The review question was clear and supported by explicit inclusion criteria. The search was reasonably extensive. However, it appeared that there was reason to strongly suspect the presence of publication bias, which may have led to an overestimation of the treatment effect. The studies were appraised using a suitable quality assessment tool which identified potential for bias in almost all of the included trials. The authors reported performing this assessment in duplicate but did not report whether similar methods were used at other stages of the review process.
There were considerable differences between the interventions and populations of the included studies. These differences appeared to be reflected in the statistical heterogeneity associated with the pooled treatment effect. A large number of subgroup analyses were used to explore this heterogeneity; it was not clear if these were pre-specified or not.
The review was reasonably well-conducted. However, there are reasons to regard the overall result as subject to uncertainty. These include the high probability of publication bias, the relatively small number of patients in the included studies, the potential for bias in the designs of these studies, and the evidence of significant heterogeneity between studies.