Keywords: network meta-analysis, selective reporting, evidence based medicine, outcome reporting bias, publication bias, meta-analysis
Evidence-based medicine aims to optimize healthcare decision-making by leveraging results from well-conducted research, and often relies on evidence synthesis in systematic reviews and meta-analyses. As such, recommendations can be biased if reported results are a selective sample of what has been collected and known to the trialists. Chan et al (2004) showed evidence of selective outcome reporting in 102 randomized trials, with 50% of efficacy outcomes and 65% of safety outcomes being incompletely reported, and statistically significant outcomes having higher odds of being reported compared with non-significant outcomes. Unfortunately, there are very few methods that properly quantify and adjust for selective reporting of outcomes. Motivated by this important methodological gap, we developed a novel EM algorithm for bias correction and an intuitive measure to quantify evidence of selective reporting in network meta-analysis. We validate the method through simulation studies and investigate the evidence of selective reporting in a published network meta-analysis comparing 12 labor induction methods, where we find that maternal and natal outcomes have different reporting patterns.