Data were collected using the publicly available Twitter analytics platform Symplur Signals [19]—a health care social media analytics platform that utilizes algorithms with natural language processing to provide in-depth information on Twitter activity. Data on topics related to COVID-19 were collected by performing specific searches categorizing the topics by organ system (Multimedia Appendix 1). Data were captured over the first 6 months of the COVID-19 pandemic, from December 1, 2019, to May 31, 2020. In an effort to capture the longitudinal evolution of social media use during this period, each month was split into half (ie, day 1-15 and day 16 to the end of the month). Data collected included total number of tweets and retweets, total number of impressions, total number of users, and user data, including the place of origin—by country (globally) and by state (within the United States). The ratios of tweets per Twitter user and impressions per tweet were also calculated (Figure 1). Definitions of these terms can be found in Multimedia Appendix 1. 

Preprint articles are research manuscripts shared publicly before peer review, which allows for rapid dissemination of information, thereby helping to inform policy and clinical practice in a timely manner. Preprint repositories have gained considerable attention over the course of the COVID-19 pandemic and have been increasingly utilized for the dissemination of crucial pandemic-related research. For our analysis, preprint articles related to COVID-19 biomedical research were identified using two popular preprint servers: medRxiv and bioRxiv. Specific search terms (see Multimedia Appendix 1) were used to identify and extract COVID-19 preprint articles for each half-month period across the 6-month study period for comparison with Twitter data. Furthermore, a follow-up review of the preprint articles pertaining to COVID-19 and gastroenterology that ultimately resulted in formal peer-reviewed journal publications was performed for the month of July 2020. This was done to account for the delay associated with the formal publication of a preprint article.

For the analysis of peer-reviewed publications, the PubMed–NCBI database was used to search for all publications pertaining to COVID-19 over the 6-month study period. The specific search terms used are detailed in Multimedia Appendix 1. All citations resulting from PubMed searches were recorded, and the search results were further filtered by half-month time intervals, identical to the search methods used for Twitter content and preprint articles for the purposes of comparison. For both preprints and publications, articles were further subgrouped by organ system topic. Duplicate publications from separate searches were individually reviewed and recategorized into the most appropriate subject group, thereby eliminating the potential for publications to be accounted for more than once. Finally, for each publication, the geographic location of the first author’s institution was recorded. 

The primary outcome of the analysis was to identify the peak activity across the 3 platforms, as this shows how efficiently one platform could disseminate information as compared to the others. “Activity” was defined by the number of tweets (via Twitter), or publications (when referring to preprint or publication databases, such as bioRxiv and medRxiv or PubMed, respectively) produced. “Dissemination” could be defined in different ways as referenced by a previous systemic review report published by the Agency for Healthcare Research and Quality (AHRQ); however, for the purpose of this study, it was defined as “the active and targeted distribution of information or interventions via determined channels using planned strategies to a specific public health or clinical practice audience” [20]. In this study, the information distributed is the content of tweets or articles published in preprint repositories or accessible via the PubMed database. Of note, Twitter impressions is a convenient way to measure the exact distribution of the tweet content, as it calculates how many users would have been sent a particular tweet based on the number of followers the user who posted the tweet originally had. Secondary outcomes included (1) peak activity in each platform overall and then by GI subtopic, (2) peak activity in each platform by geographic location, and (3) comparison of trends between the different platforms overall, as well as by GI subtopic (see Multimedia Appendix 1).

Summary statistics of baseline data for tweets, impressions, preprints and PubMed publications are presented as frequencies for categorical data unless otherwise specified. Spearman’s rank-order correlation was performed to determine the relationship between Twitter activity (ie, tweets and impressions) and PubMed publications overall, by organ system and geographic location, as well as by Twitter activity and preprint articles overall and by organ system. Analysis was performed using STATA 15 (StataCorp, LLC) software. Statistical significance was set at P<.05. All authors had access to the study data and reviewed and approved the final manuscript.

Over the 6-month study period from December 1, 2019, through May 31, 2020, 73,079 tweets were identified from a total of 44,609 users, generating 207,039,610 impressions on the topic of COVID-19. During this same period, 7164 publications pertaining to COVID-19 were found to be indexed in PubMed along with 4702 preprints archived in medRxiv and bioRxiv repositories. The overall summary of Twitter and research publication activity by half-month time interval is shown in Table 1. Twitter activity, with regard to original tweets on the topic of COVID-19 did not appear until the latter half of January 2020, which resulted in 245 original tweets. This activity progressively increased thereafter and peaked during March 16-31, 2020, with 20,660 original tweets before gradually decreasing over the remaining study interval (Figure 2).

A similar pattern of activity was observed among the number of Twitter users who posted tweets related to COVID-19, which increased from 165 users to 13,034 users between January and the latter half of March 2020. Twitter impressions followed a similar pattern, with a peak observed during the second half of March 2020. Interestingly, a second peak in impressions was apparent during the latter half of May 2020, which was not observed with regard to the number of tweets and Twitter users (Figure 3). On average, the number of tweets per Twitter user ranged from 1.48 to 1.97. Impressions generated per tweet were initially high (5874 impressions/tweet) in the latter half of January 2020 but did not peak until the latter half of May 2020 (6880 impressions/tweet). Temporal trends are further detailed in Figure 2.

