Author Reply.

Sir,

We read with great interest the letter in response to our systematic review and meta-analysis assessing the efficacy of bedaquiline versus placebo for management of multidrug-resistant tuberculosis.[1] As per our interpretation of the letter, authors were surprised by the findings of our systematic review and meta-analysis and made some suggestions on the approach for meta-analysis. For example, authors state that a fixed model, instead of random effects model should have been used due to lack of heterogeneity. Unfortunately, the authors are not aware of the latest methods for the conduct of meta-analysis. As per the Cochrane Handbook of Systematic Reviews, the choice of model should not be determined on the basis of observed heterogeneity but rather following a prespecified protocol. That is, the choice of fixed versus random effect should be decided a priori and not post hoc. Further, in the case of the fixed-effect model the key assumption is that the true effect size is same in all studies, which simply is implausible. In addition, there is never a case when two trials are exactly the same. That is, the trials may be similar but still have slight differences in terms of setting, patient populations, caregivers, etc., Therefore, when studies from the published literature are used in a meta-analysis, the random-effects model is generally considered a more plausible match.[23] On another note, the results remain unchanged regardless of the model used.

Another concern noted was that the authors of the original studies did not observe increased mortality with bedaquiline. Factually, this assessment is based on selective quoting and requires attentive reading. The authors of the study by Diacon et al. state “the development of bedaquiline for indications for which a reasonably efficacious and safe alternative exists (e.g., treatment of drug-sensitive tuberculosis or preventive treatment) should be approached with caution until more data have been collected to clarify the implication of the excess deaths.[4]” Furthermore, we assume that authors of this letter did not understand a key fact that a synthesis is considered more reliable than individual studies and is the primary reason for synthesis.[2] In line with the findings of our systematic review, the Food and Drug Administration has issued a warning related to increased mortality associated with bedaquiline[5] which was also highlighted in a recent perspective article.[6] Therefore, the observation in the letter is simply flawed and misguided. If authors of the letter have better evidence than a synthesis, the onus of bringing it in public domain lies entirely with them.

Finally, the authors provide some post hoc power analysis without understanding the context of random error and missing a key point that “absence of evidence is evidence of absence. [7]” We acknowledge that individual studies, due to multiple outcomes and comparisons can be prone to alpha error. However, is not a synthesis less prone to alpha error than any individual study due to increased sample size? A power analysis is recommended before the beginning of a trial to avoid random error and not post hoc. All the included studies in the systematic review performed a power analysis. Therefore, the argument of a post hoc power analysis to assess the probability of false positive or false negative results is a nonstarter to begin with.

In summary, we stand by the inferences of our systematic review and meta-analysis[1] which was performed as per the highest possible rigor currently being used in the field and the conclusions were based on the data obtained from synthesis (i.e., not personal opinions).