and performed meta-analysis for four types of treatment-related adverse event: gastrointestinal bleeding, hyperglycaemia/glycosuria, invasive bacterial infection (all of which could be related to corticosteroid use), and hepatitis (related to antituberculous treatment). We were unable to calculate rate ratios or summary rate ratios because the person-time over which these events were observed was unavailable.
There was no significant heterogeneity to indicate investigation of its potential sources.
To explore the possible effect of losses to follow-up on the effect estimate for the outcome death, we performed a worst case scenario analysis and compared it with an available case analysis. We assumed all participants who had dropped out of the corticosteroid group had an unfavourable outcome whereas those who had dropped out of the control group had a favourable outcome, and compared these results to an available case analysis.
We included nine trials and excluded 18 trials (Figure 1; Characteristics of included studies; Characteristics of excluded studies).
The original version of this Cochrane Review, Prasad 2000, included six trials with 595 participants (574 with follow-up, 215 deaths).
The 2008 update, Prasad 2008, added one new trial with 545 participants (535 with follow-up, 199 deaths).
We have provided a description of the included RCTs in Table 1.
The included trials were conducted in different time periods (one in the 1960s, one in the 1980s, four in the 1990s, and two between 2001 and 2007) and in different geographical regions: Thailand (Chotmongkol 1996); Egypt (Girgis 1991); India (O'Toole 1969; Kumarvelu 1994; Prasad 2006; Malhotra 2009); Philippines (Lardizabal 1998); South Africa (Schoeman 1997); and Vietnam (Thwaites 2004).
All participants were enrolled on the basis of clinical diagnosis of probable tuberculous meningitis. All included trials attempted to confirm the diagnosis by microbiological tests, but only Girgis 1991 reported the outcomes for culture-confirmed cases separately. We have described the diagnostic criteria used in each included trial in Table 2.
The trials included young children (Schoeman 1997) or adults (Kumarvelu 1994; Chotmongkol 1996; Lardizabal 1998; Thwaites 2004; Prasad 2006), or both (O'Toole 1969; Girgis 1991), and both sexes. All trials used the British Medical Research Council (MRC) system, MRC 1948, to assess baseline severity; two trials included only participants with stage II and III tuberculous meningitis (Schoeman 1997; Lardizabal 1998), while the other trials included participants with all stages of severity. Thwaites 2004 specifically reported the inclusion of HIV-positive and HIV-negative people, while Chotmongkol 1996 and Malhotra 2009 specifically reported excluding HIV-positive people.
Only Thwaites 2004 reported on drug resistance. In this trial, M. tuberculosis was cultured from the cerebrospinal fluid (CSF) or another site in 170 participants (31.2%), 85 from each group. M. tuberculosis isolates were tested for susceptibility to isoniazid, rifampicin, pyrazinamide, ethambutol, and streptomycin. Of 170 isolates, 99 (58.2%) were susceptible to all first-line drugs (51 in the placebo group and 48 in the dexamethasone group); 60 (35.3%) were resistant to streptomycin, isoniazid, or both (29 in the placebo group and 31 in the dexamethasone group); one was resistant to rifampicin alone (in the dexamethasone group); and 10 (5.9%) were resistant to at least isoniazid and rifampicin (three in the placebo group and seven in the dexamethasone group).
Six included trials used the corticosteroid dexamethasone and two trials used prednisolone (Chotmongkol 1996; Schoeman 1997). One trial, Malhotra 2009, compared both dexamethasone and methylprednisolone with placebo. We have described the dose regimens of corticosteroids used in Table 3.
Eight trials used three- or four-drug antituberculous regimens. O'Toole 1969, the earliest trial, used a two-drug regimen consisting of isoniazid and streptomycin.
Duration of antituberculous treatment varied from six months (Chotmongkol 1996; Schoeman 1997), nine months (Thwaites 2004; Prasad 2006; Malhotra 2009), 12 months (Kumarvelu 1994; Lardizabal 1998), to 24 months (Girgis 1991). In one trial, O'Toole 1969, the duration of antituberculous treatment was unclear.
Seven trials clearly described the follow-up period: two months (Lardizabal 1998); three months (Kumarvelu 1994); six months (Schoeman 1997); nine months (Thwaites 2004); 10 months (Malhotra 2009); two years (Girgis 1991); and 16 to 45 months (Chotmongkol 1996). It was unclear in O'Toole 1969 and Prasad 2006.
