Open Access of COVID-19-related publications in the first quarter of 2020: a preliminary study based in PubMed

Data source

In order to analyse publications concerning COVID-19 and their level of openness, we have chosen PubMed instead of other multidisciplinary bibliographic databases, like Web of Science (WoS) or Scopus for three main reasons:

Finally, after an overview carried out by the authors of this paper prior to this study in other databases, it was concluded that PubMed, unlike WoS, was also the one with more up-to-date Open Access information during the analysed period of this COVID-19 outbreak. So, if an article was OA or free to read, it could be reached through PubMed.

Search terms

Since during the global pandemic period, the scientific community is posting articles that are freely accessible through the NCBI, data were collected from the PubMed database in order to analyse every COVID-19-related scientific paper that is currently published (including PMC)⁹. In an attempt to evaluate the most accurate list of publications, we exported all results obtained from the suggested search queries offered by NLM (NCBI webpage), as follows: “2019-nCoV OR 2019nCoV OR COVID-19 OR SARS-CoV-2 OR (wuhan AND coronavirus)”. Only articles published from January 1^st to April 23^rd of 2020 were considered. No exclusions were made in the type of article (journal article, books, reviews, clinical trial or meta-analysis) or in the language, choosing in each case every article offered by PubMed. No preprints were found in the returned results.

In line with the objective of analysing published papers during other emergency circumstances, similar search procedures were applied to the SARS CoV-1 pandemic (query: “SARS CoV” OR “Severe Acute Respiratory Syndrome Coronavirus”; period searched: from 2003 to 2006) and MERS CoV epidemic (query: “MERS CoV” OR “Middle East Respiratory Syndrome Coronavirus”; period searched: from 2013 to 2016).

In order to determine the effect that this health emergency is having on the availability of the scientific production, we decided to compare it with the availability in a normalized situation, for which we performed the same analysis using two chronic diseases: low grade glioma (query: “low grade glioma”) and peptic ulcer (query: “peptic ulcer”), which, as seen by our search, have stable publication patterns for the last three years (from 2017 to 2019).

Data analysis

Obtained results, without exclusion, were exported and uploaded to OpenRefine, a free open source tool that helps exploration of large data sets, and has the capability to link and extend these data sets with different web services. In this study, OpenRefine was used to manage data but also as the key element in order to link our PubMed data set with Unpaywall through its application programming interface (API), the selected tool for analysing the OA content of all these data. Unpaywall (previously known as oaDOI) is a database introduced in 2016 as a service to check OA availability of journal articles identified by their Digital Object Identifier (DOI)¹⁰. Unpaywall is currently used more than 50,000 times a day and is maintained by Our Research, a non-profit company previously called Impactstory¹¹. It offers access to the OA status of scientific journals, through an open API. Unpaywall also shows license information and variable version availability from different repositories^10,11. In this study, Unpaywall data was collected via API in OpenRefine at the moment of the study (April 23^rd), but also at the review process of the work (in mid-July) after an update carried out in Unpaywall team and communicated to the authors of this paper. So both, data and methods are reviewed and updated in this version of the paper. This underlines the importance of ‘real-time science’ measurement, in a ‘real-time research’ publication process, like the one reflected in this paper.

Based on Unpaywall categorisation^12,13, four types of OA are considered: Gold, journal which publishes all its papers in Open Access without taking into account its business model. Here are included journals indexed by the Directory of Open Access Journals (DOAJ) but also other 100% OA journals precisely added by Unpaywall; Hybrid, subscription-based (non-OA) journals including some OA articles upon a fee, charged to the authors; Green, self-archived versions of a paper in a repository. It could be toll-access on the publisher page, with a free copy in an OA repository after an embargo period, or it could also be a Gold or Hybrid paper that the author has self-archived; and Bronze, articles freely available on websites hosted by their publisher, either immediately or following an embargo, but are not formally licensed for reuse¹². Unpaywall also provides information about Creative Commons (CC) licensing of each document (commonly Gold OA or hybrid journals). Copyright licenses, released by Creative Commons, are variable and range from more open - and therefore more reusable (CC0, Public Domain (PD), CC-BY or CC-BY-SA) to more restrictive ones (CC-BY-ND, CC-BY-NC, CC-BY-NC-ND or CC-BY-NC-SA)¹⁴. In addition to these types of licenses, Unpaywall also returns publisher-specific licenses (i.e. ACS-specific) as well as “implied OA” when there is an evidence that an OA license of some kind was used, but it is not reported directly on the webpage at this location.

