Determinants of downloads and citations for articles published in intensive care medicine

Medical publications undergo a comprehensive process of editor's handling, peer reviewing and editing that intends to select manuscripts that are both likely to be read and cited.


INTRODUCTION:
Three themes are key in a journal's mission and vision statement: showcasing the best science, educating the targeted readership, and framing the contemporary and key issues of our time (1). Citations and downloads are considered to represent the first and the second theme. A careful understanding of the determinants of citations and downloads are therefore important to advancing in all of the metrics that support the growing success for our journal. This study is part of a global quality improvement process for articles published in Intensive Care Medicine (2)(3)(4).

METHODS:
All papers accepted in 2012 and 2013 have been tracked through the Web-of-Science database for referencing and the Springer link statistics report for downloads from 01/01/2013 to 12/31/2014. Nonparametric correlation coefficients were used to assess the relationship between citations and downloads. Relative risk of being downloaded or cited one time have been modeled through the use of a negative binomial regression. The models were systematically adjusted on the time of exposure of the manuscript (longer for article published at the beginning of the year).
Variables introduced in the model were submitting country, manuscript category, open access, keywords, topics, number of authors, as well as h-index of first and last author.
Statistics were done using SAS 9.4 software. A p value of .05 or less was considered significant.
Downloads were independently explained by manuscript type with a major impact of conference report and expert panel, reviews and also What's new series. Other independent factors were the hindex of the last author, the use of "septic shock" in the key word list. Finally, open access papers were significantly more likely to be downloaded (Table).
Independent predictors of cites included four variables; manuscript type (Conference Reports and Expert Panel, review articles, and original manuscripts); number of authors and h-index of the first author. Last, open access papers were significantly more likely to be cited (Table).

DISCUSSION:
This study provides important elements to improve our understanding of what makes a paper read or cited. Optimizing use and identification of manuscript's keywords appears as a simple and major way to improve access to ICM articles. The number of authors directly linked to the multinational nature of the authorship should also be promoted.
Although not an open access journal, this study suggests for authors the independent impact of having their accepted manuscripts as an open access paper.
Interestingly, the article type are associated differently with downloads and cites. What's new series provides rapid up-to-date information immediately accessible for a large panel of our reader and independently explained downloads, whereas original articles were key determinants of cites.
The analysis of the altmetrics, considered to frame the contemporary and key issues of our time, remained to be evaluated. Given the important rise in cites and download in the past 6 years, a repeated analysis of the determinants of cites and downloads in 2018 will also be helpful (5,6).

Items
Relative risks (95% CI) p value Determinants of cites and downloads (multivariate analyses) P value in bold indicates the overall statistical significance within one variable with more than 2 items. NB1: The relative risk of the binomial regression indicates the relative risk of being cited (downloaded) one fold more. NB2: 5 international conferences consensus were excluded from the analysis " Consensus statement of the ESICM task force on colloid volume therapy in critically ill patients" ; " International evidence-based recommendations for point-of-care lung ultrasound" ; " The Berlin definition of ARDS: an expanded