Article type
Year
Abstract
Background: Νetwork meta-analysis (NMA) has become an increasingly popular methodology and several Cochrane Reviews comparing multiple interventions have been published over the last years. Despite the rapid methodological development of NMA, the good practice requirements of this advanced evidence synthesis tool are often ignored or improperly applied.
Objectives: To update the current empirical evidence on characteristics of published NMAs, including their methodological rigor and quality of reporting, and to evaluate empirically the extent of heterogeneity and inconsistency in networks of interventions.
Methods: We searched MEDLINE, the Cochrane Library and EMBASE to identify NMAs published in 2013. We extracted data on various network characteristics and assessed how often researchers evaluated and reported the assumptions underlying NMA appropriately. We documented the methods employed to estimate and present relative effects and treatment ranking and assessed the extent to which authors discussed the potential for bias in their findings. We estimated three indices that quantify the percentage of variability attributed to heterogeneity, inconsistency, or both, for each primary outcome.
Results: Our search identified 72 networks that met all inclusion criteria. So far we have extracted data from 40 networks: only 25% discussed how plausible the assumption of NMA is. Inconsistency was assessed statistically in less than half of the networks (43%), while issues of bias (e.g. publication bias or attrition bias) were discussed in about one in six networks. The I-squared statistic for heterogeneity or inconsistency ranged between 25% and 75% in almost half of the networks. Results from the 72 networks will be presented at the Colloquium.
Conclusions: Assumptions and potential limitations of NMA are not widely discussed in publications and this casts doubts about the credibility of their conclusions. The majority of networks in our database presented a moderate level of heterogeneity and inconsistency.
Objectives: To update the current empirical evidence on characteristics of published NMAs, including their methodological rigor and quality of reporting, and to evaluate empirically the extent of heterogeneity and inconsistency in networks of interventions.
Methods: We searched MEDLINE, the Cochrane Library and EMBASE to identify NMAs published in 2013. We extracted data on various network characteristics and assessed how often researchers evaluated and reported the assumptions underlying NMA appropriately. We documented the methods employed to estimate and present relative effects and treatment ranking and assessed the extent to which authors discussed the potential for bias in their findings. We estimated three indices that quantify the percentage of variability attributed to heterogeneity, inconsistency, or both, for each primary outcome.
Results: Our search identified 72 networks that met all inclusion criteria. So far we have extracted data from 40 networks: only 25% discussed how plausible the assumption of NMA is. Inconsistency was assessed statistically in less than half of the networks (43%), while issues of bias (e.g. publication bias or attrition bias) were discussed in about one in six networks. The I-squared statistic for heterogeneity or inconsistency ranged between 25% and 75% in almost half of the networks. Results from the 72 networks will be presented at the Colloquium.
Conclusions: Assumptions and potential limitations of NMA are not widely discussed in publications and this casts doubts about the credibility of their conclusions. The majority of networks in our database presented a moderate level of heterogeneity and inconsistency.