Scope of the analysis and limitations

Articles from dates other than the ones specified were not considered (even if PubMed includes some out-of-date articles in its results). Only articles with a DOI were taken into account, and among them, there was a proportion not recognized by Unpaywall and thus, also not considered. Hence, the exclusion criteria after Unpaywall analysis includes out-of-date and those not scanned by Unpaywall (including papers without DOI).

Also, the Unpaywall system indexes thousands of institutional and subject repositories, but there are some still missing, and the database updates periodically, so some data might have changed.

Finally, the comparison with SARS CoV-1 and MERS CoV includes certain limitations such as differences in infection or death rates (especially with MERS). Likewise, the compared period times can also be a limitation, although this comparison is useful to demonstrate the huge current production.

COVID-19 and SARS CoV-2 pandemic publications

The data obtained about SARS CoV-2 from January 1^st to April 23^rd 2020 are shown in Figure 1. In total, 6,223 articles were retrieved from PubMed. Of these 9 were from 2019, 182 did not have a DOI assigned and 420 were not recognized by Unpaywall, and so were excluded from analysis; therefore, analysis was performed on a total of 5,612 articles.

Figure 1. PubMed-hosted SARS CoV-2 related papers published in the first quarter (Q1) of 2020 and their Open Access (OA) information.

(a) Proportion of OA papers published during SARS CoV-2 pandemic. (b) Percentage of publications divided by their OA publishing mode. (c) Number of copies present in different repositories of each OA typology. (Data extracted from PubMed: 23^rd April 2020).

When analysed, we observed that the number of articles published during the selected period increases daily. Figure 1a shows that 97.4% (n=5,467) of articles were published as OA. Regarding the type of OA, 68,3% (n=3,736) are classified as Bronze OA, followed by Gold OA (19.6%), Hybrid (9.3%), and Green OA (2.8%) (Figure 1b). It is important to mention that 78.5% of the OA papers (n=4,294) have a copy in a repository, even if they are Gold (80.1%), Hybrid (89.1%) or Bronze (75.5%), which are known as shadowed Green documents¹² (Figure 1c).

In order to deeply analyse the OA situation, we also reviewed license information of all the OA papers. Figure 2 shows that most of these articles lack a license (72.1%). Most open licenses (CC0, PD, CC-BY and CC-BY-SA) are present in 13.9% of the papers, while the most restrictive ones (CC-BY-NC, CC-BY-ND, CC-BY-NC-SA and CC-BY-NC-ND), are represented in the same proportion (13.9%) of all the considered OA papers (Figure 2b). Publisher implied licenses (named as “implied OA”) are included as the most restrictive ones as the majority of these are tied to the CC-BY-NC-ND one. Attending to each OA category, as expected, 99% of all Bronze papers don’t carry any license, with exception of 23 articles present at different repositories (Figure 2b). It is remarkable that 93 of the articles classified as Gold OA (8.7%) don’t bear any license, even if they are published at an OA journal.

Related to this licensing section, when the repository copies were analysed, we observed that these copies carry a greater number of licenses (n=1,285, the 30% of all repository copies) compared to the ones located at any other source (i.e. journal page or free PDF; n=185, 15.7% of papers at these locations) (Figure 2c). More precisely, 86% of these copies are located at PubMed Central (PMC) (n=3,693) with a 79.2% of them lacking a specific license (Figure 2d). A total of 311 papers are located in other repositories different to PMC but in this case, more than 80% do have a specific licence. At this point is important to mention that after a deeper research of articles located at PMC we have observed the presence of articles that come from journals with an explicit license on its page that do not maintain it in PMC, where there is no reflected license other than a notice from each publisher stating that “access to these papers is temporary” (see Box1).

Figure 2. Licensing of Open Access (OA) SARS CoV-2 related papers hosted in PubMed first quarter (Q1) of 2020.

