How do adjuvant corticosteroids benefit patients with acute bacterial meningitis? Follow the evidence to find a clinical approach that takes the best research into account.
Case Scenario
You have just completed an assessment of an ill-looking adolescent male, who presented with a
3 day history of fever, progressive headache and altered mental status. Your physical examination and lumbar puncture confirm the presence of acute bacterial meningitis, and you initiate broad-spectrum antibiotics immediately. Your colleague approaches you and asks you if you are initiating corticosteroid treatment as well, as he remembers hearing somewhere that this additional treatment is good for preventing hearing loss and other clinical outcomes…
3 day history of fever, progressive headache and altered mental status. Your physical examination and lumbar puncture confirm the presence of acute bacterial meningitis, and you initiate broad-spectrum antibiotics immediately. Your colleague approaches you and asks you if you are initiating corticosteroid treatment as well, as he remembers hearing somewhere that this additional treatment is good for preventing hearing loss and other clinical outcomes…
Clinical question
How do adjuvant corticosteroids benefit patients with acute bacterial meningitis? What does the evidence say?
How do adjuvant corticosteroids benefit patients with acute bacterial meningitis? What does the evidence say?
The evidence
Bacterial meningitis remains a devastating illness that causes significant morbidity and mortality. Although early antibiotic treatment against causal pathogens remains the most important treatment, there is considerable interest in the potential benefits of using corticosteroids to reduce brain inflammation and sequelea thereof. Recent evidence would suggest that using adjuvant corticosteroids (CS) is particularly useful in preventing significant hearing loss, which is of considerable importance particularly in the pediatric population.
In 2007, the Cochrane Database of Systematic Reviews (CDSR) published a comprehensive review of the reported benefits of adjuvant CS in treating patients with acute bacterial meningitis. The authors reviewed 18 randomized controlled trials (RCT’s) comprising 2750 adult and pediatric patients. Their findings are summarized in Table 1. Overall, all important clinical outcomes for physicians and patients showed trends to benefits (mortality, hearing loss, long-term neurological complications), in
nearly all subgroups examined. These trends were preserved regardless of causal organism (S. Pneumoniae, H. Influenzae, N. meningitides). These benefits were less conclusive for patients especially children) from lower income countries (as defined by the United Nations Human Development Index). There was no reported increase in adverse effects associated with CS use.
nearly all subgroups examined. These trends were preserved regardless of causal organism (S. Pneumoniae, H. Influenzae, N. meningitides). These benefits were less conclusive for patients especially children) from lower income countries (as defined by the United Nations Human Development Index). There was no reported increase in adverse effects associated with CS use.
The Cochrane Collaboration remains the world’s most comprehensive and unbiased effort to synthesize clinical information for physicians to guide their practices in many medical situations. It uses comprehensive search strategies to gather all retrievable information worldwide on a given topic, rigorous screening criteria to filter out methodologically poor studies that could distort synthesized information, and reliable mathematical techniques for the volunteer expert reviewers to combine data from studies to generate a final “answer” to a given clinical question (if possible). Reviews from the Cochrane group have consistently been shown to be more conservative in their interpretation of individual study data, and the recommendations they make to guide clinical practice.
This review conclusively showed that adjuvant CS use in acute bacterial meningitis is associated with significant benefits in important clinical outcomes to both patients and clinicians. The authors recommended that all patients diagnosed with acute bacterial meningitis should receive a loading
dose of dexamethasone (0.6mg/kg) prior to or concomitantly with broad-spectrum antibiotics.
dose of dexamethasone (0.6mg/kg) prior to or concomitantly with broad-spectrum antibiotics.
