Article type
Abstract
Objectives: The reliability of network meta-analyses (NMAs) results can be limited by methodological flaws. This study aimed to identify the publication and methodological characteristics of NMAs in top medical journals over the last five years.
Methods: The Web of Science was searched to identify NMAs published in the BMJ, Lancet, JAMA, NEJM and Annals of Internal Medicine from January 2019 to December 2023. Characteristics including author, country, affiliation, publication date and journal were analyzed using the bibliometrix package in R software. Two reviewers independently extracted the included studies searching databases, statistical models, theoretical frameworks, methods used for the assumptions of heterogeneity, coherence and transitivity, and sensitivity, risk of bias (RoB), and certainty of evidence assessment.
Results: 41 NMAs were published in the BMJ, Annals of Internal Medicine and Lancet, with 11 published in both 2019 and 2022. Among total 511 authors, GUYATT GH, GE L and VANDVIK PO were the top three in number of articles published, and countries and affiliations with the highest number were Canada, China and UK, and McMaster University, University of Toronto, and Sichuan University. On average, 3.5 databases were searched per study, with Medline and Embase being the most frequently. All NMAs were included in randomized controlled studies, and two also included observational studies. 22 studies using the Cochrane RoB tool to assess the RoB, and 2 with observational studies using ROBINS-I and SIGN checklist. 34 studies conducted pairwise meta-analysis, 36 using random-effects model, 22 used frequentist frameworks, and 23 used the GRADE method to assess confidence of evidence. 21 studies did not conduct transitivity assumption, and most were using I² and τ to judge heterogeneity, using design-by treatment to judge global inconsistency, and using node-splitting to judge local inconsistency. Sensitivity analysis was mostly conducted by excluding studies with high RoB or comparing results with other statistical models and theoretical frameworks.
Conclusions: Most of the NMAs published in top medical journals over the last 5 years were published in the BMJ, with strong interactions between authors, countries and affiliations. The evaluation of transitivity assumption needs improvement. The detailed results will be presented at the conference.
Methods: The Web of Science was searched to identify NMAs published in the BMJ, Lancet, JAMA, NEJM and Annals of Internal Medicine from January 2019 to December 2023. Characteristics including author, country, affiliation, publication date and journal were analyzed using the bibliometrix package in R software. Two reviewers independently extracted the included studies searching databases, statistical models, theoretical frameworks, methods used for the assumptions of heterogeneity, coherence and transitivity, and sensitivity, risk of bias (RoB), and certainty of evidence assessment.
Results: 41 NMAs were published in the BMJ, Annals of Internal Medicine and Lancet, with 11 published in both 2019 and 2022. Among total 511 authors, GUYATT GH, GE L and VANDVIK PO were the top three in number of articles published, and countries and affiliations with the highest number were Canada, China and UK, and McMaster University, University of Toronto, and Sichuan University. On average, 3.5 databases were searched per study, with Medline and Embase being the most frequently. All NMAs were included in randomized controlled studies, and two also included observational studies. 22 studies using the Cochrane RoB tool to assess the RoB, and 2 with observational studies using ROBINS-I and SIGN checklist. 34 studies conducted pairwise meta-analysis, 36 using random-effects model, 22 used frequentist frameworks, and 23 used the GRADE method to assess confidence of evidence. 21 studies did not conduct transitivity assumption, and most were using I² and τ to judge heterogeneity, using design-by treatment to judge global inconsistency, and using node-splitting to judge local inconsistency. Sensitivity analysis was mostly conducted by excluding studies with high RoB or comparing results with other statistical models and theoretical frameworks.
Conclusions: Most of the NMAs published in top medical journals over the last 5 years were published in the BMJ, with strong interactions between authors, countries and affiliations. The evaluation of transitivity assumption needs improvement. The detailed results will be presented at the conference.