(a) Distribution of papers based on license category. Licenses were divided as: CC0, CC-BY, PD, CC-BY-SA, Implied OA, CC-BY-NC, CC-BY-NC-SA, CC-BY-ND, CC-BY-NC-ND, and those without any particular license. (b) Distribution of papers with OA license (CC0, CC-BY, PD and CC-BY-SA), restricted license (Implied OA, CC-BY-NC, CC-BY-NC-ND, CC-BY-NC-SA, CC-BY-ND and ACS Specific) or without a license of each OA typology. (c) License type (open, restricted or absent) attending each manuscript location: repository or non-repository (journal page, DOAJ or PDF). (d) Number of papers present in PubMed Central (PMC), other repositories or at both locations, together with the number of these lacking any license. (Data extracted from PubMed: 23^rd April 2020).

Furthermore, the most frequent publishers and journals during this period in relation to SARS CoV-2 were studied. The most frequent publisher is Elsevier, who published ~30% of papers, followed by Wiley (13.1%) and Springer (10.6%) (Figure 3a, note that the remaining 31.9% not showed includes publishers with lower proportion than the ones shown). In terms of journals, The British Medical Journal (The BMJ), Journal of Medical Virology and The Lancet are those with the largest number of papers: 4.2, 3.1 and 2.3% of all analysed papers, respectively (Figure 3b, representing the 6 most frequent journals).

Figure 3. Publishers and journals that published the highest number of COVID-19-related papers hosted in PubMed in Q1 of 2020.

Number and percentage of total publications distributed by most frequent publishers (a) and journals (b). (Data extracted from PubMed: 23^rd April 2020).

Based on these results, we specifically studied the COVID-19-related articles published by Elsevier, Wiley and Springer (Figure 4). All three publishers release almost all SARS CoV-based articles as OA: Elsevier: 99.8%, Wiley: 98.4% and Springer: 96.8% of their published papers about the topic in the studied period (Figure 4a). All three publishers publish the majority of their papers as Bronze OA (Figure 4b), being Wiley the one with the highest proportion, with 88.3% of its OA manuscripts (n=663). Regarding Gold OA, Elsevier and Springer have published 16.5 and 16.3%, respectively, of their COVID-19 related articles under this category, a higher proportion compared to Wiley with only 3.7% (n=28). Looking at licensing, most of the OA publications from these publishers lack a license (Figure 4c). At this point Wiley is the one with the highest number of papers without any license (89.1%, n=669) compared to Elsevier (80%, n=1,332) and Springer (74.7%, n=430) matching in each case with the corresponding number of Bronze articles. On the other hand, Springer is the one with more licensed papers (25.3%, n=146 licenses) and moreover, most of them are under CC-BY (n=124) (Figure 4c)

Attending the copies located in different repositories, Elsevier publishes a copy of 99.3% of its manuscripts in different repositories, compared to Wiley (79.9%) and Springer (83.5%), and being in all cases PMC the most used one: more than 87% of these copies are in PMC. At the same time, when looked at the licensing, it is notable that there is a big proportion of copies in the three publishers lacking licenses, being again the PMC-hosted ones those with the fewest number: 79.7% for Springer, 90.3% for Wiley and 86.4% for Elsevier (Figure 4d). As mentioned, for papers that are not shared with a specific license some publishers have decided instead, to show a specific note specifying their temporary access to their papers (see Box1).

Figure 4. Analysis of the three most frequent publishers with more Open Access (OA) COVID-19 papers hosted in PubMed in Q1 of 2020: Elsevier, Wiley and Springer.

(a) Percentage of OA publications of the most relevant publishers: Elsevier, Wiley and Springer. (b) Distribution of their open content by Gold, Hybrid, Green or Bronze status. (c) Distribution of the licensed and non-licensed articles of the three publishers. (d) Distribution of the non-licensed copies hosted in all repositories found (in salmon) versus the ones found only in PMC (in pink). (Data extracted from PubMed: 23^rd April 2020).

Publications about other coronaviruses and epidemics: SARS CoV-1 and MERS CoV

In order to compare the scientific production and OA publication during global health emergencies, both SARS CoV-1 and MERS CoV-related articles were studied using the PubMed database.

