|
Affiliation
|
Department of Medicine, University of Auckland, Private Bag 92 019, Auckland 1142, New Zealand. |
|
Abstract
|
Context: Overlapping meta-analyses on the same topic are common and often report discordant results. An Endocrine Society (ES) meta-analysis reported that vitamin D with calcium reduced the risk of falls, whereas vitamin D monotherapy had no effect. Despite meta-analysing an overlapping set of trials, we concluded that vitamin D with or without calcium had no effect on falls.
OBJECTIVE: To determine the reasons for the different conclusions from the 2 meta-analyses.
METHODS: We extracted raw data from the 25 randomised controlled trials included in the ES meta-analysis, calculated the treatment effect for each trial, compared them to the published ES meta-analysis, and determined the reason for any differences.
RESULTS: Of the 25 trials, there were differences in 14 results (56%) between the 2 meta-analyses. In the ES meta-analysis, data were used from a subset of falls or participants (4 trials), from trial completers (3 trials), or were imputed from fracture data (1 trial). Other differences resulted from use of adjusted results from original papers (2 trials), differences in incorporating data from multi-arm and factorial studies (3 trials), and inconsistent group numbers reported in original papers (2 trials). In a re-analysis of the ES meta-analysis using unadjusted analyses and all randomised participants, there was a marginal effect of vitamin D with or without calcium on falls (RR 0.95, 95%CI 0.90-1.00), with subgroup analysis showing no effect of vitamin D monotherapy (RR 1.00, 0.92-1.08), or vitamin D with calcium versus placebo/controls (RR 0.95, 0.88-1.02), but a modest effect of vitamin D with calcium versus calcium (RR 0.84, 0.76-0.92).
CONCLUSIONS: Methodological differences in utilising data from the same trials directly led to substantially different conclusions between meta-analyses about the efficacy of vitamin D supplements on falls. More detailed reporting of meta-analyses is necessary to allow readers to understand discordant results from overlapping meta-analyses.
Language: en |