MA can be valid only if findings of independently conducted research are made equally available. Third, publication bias and Abmole Dabrafenib selective Abmole GSK2636771 analysis reporting bias both affected the estimates of treatment efficacy. The impact of reporting bias in NMA is not the algebraic sum of the impact of the two biases. The substantial impact of selective analysis reporting bias is of major practical importance. In fact, contrary to publication bias, selective analysis reporting bias is less likely to be improved by trial registration only, and an additional independent statistical analysis with access to the complete raw data set, the trial protocol and the pre-specified analysis plan may be required to ensure integrity. NMAs and conventional meta-analyses both should include FDA reviews of approved drug products, which are now publicly available from the Center for Drug Evaluation and Research website, in searches for published and unpublished results. Our study might have several limitations. First, this empirical analysis relied on a particular network dedicated to a specific clinical condition, one class of drugs and one type of trial. This specificity might limit the generalizability of our findings. In this case study, reporting bias led to overestimation of effect sizes for all drugs. However, a reanalysis of meta-analyses with the addition of unpublished data from the FDA for 6 other drug classes has shown that the effect of including unpublished FDA data varies by drug and outcome, with the possibility of a decrease but also an increase in estimates of efficacy. The network of evidence from the Turner et al. dataset was limited to placebocontrolled trials, resulting in a radiating star shape. However, it corresponds to geometry of real-world networks, since simple networks of 3 different interventions without direct head-to-head trials are frequent and, in cases of three or more interventions, examples of networks with star or ladder designs have been reported frequently. Turner��s network of evidence did not include the existing head-to-head trials. Unfortunately, we do not have access to the potential unpublished results from head-to-head trials. Regulatory registration is uncommon for head-to-head trials and if industry-furnished data were available they may still suffer from selective analysis bias. Therefore, we could not perform an unbiased complete NMA. However, the addition of published and unpublished head-to-head trials in our analysis may modify the estimated impact of reporting bias. Extrapolating our results to a network with both direct and indirect evidence is not straightforward.