Scientific COVID-19–related articles indexed in PubMed as well as preprints in medRxiv and bioRxiv servers followed a similar trajectory as Twitter activity, with both the first peer-reviewed articles and the first preprints becoming available during the second half of January 2020 [21]. However, unlike Twitter activity that peaked in the second half of March 2020, publications and preprints reached peak activity around the second half of April 2020. Notably, we observed a parallel rise in the number of preprints and PubMed publications (Table 1 and Figure 2).

A moderately strong correlation was demonstrated between Twitter activity (ie, number of tweets) and number of PubMed publications (ρ=0.58), as well as between Twitter activity and number of preprints (ρ=0.57) across the study duration (P<.001 for both; Table 2).

Similarly, there was a moderately strong association between the number of Twitter impressions and PubMed publications (ρ=0.56, P<.001), as well as between the number of Twitter impressions and preprints (ρ=0.54, P<.001; Table 3).

The top 5 countries with the highest number of COVID-19–related tweets posted globally over the 6-month study period included the United States (7331/22,215, 33.0%), followed by the United Kingdom (4229/22,215, 19.3%), Spain (1527/22,215, 6.8%), Canada (1174/22,215, 5.3%), and Australia (673/22,215, 3.0%). China generated the highest number of peer-reviewed publications indexed in PubMed (1768/6352, 27.8%) throughout the study period, followed by Italy (915/6352, 14.4%), the United States (389/6352, 6.1%), France (348/6352, 5.5%), and India (303/6352, 4.8%). Figure 6 illustrates the countries with the highest Twitter activity and peer-reviewed publication activity. The top 20 countries with the highest number of tweets and the highest number of peer-reviewed publications are listed in Tables S3 and S4, respectively, in Multimedia Appendix 2. There was a strong correlation between the number of tweets and peer-reviewed publications in both the United States (ρ=0.8, P=.005) and the United Kingdom (ρ=0.8, P=.01; see Table S2 in in Multimedia Appendix 2).

Within the United States, when analyzing both Twitter activity and peer-reviewed publications, the state of New York had the highest COVID-19–related activity (11% of tweets and 39.1% of publications) during the study period, followed closely by California (10.3% of tweets and 36.6% of publications). Figure 6 illustrates the Twitter activity and publication activity in the US states. The top 20 US states with the highest number of tweets and peer-reviewed publications are listed in Tables S5 and S6, respectively, in Multimedia Appendix 2.

Twitter user data entered as the Twitter user’s self-designated health care stakeholder role was analyzed. For the topics of pulmonology or critical care, luminal gastroenterology, hepatology, IBD, and GI endoscopy, the top 2 most active stakeholder categories were doctors/physicians and researchers/academic users. Furthermore, for the topic of pancreatology, advocacy organizations and patient advocates were the most active stakeholder users. The top 15 most active users and their stakeholder roles categorized by Twitter activity for each organ system topic are further detailed in Table S7 in Multimedia Appendix 2.

Since its emergence in December 2019, the novel coronavirus SARS-CoV-2 has triggered an unparalleled global response in the fields of science, medicine, public health, and technology. Considering its highly contagious nature, along with the paucity of knowledge and current lack of effective treatment modalities to combat the infection, the need for rapid sharing and dissemination of information has been paramount. In this study, we assessed the dissemination of COVID-19–related information via preprint services, formal peer-reviewed publications, and through the global reach of the social media platform Twitter.

Specifically, we observed that during the second half of March 2020, when COVID-19 was continuing to spread rapidly prompting various nations, including the United States, to enter a state of lockdown, social media activity on Twitter was at its peak, with almost 7000 impressions per tweet analyzed. Furthermore, we observed that Twitter activity was strongly correlated with the published scientific data available to the general public. Although COVID-19 has been predominantly linked with severe pulmonary complications, approximately 20% of the conversations on social media was related to the field of gastroenterology with specific discussions related to hepatology, GI endoscopy, luminal gastroenterology, IBD, and pancreatology. Social media activity was strongly associated with the availability of published data pertaining to all gastroenterology topics with the exception of pancreatology. Finally, in our analysis of data by geographic region of publication, social media activity was most prominent in the regions most affected by the pandemic both globally (in regions where social media via Twitter is not banned) and within the United States, with strong associations between social media and publication data.

