The effects of excluding treatments from network meta-analyses: survey

摘要: Objective To examine whether the exclusion of individual treatment comparators, including placebo/no treatment, affects results network meta-analysis. Design Survey networks with trial data. Data sources PubMed and communication authors meta-analyses. Study selection methods We included that had five or more treatments, contained at least two closed loops, twice as many studies level data available. Investigators abstracted information about study design, participants, outcomes, geometry, eligible treatments. Results Among 18 involving 757 randomised controlled trials 750 possible comparisons, 11 upfront decided not to consider all comparators only 10 nodes. In 7/18 networks, there was one node whose removal caused a than 1.10-fold average relative change in estimated treatments effects, switches top three were observed 9/18 networks. Removal large changes effects (average 1.16-3.10-fold) for four originally Exclusion current uncommonly used drugs resulted substantial 1.21-fold) on systemic advanced malignancies. Conclusion Excluding meta-analyses sometimes can have important their diminish usefulness research clinicians if comparisons are missing.