There are an estimated 1.5 to 2.7 million deaths related to malaria each year (WHO 1997), and a large proportion of these deaths are preceded by cerebral malaria. About 10 to 30% of people with cerebral malaria die even when treated with effective antimalarial drugs (WHO 1990). Five to ten per cent of survivors have long term neurological disability, including epilepsy, paralysis, impaired hearing, cortical blindness, speech disorders, and learning disabilities (Warrell 1982, Brewster 1990, Bondi 1992, Meremikwu 1997).
People with cerebral malaria become unconscious (cannot be aroused even by pain) for several hours, and a high proportion have convulsions (Waller 1995, Waruiru 1996). The clinical manifestation of cerebral malaria is thought to occur because the majority of parasitized red blood cells occupy small blood vessels in the brain, which affects perfusion and metabolism in surrounding tissues (Pongparatn 1991, Pasvol 1995). Malaria reduces blood glucose levels in some individuals, and this can also cause convulsions and unconsciousness. Febrile convulsions can also occur in feverish children with malaria but, unlike in cerebral malaria, these are usually brief, and patients regain consciousness a few minutes after the convulsions (Molyneux 1995).
Convulsions due to cerebral malaria tend to be protracted, and can recur several times during the same illness. Observational studies have noted that cerebral malaria patients who have protracted or repeated convulsions have worse outcomes (Molyneux 1989, Jaffar 1997). Some researchers have also shown that people with cerebral malaria may have convulsions that are not obvious ('subtle convulsions'). These patients were shown to have abnormal electrical activity in the brain similar to those observed in convulsing patients, and may suffer from similar adverse outcomes (Molyneux 1995, Crawley 2001). Other research has shown that African children dying from cerebral malaria had deepening unconsciousness and convulsions, and raised intracranial pressure, probably resulting from brain oedema (Newton 1991, Newton 1994). It may be that convulsions in cerebral malaria contribute to more deaths by making brain anoxia (lack of oxygen) and oedema (swelling) worse, and raising the intracranial pressure.
The World Health Organization (WHO) recommends that people with cerebral malaria should be treated with antimalarial drug injections. Where necessary, they should also be treated for associated problems such as severe anaemia, using blood transfusion; low blood glucose, using glucose infusion; and convulsions, using anticonvulsant drugs (WHO 2000a).
In recent years, doctors have considered ways of further reducing the possible adverse consequences of convulsions in cerebral malaria, and have suggested that anticonvulsant drugs should be routinely given to all people with cerebral malaria whether or not they are convulsing at the time of admission (White 1995). This recommendation is based on the reasonable assumption that convulsions may themselves contribute to brain damage or increase the risk of death. Giving anticonvulsants to all cerebral malaria patients has the advantage of controlling unrecognized (as well as obvious) convulsions, and preventing further convulsions that cannot be reliably predicted. If, however, convulsions do not themselves damage the brain or make the prognosis worse, but are merely marker of a more severely affected brain, then preventing convulsions with routine anticonvulsant drugs may not improve the person's chance of surviving or escaping brain damage.
Anticonvulsant drugs are known to cause some adverse effects. For instance, phenobarbitone, phenytoin sodium, and diazepam may cause respiratory depression (Parfitt 1999). This may even lead to fatal respiratory failure in unconscious patients, especially where there are inadequate facilities for respiratory support. Although these three drugs are recommended for treating people who are convulsing (Rylance 1990, Winstanley 1992, Parfitt 1999, WHO 2000a), giving them routinely to all people with cerebral malaria could not be justified if they offer no additional benefits.
The uncertainty surrounding the routine use of anticonvulsant drugs in cerebral malaria has led us to prepare this systematic review, in which we consider whether all people with cerebral malaria should be given anticonvulsant drugs, whether or not they show visible signs of convulsing.
To evaluate the effect of routine anticonvulsant drugs in people with cerebral malaria.
Criteria for considering studies for this review
Types of studies
Randomized and quasi-randomized controlled trials.
Types of participants
People with cerebral malaria who fulfil all the following criteria:
- coma defined as non-localizing response to pain corresponding to either: (i) Blantyre coma score less than 4 or (ii) Glasgow coma score less than 11;
- presence of asexual Plasmodium falciparum in peripheral blood;
- normal cerebrospinal fluid.
