1 Trials Support Officer, York Trials Unit, University of York, York, UK.
2 Statistician, York Trials Unit, University of York, York, UK.
J Health Serv Res Policy. 2019 Jan;24(1):44-51. doi: 10.1177/1355819618797965. Epub 2018 Sep 24.
To assess the impact of retraction on the citation of randomized controlled trials.
We used an interrupted time-series with matched controls. PubMed, CINHAL, Google and the Retraction Watch Database were searched. We identified retracted publications reporting the results of randomized controlled trials involving human participants with two years of available data before and after retraction. We obtained monthly citation counts across all articles for the 24 months before and after retraction, from Web of Science. We used a Poisson segmented regression to detect changes in the level and trend of citation following retraction. We also undertook a matched control analysis of unretracted randomized controlled trials and a sensitivity analysis to account for cases of large-scale, well-advertised fraud.
We identified 387 retracted randomized controlled trial reports, of which 218 (56.3%) were included in the interrupted time-series analysis. A reduction of 22.9% (95% CI 4.0% to 38.2%, p = 0.02) was observed in the number of citations in the month after retraction, and a further reduction of 1.9% (95% CI 0.4% to 3.5%, p = 0.02) per month in the following 24 months, relative to the expected trend. There was no evidence of a statistically significant reduction among the matched controls. Authors with a large number of retractions saw a 48.2% reduction at the time of retraction (95% CI 17.7% to 67.3%, p = 0.01). Other cases had a more gradual reduction with no change at the time of retraction and a 1.8% reduction per month in the following 24 months (95% CI 0.2% to 3.4%, p = 0.03).
Retractions of randomized controlled trial reports can be effective in reducing citations. Other factors, such as the scale of the retractions and media attention, may play a role in the effectiveness of the reduction.
评估撤回对随机对照试验引用的影响。
我们使用中断时间序列和匹配对照进行研究。检索 PubMed、CINHAL、Google 和 Retraction Watch 数据库。我们确定了撤回报告人类参与者随机对照试验结果的出版物，这些出版物在撤回前两年有可用数据。我们从 Web of Science 获取了撤回前 24 个月和撤回后所有文章的每月引用计数。我们使用泊松分段回归来检测撤回后引用的水平和趋势变化。我们还对未撤回的随机对照试验进行了匹配对照分析，并进行了敏感性分析以考虑大规模、广为人知的欺诈案例。
我们确定了 387 篇撤回的随机对照试验报告，其中 218 篇（56.3%）被纳入中断时间序列分析。撤回后一个月的引用数量减少了 22.9%（95% CI 4.0%至 38.2%，p=0.02），随后 24 个月每月减少 1.9%（95% CI 0.4%至 3.5%，p=0.02），与预期趋势相比。在匹配对照中没有发现统计学上显著减少的证据。撤回数量较多的作者在撤回时减少了 48.2%（95% CI 17.7%至 67.3%，p=0.01）。其他情况的减少较为缓慢，撤回时没有变化，随后 24 个月每月减少 1.8%（95% CI 0.2%至 3.4%，p=0.03）。
撤回随机对照试验报告可以有效减少引用。其他因素，如撤回的规模和媒体关注度，可能在减少引用的效果中发挥作用。
