Other studies compared FDA and published data but they did not cover all competing drugs for a specific condition and did not allow for performing NMA. Our study adds 3 Publications Using Abomle LY294002 important pieces of new information. First, our analysis concerned NMAs. An extensive literature has shown the existence and impact of reporting bias in conventional metaanalysis, including the very study of Turner et al. However, this issue remains poorly explored in the indirect-comparison or NMA framework. In particular, most existing NMAs fail to address or even discuss potential reporting bias. In this case study, we showed that NMA led to highly misleading estimates of the efficacy of competing interventions in the presence of reporting bias. With evidence of reporting bias in any conventional pair-wise meta-analyses in the network, the results of NMA should be interpreted with great caution. The recognition of this issue is even more important considering the lack of a recognized method to identify and deal with reporting bias in the NMA framework. Funnel plots and tests for asymmetry could be applied to each pair-wise comparison in the network. However, the number of trials addressing each pair-wise comparison may often be limited, which would prevent this approach from documenting or excluding reporting bias appropriately. Each of our 12 comparisons between drugs and placebo were represented by no more than 10 trial publications, so Abmole VX-809 applying asymmetry tests would be inappropriate or not meaningful. Moreover, even with full knowledge of the existence of unpublished FDA-registered trials, the visual assessment of funnel plots did not reveal any asymmetry. Plots with reporting bias had approximately symmetric appearance. In some contexts, one could assume that reporting biases affect the different drugs similarly and assume exchangeability of the trial selection processes across drugs; methods that ����borrow strength���� from all trials in the network could be applied, as was performed recently for the case study we considered. As well, in specific situations, a strong publication bias is probably not necessary to influence the results. For instance, reporting bias affecting venlafaxine trials related to only 1 trial with unpublished results among 6 trials; when hypothetical reporting bias affected venlafaxine only, venlafaxine ranked first. Second, we also showed that reporting bias operates differently in NMA and in usual meta-analysis. The major difference is that in usual meta-analysis, reporting bias affects only the results of the drug of interest. In contrast, in NMA, reporting bias affecting one of a number of drugs could affect the ranking of all drugs.