Blantyre coma scale (Molyneux 1989) was modified from the widely used Glasgow coma scale (Teasdale 1974) to become more applicable to young children (WHO 2000a, WHO 2000b). The Blantyre coma scale has a range of 0 to 5 scores, while the Glasgow coma scale ranges from 3 to 14.
Types of interventions
Intervention: Anticonvulsant drugs started immediately after diagnosis and randomization.
Control: No routine anticonvulsant drugs, with or without placebo (anticonvulsants only given to treat convulsions according to clinical indications).
Types of outcome measures
- Death within 6 months post-randomization.
- Proportion of people experiencing convulsions within 4 weeks of randomization.
- Proportion of people still in a coma by day 7 post-randomization.
- Deficits in cognitive function (measured with standard psychometric test).
- Long term epilepsy (proportion still experiencing seizures by one year post-randomization or taking antiepileptic drugs after recovery from cerebral malaria).
- Proportion of people with a physical handicap at 12 months (measured with standard physical functioning scale)
Search methods for identification of studies
We attempted to identify all relevant studies regardless of language or publication status (published, unpublished, in press, and in progress).
We used the following topic search terms for all searching trial registers and databases: anticonvulsive agent; anticonvulsant; antiepileptic; seizure; seizure, epilepsy and convulsion; convulsion; malaria; brain malaria; and cerebral malaria. The full search strategy is available in Appendix 1.
We searched the Cochrane Infectious Diseases Group (CIDG) Specialized Register up to May 2004. Full details of the CIDG methods and the journals hand searched are published in The Cochrane Library in the section on Collaborative Review Groups.
We searched The Cochrane Central Register of Controlled Trials (CENTRAL), published in The Cochrane Library (Issue 2, 2004). This contains mainly reference information to randomized controlled trials and controlled clinical trials in health care.
We also searched the following electronic databases using search terms in combination with the search strategy for identifying trials developed by The Cochrane Collaboration and detailed in the Cochrane Reviewers' Handbook (Clarke 2000):MEDLINE (1966 to May 2004); EMBASE (1988 to May 2004); LILACS (1982 to May 2004); Science Citation Index (May 2004); and African Index Medicus (1999).
We contacted the Medical Research Council, The Gambia; Kenya Medical Research Institute; and the Wellcome Tropical Research Groups Thailand/Vietnam/Oxford for information on published, unpublished, or ongoing trials.
External referees checked the search strategy and helped to identify additional unpublished, ongoing, and planned trials. We also checked existing related reviews, and the citations of all the trials identified by the above methods. Where necessary, we contacted authors for additional information and clarification.
Data collection and analysis
We independently applied the inclusion criteria to all identified trial reports. From those trials eligible for inclusion, we independently extracted data on the methods, types of participants, interventions, and outcomes. We compared the two sets of extracted data and discussed them to ensure accuracy and completeness. We planned to seek the opinion of an editor in the Cochrane Infectious Diseases Group if we disagreed on any points. Where necessary, we requested additional data from the trial authors. Only one author (Crawley 2000) responded, providing additional information on generation of allocation sequence, and some outcome measures.
We made comparisons between the groups receiving anticonvulsant drug(s) and the control groups. While our aim was to do intention-to-treat analysis, two of the trials (White 1988, Kochar 1997) failed to clearly specify how many people were lost to follow up. We therefore did not have enough data to apply intention-to-treat analysis. We assessed heterogeneity by visual examination of forest plot and Chi-square test for heterogeneity. Where we decided to proceed with meta-analysis despite heterogeneity, we used both a fixed and random effect model. We performed a sensitivity analysis of the trials with adequate allocation concealment (White 1988, Crawley 2000) with exclusion of the inadequately concealed trial (Kochar 1997). We calculated Risk Ratio with 95% confidence interval for all the outcomes since all were dichotomous data.
Description of studies
A total of 573 participants were recruited into the 3 trials: 340 children in the Kenyan study (Crawley 2000), 185 adults in the Indian study (Kochar 1997), and 48 (mostly adults) in the Thai study (White 1988). More details are given in the 'Characteristics of included studies'.
In all three trials the intervention group received a single dose of parenteral phenobarbitone. Doses varied from 3.5 mg/kg (maximum 200 g; White 1988) to 10 mg/kg (maximum 400 mg; Kochar 1997) and 20 mg/kg body weight (Crawley 2000).