All nine trials reported death.
All but one trial reported on disabling neurological deficit in some way, although there was substantial variation in methods of assessment of this outcome between the trials (O'Toole 1969). We accepted the trial authors' definition of disability and, for the purpose of analysis, classified residual deficits into disabling or non-disabling (as shown in Table 4).
Five trials mentioned adverse events. The trials reported on a number of other immediate outcome measures we had not considered in this Cochrane review (see 'Characteristics of included studies' section).
We have listed the reasons for excluding 18 studies in the 'Characteristics of excluded studies' section.
See the'Characteristics of included studies' section, which includes a 'Risk of bias' table for each included trial. We have summarized the results of the 'Risk of bias' assessments across all included trials in Figure 2 and Figure 3.
Five included trials reported adequate methods of randomization using either computer generated sequences of random numbers or random number tables (Girgis 1991; Kumarvelu 1994; Thwaites 2004; Prasad 2006; Malhotra 2009). The remaining included trials did not clearly report the method of randomization.
We assessed four trials (O'Toole 1969; Chotmongkol 1996; Thwaites 2004; Prasad 2006) as having adequate allocation concealment, with participants allocated coded treatment packs. The remaining trials did not clearly describe allocation concealment.
Chotmongkol 1996 reported an imbalance in the severity of disease between the two groups, with the placebo group having a greater number of cases with Grade I disease and the steroid group having a greater number with Grade III disease. MRC stage 3 disease was present in 6/29 participants (20.7%) in the prednisolone group, but 4/30 participants (13.3%) in the placebo group. Conversely, stage 1 disease was present in 3/29 participants (10.3%) in the prednisolone group, but 6/30 participants (20%) in the placebo group. Both favoured the placebo group.
Four included trials had adequate blinding of participants and personnel (O'Toole 1969; Chotmongkol 1996; Thwaites 2004; Prasad 2006). Participants and personnel were not blinded in the remaining trials.
We evaluated the blinding of outcome assessors separately for the two primary outcome measures.
For death, we assessed all included trials as at low risk of bias, apart from Girgis 1991. We considered that all-cause death was unlikely to be affected by risk of bias relating to outcome assessment, and therefore we assessed included trials as at low risk of bias regardless of blinding of outcome assessors for this outcome. We assessed Girgis 1991 as having unclear risk of bias because this trial reported death as a case fatality rate, meaning that death was attributed specifically to tuberculous meningitis. The effect of misclassification of deaths as being due to tuberculous meningitis when they were in fact due to another cause on the overall estimate of mortality is unknown.
For disabling neurological deficit, we categorized unblinded outcome assessments as high risk, given the subjectivity of such assessments. Two trials blinded assessors of neurological disability and were assessed as low risk of bias (Schoeman 1997; Thwaites 2004); and two trials had unblinded outcome assessors and were assessed as high risk of bias (Kumarvelu 1994; Malhotra 2009).
Kumarvelu 1994 included 87.24% of the participants after six participants were lost to follow-up (4/24 in the corticosteroid group and 2/23 in the control group), and did not report on the reasons participants were lost to follow-up. We therefore assessed this trial as high risk of bias.
For two included trials we had access to a trial protocol (Thwaites 2004; Prasad 2006). We assessed Thwaites 2004 as at low risk of bias as the trial authors reported on all outcomes stated in the protocol in full. We assessed Prasad 2006 as at high risk of bias, as the definitions of the main outcomes were altered in the available (unpublished) data set, and adverse events were not reported. Lardizabal 1998; Malhotra 2009 and Schoeman 1997 reported all outcomes stated in the methods section in the results, so we assessed them as having low risk of bias. Chotmongkol 1996; Girgis 1991; Kumarvelu 1994 and O'Toole 1969 did not state the outcome measures in the results, so we assessed them as having unclear risk of reporting bias.
All included trials based the inclusion of participants on a clinical diagnosis of tuberculous meningitis, due to the limitations of microbiological tests to confirm the diagnosis. This means that the trials may have included some non-tuberculous meningitis cases. The direction of bias caused by such inclusions is not likely to favour corticosteroids.