In the case of the SARS CoV-1 (Severe Acute Respiratory Syndrome CoronaVirus-1) epidemic, the first case was discovered in China during November 2002¹⁵. We therefore analysed publications published in 2003, 2004, 2005 and 2006 (Figure 5). For the period from 2003 to 2006, PubMed returned a total of 2,396 articles, of which, after exclusion criteria, 1,875 were considered (476 lacked DOI and 45 were out-of-date). There was an increase in the number of publications from 2003 to 2004, with a decline onwards. The percentage of OA publications increased from 82 to 89% in the first year, maintaining a stable average of 87.6% throughout the analysed period (Figure 5a). Among these open articles, 63.1% were published as Bronze OA, 20% as Green OA, 13.3% as Gold OA, and 3.6% in hybrid journals (Figure 5b). About licensing, 82.6% (n=1,357) of the OA articles don’t carry any license (from which 24 are Gold OA), and from the licensed 17.3% (n=285), 10.7% (n=176) bear a CC-BY one (Figure 5c).

Figure 5. Publications related to SARS CoV-1 epidemic hosted in PubMed from 2003 to 2006 and their Open Access (OA) indicators.

(a) Number of total and OA publications about SARS CoV-1 epidemic during the first 4 years from the start of the epidemic. (b) OA category of the OA published articles. (c) Proportion of licensed and unlicensed OA articles. (Data extracted from PubMed: 19^th April 2020).

Next we performed the searches for the MERS CoV (Middle East Respiratory Syndrome Coronavirus) epidemic, whose outbreak began in September 2012 in Saudi Arabia¹⁶. A total of 1,069 papers were obtained for the specified period (2013 to 2016). In this case, this number is significantly lower than the one found for SARS CoV-1 over time. In 2016, the year in which most papers are registered (n=346), the percentage of these published as OA remains constant and is very high, with an average of 93.8% (Figure 6a). Unlike SARS CoV-2 and SARS CoV-1, 43.4% of MERS-related OA publications were published as Gold (Figure 6b). Almost half of the OA articles have a proper license, and among them 31% carry a CC-BY one (n=307) (Figure 6c). From the 46.4% of non-licensed articles, 23 are Gold, corresponding to the 5.3% of Gold OA articles.

Figure 6. Publications related to MERS CoV epidemic and hosted in PubMed from 2013 to 2016 and their Open Access (OA) indicators.

(a) Number of total and OA publications based on the MERS CoV epidemic during the first 4 years from the epidemic outbreak. (b) OA category of the OA published articles. (c) Proportion of licensed and unlicensed OA articles. (Data extracted from PubMed: 19^th April 2020).

As the goal is to focus on COVID-19, publishers’ and journals’ analysis has not been included for SARS CoV-1 nor MERS CoV searches as we do not consider relevant for our conclusions.

In order to determine if these results are a consequence of the current extraordinary circumstances, a control of the research was established through the analysis of open content of chronic diseases considered constant over time. We performed searches for “low grade glioma” and “peptic ulcer”, which harbour similar output levels compared to SARS CoV-1 and MERS, obtaining a constant OA proportion for each case over the last 3 years (Figure 7). This rate is low for all cases, with an average of 54.9% and 53% for low grade glioma (Figure 7a) and peptic ulcer (Figure 7b), respectively. In addition, articles concerning both diseases were mostly published as Gold OA, with a 51.3% and 60.1% of the OA articles in each case (Figure 7a and 7b). Attending to licensing, the proportion of licensed papers is 61.9% for low grade glioma and 66.21% for peptic ulcer (Figure 7c). Moreover, as a result of the high number of Gold OA papers, the proportion of CC-BY licenses is high for both cases (36.7% and 29.4%). It is important to underline that the number of repository copies for both controls represents 87.5% (n=791) and 77.7% (n=745) of the OA papers for low grade glioma and peptic ulcer, respectively.

Figure 7. Analysis of the number and OA properties of papers about two chronic diseases: low grade glioma and peptic ulcer.

Number of publications, OA proportion and category of articles related to low grade glioma (a) and peptic ulcer (b) during 2017, 2018 and 2019. (c) Proportion of licensed and unlicensed OA articles of both diseases. (Data extracted from PubMed: 20^th April 2020).

Underlying data

Zenodo: “Open Access of COVID-19 related publications in the first quarter of 2020: a preliminary study based in PubMed”^17,18 The paper reports only the filtered data. The underlying raw data (Excel data file containing Unpaywall results of each research query) are available in Zenodo:

Version 1: http://doi.org/10.5281/zenodo.3826038 under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0)¹⁷.

Version 2: http://doi.org/10.5281/zenodo.3959950 under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0)¹⁸.