Correction notice to paper “Agreement of treatment effects for mortality from routinely collected data and subsequent randomized trials: meta-epidemiological survey”BMJ 2018; 362 doi: https://doi.org/10.1136/bmj.k3210 (Published 17 August 2018) Cite this as: BMJ 2018;362:k3210
In response to this research paper by Hemkens and colleagues (BMJ 2016;352:i493, doi:10.1136/bmj.i493), Franklin and colleagues published a paper in Epidemiologic Methods (2017;6:20160018, doi:10.1515/em-2016-0018), arguing that the inversion rule used by Hemkens and colleagues in their analysis may have positively biased the estimates of the relative odds ratio and therefore their conclusion was invalid. For this reason, Franklin and colleagues submitted a rapid response to The BMJ on 13 November 2017 requesting that the authors and The BMJ retract the paper. To address the issue, The BMJ’s editors and statisticians met on several occasions to review the original methods and their criticisms. The advice of outside experts was also sought.
After much deliberation, The BMJ’s statisticians and editors concluded that, as Franklin and colleagues argue, the strategy of reversing some randomised controlled trial (RCT) results on the basis of routinely collected data (RCD) to assess the difference between results from RCD and RCT studies is a biased approach, and there were several steps that the authors could take to address this source of bias. The BMJ asked the authors to reanalyse their data and to modify their conclusions, as necessary, based on the results of the new analysis. The reanalysis did not change the paper’s results or interpretation, and our decision is that a correction rather than a retraction is warranted.
Readers should be directed to the authors’ correction and reanalysis (appendix 1), which can be found in this Correction’s supplementary materials. Commentaries from internal and external reviewers, correspondence with the authors, and a timeline of the review process (appendix 2) are also available, and further information on the issue can be found in an accompanying summary (doi:10.1136/bmj.k3213) and commentary article (doi:10.1136/bmj.k3259).