Cabanac Guillaume, Oikonomidi Theodora, Boutron Isabelle
Computer Science Department, IRIT UMR 5505 CNRS, University of Toulouse, 118 route de Narbonne, 31062 Toulouse cedex 9, France.
Inserm, Université de Paris, Centre of Research in Epidemiology and Statistics (CRESS), 75004 Paris, France.
Scientometrics. 2021;126(6):5285-5304. doi: 10.1007/s11192-021-03900-7. Epub 2021 Apr 18.
Preprints promote the open and fast communication of non-peer reviewed work. Once a preprint is published in a peer-reviewed venue, the preprint server updates its web page: a prominent hyperlink leading to the newly published work is added. Linking preprints to publications is of utmost importance as it provides readers with the latest version of a now certified work. Yet leading preprint servers fail to identify all existing preprint-publication links. This limitation calls for a more thorough approach to this critical information retrieval task: overlooking published evidence translates into partial and even inaccurate systematic reviews on health-related issues, for instance. We designed an algorithm leveraging the Crossref public and free source of bibliographic metadata to comb the literature for preprint-publication links. We tested it on a reference preprint set identified and curated for a living systematic review on interventions for preventing and treating COVID-19 performed by international collaboration: the COVID-NMA initiative (covid-nma.com). The reference set comprised 343 preprints, 121 of which appeared as a publication in a peer-reviewed journal. While the preprint servers identified 39.7% of the preprint-publication links, our linker identified 90.9% of the expected links with no clues taken from the preprint servers. The accuracy of the proposed linker is 91.5% on this reference set, with 90.9% sensitivity and 91.9% specificity. This is a 16.26% increase in accuracy compared to that of preprint servers. We release this software as supplementary material to foster its integration into preprint servers' workflows and enhance a daily preprint-publication chase that is useful to all readers, including systematic reviewers. This preprint-publication linker currently provides day-to-day updates to the biomedical experts of the COVID-NMA initiative.
预印本促进了未经同行评审的研究成果的开放和快速传播。一旦预印本在经过同行评审的期刊上发表，预印本服务器会更新其网页：添加一个指向新发表作品的显著超链接。将预印本与已发表文献关联起来至关重要，因为它能为读者提供经过认证的最新版本的作品。然而，主流的预印本服务器未能识别所有现有的预印本 - 已发表文献链接。这一局限性要求对这项关键的信息检索任务采取更全面的方法：例如，忽视已发表的证据会导致对健康相关问题的系统评价不完整甚至不准确。我们设计了一种算法，利用CrossRef公开且免费的书目元数据来源来梳理文献，以查找预印本 - 已发表文献链接。我们在为一项由国际合作开展的关于预防和治疗新冠肺炎干预措施的实时系统评价所确定和整理的参考预印本集上对其进行了测试：即COVID - NMA计划（covid - nma.com）。该参考集包含343篇预印本，其中121篇在经过同行评审的期刊上发表。虽然预印本服务器识别出了39.7%的预印本 - 已发表文献链接，但我们的链接器在未参考预印本服务器任何线索的情况下识别出了90.9%的预期链接。在这个参考集上，所提出的链接器的准确率为91.5%，灵敏度为90.9%，特异性为91.9%。与预印本服务器相比，准确率提高了16.26%。我们将此软件作为补充材料发布，以促进其融入预印本服务器的工作流程，并加强对所有读者（包括系统评价者）都有用的每日预印本 - 已发表文献追踪。这个预印本 - 已发表文献链接器目前为COVID - NMA计划的生物医学专家提供每日更新。
