Our study has some limitations. Although we did our best to include all available trials and retrieve unpublished data, we cannot rule out the possibility of missing information. The latest update of studies included in the network meta-analysis was in April, 2017. We did a PubMed search in May, 2018, and found only three additional studies that met our inclusion critieria.48, 49, 50 Since we already had 133 included studies, we decided that adding these three studies would not have changed the final results materially. Additionally, some nodes in our network included only few studies. To adhere to the assumption of transitivity and reduce the risk of biased estimates (for instance, those that included enrichment designs), we had to discard many studies that were initially selected as potentially relevant (appendix pp 26–235). Most included studies compared an active drug with placebo and the number of actual head-to-head trials was quite small, so comparative efficacy between interventions was frequently based on indirect comparisons.