See: Summary of findings for the main comparison Any corticosteroid compared to control for tuberculous meningitis
All nine included trials reported on death (Figure 4). The two largest trials, Girgis 1991 and Thwaites 2004, had more than 150 deaths in each, and the remaining trials were small trials with fewer deaths. Overall, the direction of effect indicated a benefit of steroids, with no statistical heterogeneity: the I² statistic was 0%.
The pooled analysis found that there were 25% fewer deaths with corticosteroids (RR 0.75, 95% CI 0.65 to 0.87; nine trials, 1337 participants, Analysis 1.1). The median death rate across trials was 41% without corticosteroids, which translates to a 10% absolute risk reduction with corticosteroids when applying this relative risk. This summary estimate of effect was deemed to be high quality evidence using the GRADE approach (see Summary of findings for the main comparison).
Eight trials reported on disabling neurological deficit (Figure 5). In both the intervention and control groups there were fewer events compared with death, and there was no difference between the two groups detected at two to 24 months follow-up (RR 0.92, 95% CI 0.71 to 1.20; eight trials, 1314 participants, Analysis 1.2). This summary estimate of effect was deemed to be low quality using the GRADE approach, because half the trials were at high risk of bias due to lack of blinding of outcome assessors and the estimate was imprecise.
Eight trials reported data from which we could derive a combined outcome incorporating death and disabling neurological deficit (Chotmongkol 1996; Girgis 1991; Kumarvelu 1994; Lardizabal 1998; Malhotra 2009; Prasad 2006; Schoeman 1997; Thwaites 2004). For this outcome, the overall estimate showed a reduction in the risk of death or disabling residual neurological deficit with corticosteroids (RR 0.80, 95% CI 0.72 to 0.89; eight trials, 1314 participants, Analysis 1.3). This effect mirrors the results of the mortality analysis which is the main contributor of events.
Only one recently published trial, Thwaites 2004, reported the long-term outcome of people with tuberculous meningitis randomized to receive either dexamethasone or placebo. The primary long-term outcome was survival during the five years follow-up, while secondary outcomes were status of disability and TB relapse. Fifty participants (9.4%) were lost to follow-up by the end of the follow-up period. The participants in the dexamethasone arm fared better on two-year survival rate (0.63 versus 0.55; risk difference 0.8, 95% CI 0.00 to 0.16; P = 0.07), but this advantage was lost at five years (0.54 versus 0.51; risk difference 0.03, 95% CI −0.06 to 0.12; P = 0.51). Analysis of hazard ratios by stage of disease at presentation suggested that benefit of dexamethasone in MRC stage I disease tended to persist longer with five-year probability of survival being 0.69 versus 0.55 (risk difference 0.14, 95% CI −0.01 to 0.29; P = 0.07). However, the test of interaction between disease severity and effect size was not statistically significant (P = 0.46 for zero to three months and P = 0.18 after three months). For disability, the follow-up study reported similar numbers with severe persistent neurological disability in both the steroid and non-steroid groups.
Of the six included trials that mentioned adverse events (O'Toole 1969; Kumarvelu 1994; Chotmongkol 1996; Schoeman 1997; Thwaites 2004; Malhotra 2009), three trials reported on incidence (O'Toole 1969; Thwaites 2004; Malhotra 2009; Figure 6). O'Toole 1969 reported four different adverse events (gastrointestinal bleeding, glycosuria, infections, and hypothermia), which occurred in both groups (Table 5). Thwaites 2004 reported on several adverse events, which were divided into "severe" and other events (Table 5). Malhotra 2009 reported incidences of hepatitis, anti-epileptic toxicity, gastrointestinal bleeding, and paradoxical tuberculoma in both groups. Schoeman 1997 had "serious side effects" as an outcome measure and reported "no serious side effects of corticosteroid therapy".
Meta-analyses examining gastrointestinal bleeding, hepatitis, hyperglycaemia, and invasive bacterial infection did not demonstrate a difference in the incidence of these events between the corticosteroid and placebo groups (Analysis 1.4). However, the meta-analysis is not sufficiently powered to detect a significant difference in adverse events between groups, so the results should be interpreted with caution.
We explored whether heterogeneity was explained within two main pre-specified subgroups.
