Improved dynamics of sharing research findings in the COVID-19 epidemic compared with the SARS and Ebola epidemics

To determine the speed of sharing research data and results in the COVID-19 epidemic, we measured the time-lag between the epidemic start time and the rise in the number of coronavirus-related documents indexed in the PubMed database. The PubMed database enables easy access to the Create Date (CRDT) of each article, which is the date that the article was added to the database [15]. This date is important because the articles become searchable for the research community since then. We determined the number of PubMed publications pertaining to coronaviruses indexed in each week by finding articles that have the term “coronavirus” in their “title/abstract” and indexed in “Date-Create (CRDT)” of the week range. Forty-eight weeks before the epidemic start-time was set as the control period (from 6th January to 8th December) [1, 16], and the number of documents per week for this period was assumed as the baseline level. Afterward, the number of documents per week after the epidemic start-time was analyzed, and the week in which the numbers of documents increased significantly compared to the baseline level was determined. Then, the time range between this week and the epidemic start-time and declare-time was calculated. We also did the same for the SARS and 2014 Ebola epidemic in West Africa by changing the time scale to month. Therefore, forty-eight months before each epidemic start-time was considered as the control (1st November 1998 to 1st November 2002 for the SARS [17], and 1st December 2009 to 1st December 2013 for the Ebola epidemic [18]). The time frame change was made because of the low number of publications per some weeks for these two epidemics (even zero for some records). In the case of the Ebola epidemic, the term “Ebola” was searched in the “title/abstract.”

To classify the topics investigated in the articles before and during the COVID-19, SARS, and Ebola epidemics, we used the PubMed Medical Subheading (MeSH) database. MeSH terms and subheadings are controlled vocabularies for indexing and searching biomedical literature, which is used as an indicator for the topic of an article [19]. Eleven Mesh subheadings were chosen to classify the articles’ topics into eleven categories: diagnosis, drug therapy, epidemiology, etiology, genetics, immunology, microbiology, prevention and control, statistics & numerical data, transmission, and organization & administration. Each category’s exact definition is presented in the PubMed MeSH database in the subheadings part [20]. For the COVID-19, the year 2019 was considered before the epidemic period, and the year 2020, up to 10th April, was considered the epidemic course. For the SARS and Ebola epidemics, the years 2002 and 2013 were considered as before the epidemics period, respectively and, the years 2003 and 2014 were considered the epidemic course. This consideration’s logic was based on the previous searches, which showed a marked rise in the number of relevant articles in 2003, 2014, and 2020.

To find the number of articles having each topic in them, we found the number of PubMed publications that indexed in the desired “Date-Create” and have the term “coronavirus” (“Ebola” in the case of the Ebola epidemic) in their “title/abstract” and assigned to the desired MeSH subheading. Samples from the total search results were screened manually to assess the validity of the search method. After approving the search method’s validity, all of the results were included in the analysis, and no exclusion was performed. A similar process was done for finding the records pertaining to the SARS and Ebola epidemics. Notably, each article could be assigned to a few of these subheadings or none of them; therefore, the sum of the subheadings’ frequency was not equal to the total number of articles indexed.

Finally, to determine the topics that were more investigated during the epidemic courses, two measures were compared between before and during each epidemic: First, the rank of each topic in frequency, and second, the proportion of each topic frequency to the total number of search results (relative frequency).

For analyzing the rise in the number of publications after each epidemic emerged, we omitted outlier records in the control period by considering any value that lies more than one and a half times the interquartile range (IQR) beyond the first and the third quartiles. Therefore, two records were omitted from the SARS and Ebola control periods (October 2011 and December 2012 in the Ebola and December 1998 and January 2002 in the SARS epidemic control period). Then, the Shapiro-Wilks test was performed to confirm that all of the control periods follow a normal distribution [21]. The records belong to after each epidemic start-time was analyzed using a one-sample z-test to determine the first record that shows a significant increase compared to the control period. All Data are summarized as mean (SD), and we considered differences at p-value < 0.01. Data analysis was conducted using Python (version 3.6). The SciPy library (version 1.4.1). Data visualization was performed using Tableau Desktop (version 2020).