Application of the evidence
A reader may struggle with the difference between “statistical” significance and “clinical” significance of reported results. “Statistical significance” is a mathematical concept that demonstrates that the difference between two interventions being compared is mathematically “true,” such that it is
unlikely that the difference is due to chance. For example, a comparison of antibiotics vs. analgesics for acute otitis media may show a statistically significant improvement of 10% (ranging from 4%–14%) favoring the use of antibiotics to prevent pain on day 1 of treatment. “Clinical significance,” however, is a subjective concept based on the importance of the outcome of interest to both clinicians and patients, and how beneficial the treatment effect really is. Again, using the otitis media example, if the 10% reported difference in day 1 control was based on 90% improvement in the antibiotics-
treated patients, and 80% improvement in the analgesics-treated patients, then this may make clinicians and patients rethink how important that 10% importance really is. In another situation, however, such as reperfusion therapy for acute ischemic stroke, a statistically significant 10% improvement in short-term death would also be clinically significant for patients and clinicians. The
point here is that mathematical significance of the presented results must be interpreted in the context of the clinical situation being dealt with, and the preferences of the treating physician and receiving patient.
unlikely that the difference is due to chance. For example, a comparison of antibiotics vs. analgesics for acute otitis media may show a statistically significant improvement of 10% (ranging from 4%–14%) favoring the use of antibiotics to prevent pain on day 1 of treatment. “Clinical significance,” however, is a subjective concept based on the importance of the outcome of interest to both clinicians and patients, and how beneficial the treatment effect really is. Again, using the otitis media example, if the 10% reported difference in day 1 control was based on 90% improvement in the antibiotics-
treated patients, and 80% improvement in the analgesics-treated patients, then this may make clinicians and patients rethink how important that 10% importance really is. In another situation, however, such as reperfusion therapy for acute ischemic stroke, a statistically significant 10% improvement in short-term death would also be clinically significant for patients and clinicians. The
point here is that mathematical significance of the presented results must be interpreted in the context of the clinical situation being dealt with, and the preferences of the treating physician and receiving patient.
The effect of an intervention is an estimate of the truth in that we can never know the true effect unless we test it on absolutely everyone with the condition. In most papers, the effect of an intervention is reported as some “measure of association,” that is a single number which represents a mathematical relationship between outcomes of the groups in the study. Examples of measures of
association include relative risk (RR), odds ratio (OR), relative risk reduction (RRR), absolute risk reduction (ARR) and numberneeded- to-treat (NNT).
association include relative risk (RR), odds ratio (OR), relative risk reduction (RRR), absolute risk reduction (ARR) and numberneeded- to-treat (NNT).
We often refer to the measure of effect as the point estimate, and, because it’s an estimate, we include a range of values (confidence interval) in which the truth is likely to lie. If it’s a 95% confidence interval then there’s a 95% probability that the truth lies within the given interval of values. If it’s a 99% confidence interval, the probability is 99% that the truth lies within the given interval of values. It’s the confidence interval, therefore, that determines the precision of the point estimate: a narrow interval is considered precise whereas a wide interval is less precise and more likely to include a value or values that are not statistically significant.
Another important concept is understanding how the confidence interval relates to the measure of association being reported. Both the relative risk (RR) and odds ratio (OR) are proportions, so if the confidence interval around the RR or OR includes the value 1.00, it means there is no difference.
Similarly, the RRR and ARR are differences, so if their confidence intervals include the value of 0.00, then again there is no difference between the two interventions being compared in a study. For a statistician, a study result is “statistically significant” shows some important difference, and the associated confidence intervals do NOT include the value of “no difference.”
Similarly, the RRR and ARR are differences, so if their confidence intervals include the value of 0.00, then again there is no difference between the two interventions being compared in a study. For a statistician, a study result is “statistically significant” shows some important difference, and the associated confidence intervals do NOT include the value of “no difference.”
In this review, the relative risks (RR) all show pretty convincing benefits supporting the use of adjuvant CS. However, for some of the outcomes of interest, the 95% confidence intervals around the RR estimate approach or include 1.00, suggesting that there may be no benefit at all. Statistical “purists” would maybe disparage these results, saying that the confidence intervals approaching the
“no–difference” value negates any potential value for the intervention being discussed. “Real-world” clinicians, however, may be more forgiving of the statistical uncertainty, in favour of the beneficial applicability to patient care. Given the devastating nature of these complications from meningitis, most clinicians would likely be willing to offer CS treatment, even if the benefits are not as large
as reported by the relative risk (RR). In this regard, judicious use of evidence is meant to complement clinical thinking and risk stratification for potential adverse sequelae for these types of patients.
