Assessment of Concordance and Discordance Among Clinical Studies Posted as Preprints and Subsequently Published in High-Impact Journals

This cross-sectional study examines the concordance between clinical studies posted as preprints and subsequently published in high-impact journals, including key study characteristics, reported results, and study interpretations.


Introduction
Dissemination of clinical and health science research as preprints-that is, preliminary reports of studies that have not yet undergone peer review-has grown rapidly since the launch of the medRxiv preprint server in 2019. 1,2 Although not all preprints will subsequently be published in peerreviewed journals, among those that are, the extent of the changes to studies' reporting of design, conduct, and results remains unknown. 3 Accordingly, in this cross-sectional study, we examined the concordance between clinical studies posted as preprints and subsequently published in highimpact journals (preprint-journal article pairs), including key study characteristics, reported results, and study interpretations.

Methods
This study did not require institutional review board approval because it was based on publicly available information, in accordance with 45 CFR §46. Informed consent was not needed because no patient data were used.
We used the bioRxiv/medRxiv Application Programming Interface to identify all preprintjournal article pairs from medRxiv's inception on June 11, 2019, through August 13, 2020. We limited our sample to clinical trials, prospective and retrospective observational studies, and systematic reviews with meta-analyses reporting health-related outcomes that were published in peer-reviewed journals with a 2019 impact factor greater than 10 according to InCites Journal Citation Reports. For preprint-journal article pairs, we abstracted the number of authors; first and senior authors; funding, conflict of interest, and institutional review board disclosure statements; and sample size (number of patients for trials and observational studies or number of studies for meta-analyses). In addition, we identified the prespecified primary end points, corresponding results from inferential analyses, and overarching abstract-level conclusions and study interpretations; for studies without clearly defined end points, we recorded all abstract-level end points and results or categorized the study as purely descriptive. Building on previously published methods, 4 disclosure statements, sample size, number of authors, and end points were considered concordant if they contained the same information or had numerical equivalence. Results from inferential analyses were considered discordant if effect estimates and/or 95% CIs or P values changed. For preprint-journal article pairs with descriptive results, we did not determine whether results were concordant or discordant.
Descriptive analyses were conducted using Excel software version 16.0 (Microsoft) from November 2020 to January 2021.

Results
Among 8941 preprints posted on medRxiv, 944 had been subsequently published in a peer-reviewed journal as of our search date. Of these, 47 were clinical studies published in a high-impact journal (

Discussion
Although preprints are preliminary reports that should be critically assessed, similar to prior studies of text content, 5 our findings suggest that the overarching study interpretations, as well as other study design details described in preprints of clinical studies that are subsequently published in highimpact journals, did not change markedly. Our study was limited to medRxiv preprints published in   a Other refers to preprint-journal article pairs where more information is disclosed for certain authors in the preprint (or journal article) and less information is disclosed for other authors in the preprint (or journal article). When evaluating funding and COI disclosures, we only compared authors who were concordant between preprints and journal articles.
b Sample size, results for primary end points, and study interpretations were classified as could not be compared when preprint-journal article pairs contained unclear information or reported descriptive results (ie, nonnumerical or noninferential analyses).
c Four preprint-journal article pairs had ascertainment time changes.