The phenobarbitone was given intramuscularly to all the participants in two trials (White 1988, Kochar 1997). In the third, the first 23 (6.8%) participants received intravenous phenobarbitone while the rest received intramuscular phenobarbitone (Crawley 2000). Two trials compared phenobarbitone with placebo (Crawley 2000, White 1988), while the third trial had a no treatment control group (Kochar 1997).
All three trials reported deaths and number of people experiencing convulsions within four weeks of hospitalization. Crawley 2000 reported patients who experienced three or more episodes of convulsions. Only one trial reported neurological problems, including visual impairment, physical handicap, and developmental delay (Crawley 2000). None provided data on long term epilepsy.
Risk of bias in included studies
Crawley 2000 used a sequentially numbered randomization register to generate the allocation sequence. The method used to generate allocation sequence was unclear in White 1988, and suboptimal in Kochar 1997, where the participants were allocated to receive either phenobarbitone or nothing on alternate days.
Allocation was concealed using sealed envelopes in White 1988 and by using a central code in Crawley 2000; the staff and participants were blinded in both. Allocation was not concealed in Kochar 1997 and the clinicians were not blind to the allocated group.
In Crawley 2000, 32 (9.4%) participants were either excluded after randomization (11) or lost to follow up (21). The authors of the other two trials did not state categorically that no participants were lost follow up. It is possible that some losses to follow up were not reported. The authors performed intention-to-treat analysis on their primary outcome of convulsion after randomization.
Effects of interventions
We found statistically significant heterogeneity for this outcome of death within six months post-randomization (Chi-square = 9.91; p = 0.0071, Analysis 1.1). The overall fixed and random effects estimates of the Risk Ratio were 1.16 (95% confidence interval 0.85 to 1.58) and 1.30 (95% confidence interval 0.58 to 2.88) respectively. Both estimates indicate no statistically significant difference.
The scope for investigating sources of heterogeneity is limited because only three trials were included. We carried out a sensitivity analysis of the two studies with adequate allocation concealment (Crawley 2000, White 1988) and found no detectable statistically significant heterogeneity (Chi-square 0.26; p = 0.61, Analysis 1.2). The overall fixed and random effect estimates of the Risk Ratio of death of 2.00 (95% confidence interval 1.20 to 3.33) and 1.98 (95% confidence interval 1.19 to 3.28) respectively, showed that in the adequately concealed trials, the risk of death was increased if phenobarbitone was used.
We found no heterogeneity for this outcome of proportion of people experiencing convulsions within four weeks of randomization. The pooled estimate shows that participants treated with phenobarbitone were statistically significantly less likely to have convulsions within four weeks of going to hospital (Risk Ratio 0.30; 95% confidence interval 0.19 to 0.45; fixed effect model; Analysis 1.3). Crawley 2000 also reported that the placebo group was also more likely to have prolonged convulsions (23/170) than the phenobarbitone group (9/170). The other trials did not provide such detailed information on the pattern of convulsions.
None of the trials reported a standard assessment of physical handicap at 12 months follow up. However one trial (Crawley 2000), reported that the risk of neurological problems (described by authors as "permanent sequelae") up to the third month of follow up was not statistically significantly different between the phenobarbitone (9/170) and the placebo (15/170) groups (Risk Ratio 0.60; 95% confidence interval 0.27 to 1.33, Analysis 1.4).
None of the trials provided data on coma by day 7, deficits in cognitive function, or long term epilepsy. Additional information supplied by Crawley 2000 confirmed that none of the people in that trial was still in coma by day 7.
Of the three trials included in this review, two used adequate methods of allocation concealment and were double blinded (White 1988, Crawley 2000). The third trial used a quasi method of randomization, had no allocation concealment, and was not blinded (Kochar 1997). Although no losses to follow up were described in two trials (Kochar 1997, White 1988), there were no categorical statements that all the randomized participants were followed up; this left us uncertain as to whether the analyses are truly intention-to-treat. Crawley 2000 reported the loss of 5% of the randomized participants to follow up. The sample sizes of the treatment (n = 102) and control (n = 83) groups in Kochar 1997 differ remarkably, suggestive of possible bias in allocation or follow up of participants. In summary, two of the three included trials were of good methodological quality (White 1988, Crawley 2000), while the third was of poor methodological quality and prone to greater bias (Kochar 1997).
