Article type
Year
Abstract
Background: The risk of including duplicate or redundant publications in meta-analyses and systematic reviews is gaining increasing recognition. However, the impact of duplicate publications on estimates of outcomes in meta-analysis remains to be explored. Objective: To determine the impact of including duplicate publications on estimates of clinical outcomes in a meta-analysis of off-pump versus on-pump coronary artery bypass surgery.
Methods: While performing a meta-analysis of randomized controlled trials of off-pump versus on-pump coronary artery bypass surgery, all trials that met inclusion criteria for the meta-analysis were explored for the possibility of containing previously published data. Original investigators were contacted to confirm suspected duplicates. Trials were categorized as non-covert duplicates when the publication declared the duplication and/or cited the original publication. Trials were categorized as covert when there was no citation of the original publication. After confirmation of duplicate publications, meta-analysis was performed to estimate clinical outcomes with and without duplicate publications to explore the impact of redundant data. Sensitivity analysis was performed by definition of duplicate publication. For each of the following clinical outcomes, odds ratios and 95% confidence intervals were calculated using the fixed effect or random effects model, as appropriate based on absence or presence of heterogeneity: mortality (primary endpoint), stroke, myocardial infarction, renal failure, and transfusions at up to 30 days. For each analysis, heterogeneity was estimated using the chi-square test for heterogeneity and the I-squared.
Results: Of 44 randomized trials that met the inclusion criteria, a total of 15 (34%) were subsequently identified as being duplicate publications. Of these, 10 were covert and 5 were non-covert duplications (23% and 11% of the total number of trials, respectively). Each duplicate publication had at least one original author from the original publication. For mortality, the estimates were similar with duplicates included [OR 0.85, 95%CI 0.46-1.57] and without duplicates [OR 0.86, 95%CI 0.48-1.54]. Similarly, for stroke, myocardial infarction, atrial fibrillation and transfusions, including the duplicate data did not materially affect the resulting estimates of OR and 95%CI. The p-value for heterogeneity and the I-squared remained similar with and without duplicates. In sensitivity analyses of covert duplications only, the results were similar as for when both categories of duplications were included.
Conclusion: Multiple publications of randomized trials represented over 30% of all included trials for this meta-analysis. Nevertheless, effect estimates and heterogeneity were not materially affected by the inclusion of duplicate data, and no conclusions would have changed with or without inadvertent inclusion of redundant data. Nevertheless, caution is warranted in generalizing these results to other meta-analyses. Conclusion: While duplicate publications give the illusion of increased numbers of relevant trials, inadvertent inclusion of redundant data does not invariably impact estimates of outcomes and conclusions. Further exploration of the impact of redundant data on meta-analyses is warranted.
Methods: While performing a meta-analysis of randomized controlled trials of off-pump versus on-pump coronary artery bypass surgery, all trials that met inclusion criteria for the meta-analysis were explored for the possibility of containing previously published data. Original investigators were contacted to confirm suspected duplicates. Trials were categorized as non-covert duplicates when the publication declared the duplication and/or cited the original publication. Trials were categorized as covert when there was no citation of the original publication. After confirmation of duplicate publications, meta-analysis was performed to estimate clinical outcomes with and without duplicate publications to explore the impact of redundant data. Sensitivity analysis was performed by definition of duplicate publication. For each of the following clinical outcomes, odds ratios and 95% confidence intervals were calculated using the fixed effect or random effects model, as appropriate based on absence or presence of heterogeneity: mortality (primary endpoint), stroke, myocardial infarction, renal failure, and transfusions at up to 30 days. For each analysis, heterogeneity was estimated using the chi-square test for heterogeneity and the I-squared.
Results: Of 44 randomized trials that met the inclusion criteria, a total of 15 (34%) were subsequently identified as being duplicate publications. Of these, 10 were covert and 5 were non-covert duplications (23% and 11% of the total number of trials, respectively). Each duplicate publication had at least one original author from the original publication. For mortality, the estimates were similar with duplicates included [OR 0.85, 95%CI 0.46-1.57] and without duplicates [OR 0.86, 95%CI 0.48-1.54]. Similarly, for stroke, myocardial infarction, atrial fibrillation and transfusions, including the duplicate data did not materially affect the resulting estimates of OR and 95%CI. The p-value for heterogeneity and the I-squared remained similar with and without duplicates. In sensitivity analyses of covert duplications only, the results were similar as for when both categories of duplications were included.
Conclusion: Multiple publications of randomized trials represented over 30% of all included trials for this meta-analysis. Nevertheless, effect estimates and heterogeneity were not materially affected by the inclusion of duplicate data, and no conclusions would have changed with or without inadvertent inclusion of redundant data. Nevertheless, caution is warranted in generalizing these results to other meta-analyses. Conclusion: While duplicate publications give the illusion of increased numbers of relevant trials, inadvertent inclusion of redundant data does not invariably impact estimates of outcomes and conclusions. Further exploration of the impact of redundant data on meta-analyses is warranted.