“no–difference” value negates any potential value for the intervention being discussed. “Real-world” clinicians, however, may be more forgiving of the statistical uncertainty, in favour of the beneficial applicability to patient care. Given the devastating nature of these complications from meningitis, most clinicians would likely be willing to offer CS treatment, even if the benefits are not as large
as reported by the relative risk (RR). In this regard, judicious use of evidence is meant to complement clinical thinking and risk stratification for potential adverse sequelae for these types of patients.
Case summary
You discuss the steroid issue with your colleague, and quickly both agree that giving adjuvant CS is an excellent intervention for this young patient in whom you have confirmed the diagnosis of acute bacterial meningitis. As you admit the patient to the hospital, your consultant compliments you on your thorough assessment and management strategy for the benefit of this patient.
You discuss the steroid issue with your colleague, and quickly both agree that giving adjuvant CS is an excellent intervention for this young patient in whom you have confirmed the diagnosis of acute bacterial meningitis. As you admit the patient to the hospital, your consultant compliments you on your thorough assessment and management strategy for the benefit of this patient.
Next page:
Learn to apply the formulae of evidence-based medicine
{mospagebreak title=Learn to apply the formulae of evidence-based medicine}
As stated earlier, “measure of association” is a numeric representation (the “effect estimate”) of the relationship of exposures and outcomes in an interventional study. The exposures in this articles are adjuvant corticosteroids compared to standard care. The outcomes of interest are multiple, including death, long-term neurologic sequelae, hearing loss, and so on.
The relationship of exposures and outcomes can be expressed in standard 2×2 tables as shown below. We’ve chosen a fictitious test and results to illustrate how the formulae are applied.
The measures of association can be represented in a number of ways:
1) Odds Ratio – the odds of an outcome is the ratio of probability of an event occurring to the probability of the event not occurring. As such, the odds ratio (OR) is the cross product of the individual outcomes in the table above OR=AD/BC.
Example: OR = 200/600 = .33
Example: OR = 200/600 = .33
2) Relative Risk – the risk of an outcome is simply the probability of an event occurring, without comparing it to another probability. A relative risk (RR), however, does compare such probabilities. The relative risk is the incidence of the outcome in the treatment group (exposed) compared to the incidence of outcome in the control (nonexposed) group:
RR = [A/(A+B)]/[C/(C+D)] = A(C+D)/C(A+B)
Example: RR = [20/(20+30)]/[20/(20+10)] = 0.60
Example: RR = [20/(20+30)]/[20/(20+10)] = 0.60
**Note: Both the OR and RR are proportions, so if either of them equals 1.00, or the confidence interval around this effect estimate includes 1.00, then there is no difference between the two groups for the outcome being measured.
3) Relative Risk Reduction (RRR) – this term quantifies the difference between two risk estimates, usually showing how much the treatment reduced the outcome of interest relative to the baseline rate in the control group. It is usually expressed as a percentage: RRR = ([C/(C+D)] – [A/(A+B)])
/ [C/(C+D)]
/ [C/(C+D)]
Example: RRR=([20/20+10)] – (20/20+30)]/[20/(20+10)] = 0.39
4) Absolute Risk Reduction (ARR) – This term quantifies the actual difference
between the outcome rates in the two
groups. These numbers tend to be much
more conservative than the RRR above. It is
represented by the following formula: ARR
= [C/(C+D)] – [A/(A+B)].
Example: [20/(20+10)] – [20/(20+30)] = 0.27
between the outcome rates in the two
groups. These numbers tend to be much
more conservative than the RRR above. It is
represented by the following formula: ARR
= [C/(C+D)] – [A/(A+B)].
Example: [20/(20+10)] – [20/(20+30)] = 0.27
5) Number Needed to Treat (NNT) – this term expresses the ARR in the context of how many patients need to be treated to realize one beneficial outcome. Numerically, it is the inverse of the ARR: NNT = 1/ARR. As expected, the higher the ARR, the lower the NNT to generate one outcome. This is good if the outcome is desirable, but bad if the outcome is not desired (eg. NNH – number needed to harm).
**Note: Both the RRR and ARR are differences (not proportions as OR and RR), so if the value of the difference (or the confidence interval around the estimate) includes 0.00, then there is no difference
between the two groups.
between the two groups.