The importance of being specific-a meta-analysis evaluating the effect of antibiotics in acute otitis media

摘要

Background: Whether acute otitis media (AOM) should be the cause for antibiotic treatment has been a matter of debate during the last decades. Treatment guidelines are based on less than twenty trials that have found the effect of antibiotics on symptomatic outcomes in AOM, such as pain, to be very modest. Two recent trials found a more substantial effect of antibiotics when they looked at treatment failure as the outcome. That the effect varies with the chosen outcome may not only be because the true effect is different but also because different outcomes are more or less specific for the disease in question. Objective: The purpose of this study was to perform a meta-analysis to calculate a composite risk ratio for treatment failure in AOM and also to investigate whether the specificity of treatment failure as an outcome differs from that of symptomatic outcomes, such as pain. Methods: Trials evaluating the effect of antibiotics in AOM and reporting the number of treatment failures were identified and a fixed-effects meta-analysis was performed. In addition, the literature was searched for articles providing direct or indirect figures on the specificity of different outcomes in AOM trials. A hypothetical study was designed to show how differences in sensitivity/specificity of inclusion/outcome criteria affect the results of a trial. Results: The meta-analysis yielded a composite risk ratio of 0.4 (95% CI 0.35-0.48), p < 0.001 for the effect of antibiotics on treatment failure. Based on data from the literature, the specificity of treatment failure was estimated to 92-100%. The hypothetical study showed how a non-specific outcome biases the effect estimate towards the null, whereas other kinds of misclassification only decrease precision. Conclusion: Future trials should focus on improving diagnostic criteria to increase precision but primarily, they should focus on choosing a specific outcome in order not to get a biased effect estimate.