For severity of illness, we stratified the results on death by the severity of illness (MRC stages I, II, and III) in Figure 7. The effect of corticosteroids appeared to be consistent across all stages of the disease although the analysis is relatively underpowered (stage I RR 0.50, 95% CI 0.29 to 0.85; six trials, 305 participants); stage II (RR 0.72, 95% CI 0.56 to 0.93; seven trials, 581 participants); and stage III (RR 0.69, 95% CI 0.54 to 0.88; eight trials, 651 participants, Analysis 2.1).
For HIV status, one trial specifically mentioned that 98 of the included participants were HIV-positive (Thwaites 2004). Analyses stratifying the outcomes of death and disabling neurological deficit did not detect any large differences, and so showed no apparent effect of HIV status on the effect estimates, but the analysis is underpowered (Analysis 3.1; Analysis 3.2; Figure 8).
Six trials reported on losses to follow-up (Kumarvelu 1994; Lardizabal 1998; Malhotra 2009; O'Toole 1969; Schoeman 1997; Thwaites 2004), with two trials reporting no losses to follow-up (Lardizabal 1998; O'Toole 1969). We performed a worst case scenario analysis, assuming that all participants lost to follow-up in the corticosteroid group died while those in the control group survived (Analysis 4.1). Under this extreme assumption, there was still a reduction in deaths with corticosteroids (RR 0.80, 95% CI 0.66 to 0.96), and the estimate was similar to the available case analysis (RR 0.71, 95% CI 0.59 to 0.86). Thus, losses to follow-up are unlikely to have introduced bias in favour of corticosteroids.
Six included trials date to the period when registry of clinical trials was not mandatory or routine. Protocols of the included trials were unavailable except for two trials (Prasad 2006; Thwaites 2004). For five trials where the outcomes were not clearly specified in the methods section, we assessed the risk of reporting bias as unclear. We assessed three trials as at low risk of reporting bias as all outcomes specified in the protocol or methods were reported (Schoeman 1997; Thwaites 2004; Malhotra 2009). We assessed one trial as at high risk of bias, as outcome definitions were changed in the reported data (unpublished), and adverse events were not reported (Prasad 2006). Overall, the main analysis is unlikely to have been affected by reporting bias.
We have presented a funnel plot of the included trials in Figure 9. It refers to the outcome death and values below one favour corticosteroids. There is no obvious evidence of publication bias, but the number of included trials was low.
See 'Summary of findings' table 1 (Summary of findings for the main comparison).
Nine trials met the inclusion criteria. At follow-up from 2 to 24 months, steroids reduce deaths by one quarter. Disabling neurological deficit is less common in survivors, and steroids may have little or no effect on this outcome; even taking the upper confidence limit of 20% increased risk, this is probably not quantitatively important when compared to the reduced mortality. There was no difference between groups in the incidence of adverse events, which included gastrointestinal bleeding, invasive bacterial infections, hyperglycaemia and hepatitis, although adverse events were not reported in all studies.
One trial followed up participants for five years. The effect on death and was no longer apparent at this time-point, and there was no difference in disabling neurological deficit detected.
One trial included human immunodeficiency virus (HIV)-positive people. The stratified analysis by HIV status in this trial showed no heterogeneity, with point estimates for death similar to HIV-negative participants in the same trial.
The trials included male and female children and adults, most of whom were HIV-negative. Thwaites 2004 reported that they included 98 HIV-positive participants, but they did not stratify the randomization for this subgroup; therefore the results for this subgroup should be interpreted with caution. The effect of corticosteroids was not significantly different between HIV-positive and HIV-negative participants, but the trial lacked the power to detect such a difference if one did exist due to the low number of HIV-positive participants.
Though the included trials varied in their use of bacteriological confirmation of diagnosis, there is reasonable evidence to suggest that the trial participants had tuberculous meningitis. Moreover, the intention-to-treat analysis in clinically diagnosed participants provides assurance that use of corticosteroids on the basis of clinical diagnosis does more good than harm. This is important because the decision to use corticosteroids is usually taken on a purely clinical basis when culture reports are unavailable and it is the balance of benefit and risk of such a decision that needs to be determined to set a clinical policy. The proportion of confirmed cases is mentioned only to provide confidence in the clinical diagnosis made by the investigators. Separate analysis of culture-positive cases is probably less relevant for clinical decision making.
All included trials were conducted in high TB burden settings, in specialist referral hospitals.
We used the GRADE approach to assess the quality of the evidence for the two primary outcomes at two to 24 months follow-up, and at five years follow-up (Summary of findings for the main comparison).