In the longitudinal assessment of publication activity, the authors observed that peak social media activity predated peak PubMed publications by approximately 30 days. One possible explanation for this lag interval is the technical delay that typically occurs after acceptance of a peer-reviewed manuscript by a journal and prior to indexing in PubMed. It is worth highlighting that around the peak of PubMed publication activity, we also witnessed a parallel rise in the preprint repository activity. Historically, the use of preprint repositories has allowed researchers to “claim the space” or even to “publish first” in contentious fields of science and research. However, in the face of an evolving pandemic, preprint repositories are serving as a new mode of scientific communication, bypassing the typical lengthy peer-review process and thus allowing for faster dissemination and communication of research and clinical findings related to COVID-19. In fact, in March 2020, when the World Health Organization officially declared COVID-19 a pandemic, 8830 biomedical preprints were published, a 142% increase from the previous year, and medRxiv page views had increased to 15 million a month, as compared to 1 million a month prior to the start of the pandemic [22]. As of July 1, 2020, we were able to document that nearly half of all preprint articles related to COVID-19 and gastroenterology have subsequently made their way into reputable scientific journals, further supporting this theory.

Notably, we observed that physicians, nonmedical doctors, and scientific researchers constituted the lead stakeholder activity for social media use overall during the COVID-19 pandemic. This observation suggests that even during the early stages of the COVID-19 pandemic, social media became an increasingly sought-after tool, likely for the purpose of communicating medical and scientific information. This study confirms—by documenting health care stakeholder activity—that the scientific and medical communities leveraged social media platforms during the early stages of the COVID-19 pandemic more so than other health care stakeholders such as patients or advocacy groups. Previous studies performed during nonpandemic times have demonstrated conflicting evidence regarding the impact of social media on scientific publications related to citation impact, metrics, and viewability [23-26]. This study clearly illustrated that social media activity was greater and also peaked earlier than any of the publication modalities, which is important to consider when investigating how social media could potentially impact research and scientific work in future.

The gastroenterology subtopic of pancreatology is a good example of this potential influence, or rather, the lack thereof. Although pancreatic manifestations (eg, acute pancreatitis) of COVID-19 have been less commonly reported in the literature as compared to other GI symptoms, we unexpectedly observed that Twitter activity on the topic of pancreatology was the highest among all gastroenterology subtopics. However, there was no statistically significant correlation between Twitter and publication activity or between Twitter and preprint activity on the topic of pancreatology, thus implying that the rise of social media activity did not necessarily reflect the same rise in activity in bioRxiv, medRxiv, or PubMed activity. This could possibly be explained by reviewing the stakeholder data for pancreatology-related tweets. For each of the other gastroenterology subfields, stakeholder demographics fell largely under the categories of physicians, nonmedical doctors, or researchers and academic users; however, for the topic of “pancreas,” the most active stakeholders appeared to be advocacy organizations or patient advocates. Upon further review of the top 50 associated hashtags used for the “pancreas” topic, those related to cystic fibrosis support and awareness groups were the most common. Patient advocacy organizations not only play an important part in espousing awareness of specific diseases and patient populations, but they also serve an essential role in the dissemination of information on their behalf, typically in the form or raising awareness on the internet, lobbying directly for change within the government of other institutions, and via marketing and outreach, but likely only indirectly in the context of research activity. Therefore, this increase in Twitter activity may not translate to a proportionate increase in research publication activity on the topic of COVID-19 and pancreatology as demonstrated in this study. Using this topic as an example, further investigation diving deeper into the content of tweets, specific users, and how they translate directly into published work would be helpful to document the direct impact social media has on research activity.

There are several limitations to this study that are worth noting. The first major limitation is that social media platforms, such as Twitter, are not available in several countries, including China. This could help to explain China’s lead in research publication activity as compared with other regions, as publication is likely one of the primary modalities used in this country for disseminating information. Other social media platforms, including WeChat and Sina Weibo, are used in China; however, information regarding the use of these other social media platforms within China is limited to date. Future studies are needed to assess social media activity on these alternative platforms and their association with publications as well as how they compared to the use of Twitter in other nations, such as the United States (where Twitter is not banned). Second, although we were able to account for and reassign duplicate publications for the various categorizations performed, a similar approach could not be guaranteed for Twitter data. The very nature of tweets allows for other users to publish an original tweet to their account generating additional impressions as it is sent to their followers (also known as a “retweet”). Therefore, limiting duplicate tweets may artificially decrease activity. More importantly, certain tweets, or even retweets, may have been assigned to more than one topic area owing to limitations of the Symplur software. Hence, we were unable to limit tweets to a single topic, and as such, this may have artificially boosted the overall number in certain organ system topics, thereby potentially skewing the results of this study.

In conclusion, this study demonstrates patterns in the utility of social media and publications—both in preprint repositories as well as peer-reviewed journals—for the rapid dissemination of information during a global pandemic of infectious disease. As the world faces this unprecedented public health emergency, this study has reflected on shifting worldwide trends from solely traditional methods of disseminating information (ie, via publications) to more contemporary methods, specifically among the GI community. social media tools like Twitter can be an effective method for educating and informing audiences in real time and via an interactive approach, a feat that cannot always be achieved with more conventional methods (ie, scientific publications). The new media age has resulted in a number of novel avenues for the distribution of information, including Twitter. Utilizing a single modality for dissemination of health care discoveries or information has been shown by the AHRQ to not be as effective as utilizing multiple modalities. It may therefore be time for the medical and scientific communities to cultivate formal social media platforms as effective tools for data sharing and collaboration to augment existing modalities of archival publication.