The epidemiology of errors in data capture, management, and analysis: A scoping review of retracted articles and retraction notices in clinical and translational research

Abstract

Abstract Introduction: To better understand and prevent research errors, we conducted a first-of-its-kind scoping review of clinical and translational research articles that were retracted because of problems in data capture, management, and/or analysis. Methods: The scoping review followed a preregistered protocol and used retraction notices from the Retraction Watch Database in relevant subject areas, excluding gross misconduct. Abstracts of original articles published between January 1, 2011 and January 31, 2020 were reviewed to determine if articles were related to clinical and translational research. We reviewed retraction notices and associated full texts to obtain information on who retracted the article, types of errors, authors, data types, study design, software, and data availability. Results: After reviewing 1,266 abstracts, we reviewed 884 associated retraction notices and 786 full-text articles. Authors initiated the retraction over half the time (58%). Nearly half of retraction notices (42%) described problems generating or acquiring data, and 28% described problems with preparing or analyzing data. Among the full texts that we reviewed: 77% were human research; 29% were animal research; and 6% were systematic reviews or meta-analyses. Most articles collected data de novo (77%), but only 5% described the methods used for data capture and management, and only 11% described data availability. Over one-third of articles (38%) did not specify the statistical software used. Conclusions: Authors may improve scientific research by reporting methods for data capture and statistical software. Journals, editors, and reviewers should advocate for this documentation. Journals may help the scientific record self-correct by requiring detailed, transparent retraction notices.