We graded the quality of the estimate of effect for the outcome death at two to 24 months follow-up as high. We assessed the estimate of effect as being at low risk of bias, as while there are some included trials that did not clearly report on the randomization method or allocation concealment, or both, the two largest included trials had few concerns and showed a consistent effect. The trials provided evidence of benefit for all age groups. Although only one trial reported on outcomes for people living with HIV, there was no obvious qualitative heterogeneity. We did not find any serious imprecision. We graded the estimate of effect for death at five years follow-up as moderate, and downgraded by one for indirectness as the data came from a single trial conducted in a high quality healthcare unit in a setting with high levels of endemic infectious diseases and poverty.
We assessed the quality of the estimate of effect for the outcome disabling neurological deficit as low quality. The lack of blinding of outcome assessors of disabling neurological deficit in four of the eight trials reporting this outcome led us to downgrade the quality of evidence by one for risk of bias. There was imprecision of this estimate relating to the small number of events, which led us to downgrade by one. We graded the estimate of effect for disabling neurological deficit at five years follow-up as very low quality, and downgraded by one for indirectness as the data was from a single trial (as for the outcome death, see above) and by two for imprecision as there were few events and the CI ranged from substantive harms to substantive benefits of corticosteroids.
We have attempted to limit bias in the review process. The Cochrane Infectious Diseases Group Information Specialist conducted the literature search, and it is unlikely that these searches missed any major trials; however, we cannot rule out the possibility that we missed some small unpublished trials. The funnel plot did not assist with this because there were too few included trials. To limit bias in the trial selection process and data extraction, we independently examined the search results, determined study selection, and extracted data.
Questions remain about the mechanism by which corticosteroids improve clinical outcomes, and advances in understanding of these mechanisms have led to a suggestion that some people may benefit from corticosteroids while others do not, and some may even be adversely affected by steroids (Thwaites 2013). Leukotriene A4 hydrolase (LTA4H) has been implicated in the pathogenesis of mycobacterial infection through its effect on the equilibrium between pro- and anti-inflammatory eicosanoids. Tobin et al. showed that both low- and high-LTA4H expression zebrafish morphants show increased mycobacterial bacterial burden compared with wildtype controls (Tobin 2010; Tobin 2012). Low-LTA4H expression led to increased lipoxin A4 production and dampening of the early tissue necrosis factor-alpha (TNF-α) response, and high-LTA4H morphants showed increased macrophage lysis despite early control of intracellular mycobacterial replication by TNF-α, with subsequent extracellular mycobacterial growth. Both of these states led to uncontrolled mycobacterial replication. Thus, hypersusceptibility to mycobacterial infection is associated with both inadequate and excessive inflammatory responses.
The use of dexamethasone in the zebrafish morphants rescued high-LTA4H animals but led to increased susceptibility in low-LTA4H animals (Tobin 2012). In people, the LTA4H transcription level is regulated by a polymorphism in the gene promoter at SNP rs17525495, with rs17525495 TT associated with high LTA4H protein expression, rs17525495 CC associated with low expression, and rs17525495 CT intermediate expression. Genotyping performed on 182 participants from a series of clinical studies in Vietnam demonstrated that people with the TT genotype (high LTA4H, hyperinflammatory) had the highest mortality amongst participants who did not receive dexamethasone, but the lowest in the dexamethasone group; the people with the CC genotype (low LTA4H, hypoinflammatory) had the highest mortality in the dexamethasone group (Tobin 2012). These results suggest that LTA4H genotype may have an important influence on whether or not steroids are effective in tuberculous meningitis, at least in this population.
Further investigation into the relationship between LTA4H expression in people, dexamethasone use, and outcomes in people with TB meningitis is needed to determine whether dexamethasone use is associated with harm in the subset of people with LTA4H deficiency, and whether genotyping people for LTA4H at diagnosis is useful to guide treatment with corticosteroids. Other drugs that target parts of this inflammatory pathway, such as thalidomide, adulimumab and infliximab, have been used as rescue therapy in people with severe inflammatory complications of TB meningitis, but few clinical trials have been conducted on the use of these agents, and all these potent immunosuppressive drugs have the potential to cause harm as well as benefit (Schoeman 2001; Schoeman 2004; Schoeman 2010; Jorge 2012; Lee 2012; Molton 2015).
