Sputum monitoring during tuberculosis treatment for predicting outcome

David Horne and colleagues’ systematic review and meta-analysis, of sputum monitoring during tuberculosis treatment for predicting outcomes, deals with an important topic. However, we disagree with some methodological aspects, which might have introduced important biases. We believe, on the basis of relevant references on systematic reviews and meta-analysis, that the search strategy used was not the most appropriate; it led to the high number of studies selected in the initial search (n=12 369) and makes reproducibility of the search questionable. Moreover, a one-by-one selection of relevant trials among 12 369 titles or abstracts is much more prone to mistakes. In our opinion, the strategy could have been more specifi c (including more terms or restrictions), without loss of search sensitivity. Another concern is about the fi gure used to show the study selection process. From the description of the included studies, how 15 papers resulted in 28 studies was unclear; the study selection information was not clearly explained during the development of the paper either, which allows for diff erent potential interpretations by readers. It was also unclear as to how the authors dealt with non-English papers. Initially exclusion criteria included a language restriction (only English), but in the second search round no language restrictions were used. The results of this second look, without language restrictions, were not mentioned. Considering that the studied disease largely aff ects underdeveloped non-Englishspeaking countries, relevant papers were probably not included in the meta-analysis. Finally, there was no information about whether the investigators attempted to contact the authors of studies to complete missing data, to avoid the exclusion of these studies. How unpublished studies were searched for (grey literature search) and handled were not mentioned either. Since the results of unpublished studies can systematically diff er from published and widely available studies, we believe this search is very important to prevent selection (publication) bias and ensure the systematic review’s quality.