The mean (SD) number of articles per week for the control period (from 6th January to 8th December) was 9.8 (2.9), and this level was considered as the baseline. After the epidemic start-time on 8th December 2019 [1, 16], the first week that showed a significant increase in the number of articles than the baseline was 19-25th December (documents number =21, p< 0.001) and, such increase was repeated in the next weeks (Fig. 1). Notably, 29th December to 4th January also showed a significant increase in the number of articles (documents number= 17, p< 0.01); however, this result was not repeated in the next 2 weeks, and we did not consider this record as a significant increase. Therefore, the time range between rising in the number of articles and the epidemic start-time was 6 weeks and, the time range between this rising and notifying to WHO of the epidemic (31st December 2019 [22]) was 3 weeks.

The number of search results for before and during the COVID-19 epidemic was 874 and 1745, respectively. The three most frequent MeSH subheadings assigned to the articles about coronaviruses indexed in 2020, up to 10th April, were etiology (340 articles), statistics & numerical data (272 articles), and epidemiology (270 articles) (Fig. 2). In contrast, the three most common MeSH subheadings for articles of the year 2019 were etiology (233 articles), microbiology (171 articles), and genetics (125 articles). Statistics & numerical data, epidemiology, transmission, diagnosis, and organization & administration were five topics that their ranks in frequency were increased in 2020 than in 2019. Besides, the increase in relative frequency was significant for all these topics, along with prevention and control (0.041 vs. 0.071, p-value < 0.001).

The mean (SD) number of articles per month added to the PubMed database in the control period (1st November 1998 to 1st November 2002) was 6.7 (3.27). We considered this number of articles per month as the baseline. In the epidemic course, which started in November 2002 [17], the first month, which showed a significant increase in the number of articles compared to the baseline, was April 2003 (documents number= 17, p< 0.001) (Fig. 3). This significant increase than the baseline was also repeated in the next few months. Therefore, the rise in the number of indexed articles occurred approximately 5 months after the epidemic started, and 2 months after the epidemic was reported by WHO (February 2003 [8]).

The number of papers included in the study for before and during the SARS epidemic was 282 and 886, respectively. The three out of eleven most common MeSH terms assigned to the articles that are about coronaviruses and indexed in 2003 were as follow etiology (234 articles), microbiology (122 articles), and genetics (121 articles) (Fig. 4). These three MeSH subheadings were also the most frequent ones in 2002, and the frequency was 95 articles for etiology, 62 articles for genetics, and 49 articles for microbiology. The rank of five topics in frequency was increased in 2003 than 2002 that comprises: microbiology, statistics & numerical data, epidemiology, diagnosis, and drug therapy. However, diagnosis and transmission were the topics that their relative frequency was increased significantly (0.028 vs. 0.088 for diagnosis, 0.007 vs. 0.042 for transmission, p-value < 0.01).

The mean number of articles per month indexed in the control period (1st December 2009 to 1st December 2013) was 8.1 (2.62), and this level was considered as the baseline. In the epidemic course, a sustained significant increase in the documents number compared to the baseline was started from July 2014 (documents number=17, p< 0.001) (Fig. 5). April 2014 also showed a significant increase in the documents number than the baseline (documents number=18, p< 0.001); nevertheless, this increase was not repeated in the next 2 months. This epidemic started in West Africa in December 2013 and was declared in March 2014 by WHO [18, 23, 24]; therefore, the sustained rise in the number of indexed articles occurred 7 months after the epidemic start-time and 4 months after the epidemic declare-time.

The number of papers included in the study for before and during the Ebola epidemic was 262 and 2377, respectively. The three most frequent MeSH subheadings assigned to the articles that are about the Ebola virus disease and added to the PubMed database in 2014 were statistics and numerical data (421 articles), epidemiology (407 articles), and prevention and control (373 articles) (Fig. 6). However, in 2013, the most common MeSH subheadings in the articles that are about the Ebola virus disease were etiology (72 articles), genetics (47 articles), and immunology (34 articles). The six topics that their ranks in frequency were increased comprises statistics & numerical data, epidemiology, prevention & control, organization & administration, transmission, and diagnosis. Similarly, the increase in relative frequency was significant for these topics except diagnosis (0.015 vs. 0.044, p-value= 0.012).