There is high quality evidence of the benefit of corticosteroids in preventing death in people with tuberculous meningitis. This effect is probably attenuated over time, as five-year follow-up data from one trial suggests this, but there may be confounding factors leading to this observation. Corticosteroids appear to reduce mortality in people with TB meningitis, regardless of the British Medical Research Council (MRC) stage at presentation. Corticosteroids may have no effect on rates of disabling neurological deficit in people who survive TB meningitis, but the confidence interval around this estimate includes increased risk of this outcome. However, given the benefit associated with reduced risk of death, this is unlikely to be quantitatively important when considering whether or not to use corticosteroids in patients with TB meningitis. There is uncertainty about whether or not corticosteroids are beneficial for HIV-positive people with TB meningitis due to the lack of direct evidence in this group. Corticosteroids may not be associated with increased risk of adverse events, but there is uncertainty related to the limited reporting of adverse events in the included trials.
Further research is unlikely to add to certainty about the effect of corticosteroids in people with tuberculous meningitis who are HIV-negative in preventing death.
In people that are immunosuppressed, such as people living with HIV, it is unclear whether corticosteroids are of benefit. As corticosteroids could lead to greater risk of harm in these people, further research would be useful to provide clear guidance for treatment.
Another question that remains unanswered is the optimum choice of corticosteroid drug and dosing regimen. Given the fact that use of corticosteroids carries the risk of adverse events, and that many of these are dose-dependent, further research examining this question would be beneficial.
We thank Estée Török and Marcel Wolbers for providing additional data from the follow-up study of participants from Thwaites 2004, and Artemio Roxas Jr. for providing access to Lardizabal 1998. Hannah Ryan, Paul Garner, and the editorial base for the Cochrane Infectious Diseases Group are funded by the UK Department for International Development (DFID) in a grant related to evidence synthesis for the benefit of developing countries (Grant: 5242). The views expressed in this review do not necessarily reflect UK government policy. We thank the All India Institute of Medical Sciences, New Delhi, India for providing infrastructure support.
|Search set||CIDG SRa||CENTRAL||MEDLINEb||EMBASEb||LILACSb|
|3||steroids||corticosteroid*||TB||TB||1 or 2|
|4||corticosteroids||glucocorticoid*||1 or 2 or 3||1 or 2 or 3||steroid*|
|8||1 or 2||2 or 3 or 4 or 5 or 6 or 7||glucocorticoid*||glucocorticoid$||4 or 5 or 6 or 7|
|9||3 or 4 or 5 or 6 or 7||1 and 8||hydrocortisone||hydrocortisone||3 and 8|
|10||8 and 9||—||dexamethasone||dexamethasone||—|
|14||—||—||5-13/or||4 and 13||—|
|15||—||—||4 and 14||Limit 14 to human||—|
|16||—||—||Limit 15 to human||—||—|
aCochrane Infectious Diseases Group Specialized Register.
bSearch terms used in combination with the search strategy for retrieving trials developed by Cochrane (Lefebvre 2011); upper case: MeSH or EMTREE heading; lower case: free text term.
Kameshwar Prasad (KP) developed the first published version of this Cochrane Review (Prasad 2000). During the 2008 update, KP screened the search results, assessed methodological quality, extracted and analysed data, interpreted the results, and rewrote several sections of the review. MB Singh also screened the search results, assessed methodological quality, extracted data, and entered data into RevMan (RevMan 2014). For the 2015 update, Hannah Ryan (HR) re-extracted and analysed the data, revised the 'Risk of bias' assessment, constructed a 'Summary of findings' table with GRADE assessment, and revised the Background, Results, and Discussion sections.
KP is a co-author of two of the included trials (Kumarvelu 1994; Prasad 2006). HR independently conducted 'Risk of bias' assessments and data entry and interpretation with the CIDG Co-ordinating Editor, Paul Garner.
Antitubercular Agents [*therapeutic use]; Chemotherapy, Adjuvant; Dexamethasone [therapeutic use]; Glucocorticoids [*therapeutic use]; Hydrocortisone [therapeutic use]; Prednisolone [therapeutic use]; Randomized Controlled Trials as Topic; Tuberculosis, Meningeal [*drug therapy; mortality]
Adult; Child; Humans
References to studies included in this review