We found statistically significant heterogeneity for our primary outcome of death within six months post-randomization. This is an indication that the data sets from the respective trials differ so much that combination of the data sets in meta-analysis would give misleading results. We tried to explore the source of this heterogeneity, but found ourselves limited because only three trials were eligible for inclusion in this review. The differences in the methodological quality of trials could explain this heterogeneity, since when we analysed the two trials of good methodological quality, we found no statistically significant heterogeneity. However, given the low power of the test for heterogeneity and the small size of the trials, this does not necessarily mean that heterogeneity is absent. Pooled estimates suggest that mortality will be increased if people are treated with phenobarbitone compared to placebo. This is statistically significant for the subgroup of the two trials of good methodological quality and suggests that the risk of death doubles, with a number needed to harm of approximately 10.
Another potential source of heterogeneity is the dose of phenobarbitone. Crawley 2000 used the highest dose (20 mg/kg) and has the highest mortality rate. This is biologically plausible, as higher doses of phenobarbitone are more sedating and can cause depression of the respiratory centre. However, there was no clear trend among the three included trials that indicated mortality increases as the dose of phenobarbitone increases. Crawley 2000 also gave other anticonvulsants to participants who convulsed for longer than five minutes. Such participants first received intravenous diazepam (0.3 mg/kg/dose; maximum of two doses), and subsequently intramuscular paraldehyde (0.2 ml/kg), followed by phenytoin infusion (20 mg/kg) if convulsions persisted. A commentary on this study has raised a question on the possible effect of the additional doses of diazepam on the participants (Molyneux 2000). Kochar 1997 also gave participants additional diazepam and phenytoin to control convulsions while White 1988 used diazepam alone. It would therefore be difficult to attribute the observed heterogeneity to these additional anticonvulsant drugs alone.
The age of trial participants may have contributed to the heterogeneity. Crawley 2000 included young children while the other two trials included mainly adults. It is possible that younger children are more susceptible to the sedative effects of phenobarbitone and hence to respiratory depression and death. Extrapolating this observation to adult populations may be misleading given that the trial with the highest number of adults (Kochar 1997) was of poor methodological quality. Because only three trials were eligible for inclusion in this review, and because we only had access to aggregate level data, we were unable to explore the sources of heterogeneity further.
We found no heterogeneity for convulsions within four weeks of follow up, our second primary outcome. Phenobarbitone was associated with statistically significantly fewer convulsions during follow up. This effect is however outweighed by the possibility of a higher risk of mortality.
Crawley 2000 provided data on physical handicap; there was no statistically significant difference between the treatment and placebo groups. We found no data on the other outcomes that we had planned to assess in this review namely, coma by day 7; deficits in cognitive function; and long term epilepsy.
Implications for practice
Giving phenobarbitone routinely to people with cerebral malaria reduces the number of convulsions, but may increase the risk of death. Phenobarbitone should not be given routinely to people with cerebral malaria until there is more evidence on how it can be used without causing harm.
Implications for research
Anticonvulsant drugs have potential for benefit. Larger clinical trials using lower doses of phenobarbitone or other anticonvulsant drugs, such as phenytoin, are needed to determine whether or not routine anticonvulsant drugs are beneficial in treating cerebral malaria. It is also necessary to consider the cost implications of any choice of anticonvulsant drugs used in these trials because many people in countries where falciparum malaria is endemic may be unable to afford expensive anticonvulsant drugs. In this sense, phenobarbitone has the advantage of being relatively cheap.
We gratefully acknowledge the very helpful contributions of Professor Malcom Molyneux, Professor Kevin Marsh, and Dr Jane Crawley. The reviewers have received technical assistance from both the Cochrane Infectious Diseases Group and the Cochrane Epilepsy Group.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Index terms
Appendix 1. Search methods: detailed search strategies
Last assessed as up-to-date: 20 May 2004.
Protocol first published: Issue 2, 2000
Review first published: Issue 2, 2002
Contributions of authors
Martin Meremikwu raised the research question and wrote the first draft of the protocol. Subsequently both reviewers (Martin Meremikwu and Tony Marson) contributed to the preparation of the protocol, literature search, selection of trials, data extraction, analysis, and writing the text of the review.
Declarations of interest
We certify that we have no affiliations with or involvement in any organization or entity with a direct financial interest in the subject matter of the review (eg, employment, consultancy, stock ownership, honoraria, expert testimony).
Sources of support
- Liverpool School of Tropical Medicine, UK.
- Department for International Development, UK.
- Department for International Development, UK.
- University of Calabar, Nigeria.
Medical Subject Headings (MeSH)
MeSH check words
* Indicates the major publication for the study