To analyze posts by medical professionals on social media, we used a mixed methods approach blending the capabilities of a social media analytics tool, Crimson Hexagon, with the capabilities of a qualitative analysis tool, NVivo (QSR International), for data collection and analysis (Figure 1).

In general, the methodology started with data collection. The researchers agreed on a data range of interest, target social media platform, target users, keywords used to search for online posts, and restrictions to impose. Second, qualitative analysis was conducted using NVivo to explore the collected data to identify relevant tweets, infer prominent concepts, and then identify the main themes in the data. Qualitative analysis can be used to uncover important concepts and develop an understanding about a phenomenon [14]. A popular method for qualitative analysis is qualitative coding [15], which was adopted in this study. Qualitative coding is the process of assigning descriptive or inferential labels to chunks of data, which may assist concept development [16,17]. Third, a data analytics tool, Crimson Hexagon, was used for opinion analysis of the predefined categories. Crimson Hexagon, a social media analytics company that is now part of Brandwatch, employs unsupervised and supervised machine learning techniques and a text analysis model developed by Hopkins and King [18].

Our target social media platform for data collection was Twitter. Initially, we identified 119 medical professionals who were actively discussing the COVID-19 pandemic on Twitter. The medical professionals were identified by searching the Onalytica website, which specializes in providing influencer marketing software, and finding a list of top health care professionals ranked by influence score [19]. Also, we used the Johns Hopkins Coronavirus Resource Center [1], which provides a comprehensive COVID-19 case tracker as well as other useful information about COVID-19, including the Johns Hopkins COVID-19 Experts/Centers account on Twitter. The Twitter IDs of the medical professionals were used to identify the target users. Next, using Crimson Hexagon with the search query shown in Figure 2, we extracted all tweets for the identified medical professionals between December 1, 2019 and April 1, 2020. A total of 10,096 English tweets were collected. The key advantage of using a social media analytics platform such as Crimson Hexagon is that it provides access to the “Twitter firehose” (ie, every public tweet ever posted on Twitter in any language and from any geographic location that meets the search criteria).

For the data analysis, we started by identifying relevant tweets. To do this, a random subset of 250 tweets were analyzed by three researchers to determine which tweets were relevant to the COVID-19 pandemic. Three researchers independently labeled the tweets as relevant or not relevant. To ensure that the obtained results were reliable and consistent, we followed a case study protocol and established interrater reliability. We obtained a Fleiss κ value of 0.628, which is at the bottom of the range that reflects substantial agreement (0.61-0.80) and is just above the range that reflects moderate agreement (0.41-0.60) [20].

Next, we performed automatic coding of the relevant tweets using NVivo, a tool that helps organize and analyze a wide variety of data, including but not limited to documents, images, audio, video, and social medial content [21]. Automatic coding was used to assist the process of concept development related to the COVID-19 pandemic. The results of the automatic coding process are shown in Multimedia Appendix 1. Once the main codes and subcodes were identified, three researchers worked together via a combination of inductive and deductive thinking to identify conceptual categories [22]. The goal was to build a descriptive, multi-dimensional preliminary framework for later analysis. Based on the main code and subcode analysis, we were able to identify eight main categories/themes, namely information and knowledge, symptoms and illness, fighting misinformation, infection and transmission, testing, actions and recommendations, the health care system, and immunity.

We initially used Crimson Hexagon to define the categories from the qualitative analysis and the associated trained algorithm to explore the medical professional opinion surrounding the COVID-19 pandemic outbreak. Multimedia Appendix 1 describes each of the categories, lists the keywords delineating each of the categories, and provides a representative tweet for each. Using Multimedia Appendix 1 as a codebook, we manually labeled the categories and automatically distributed 10,096 tweets over 9 categories: the 8 categories from qualitative analysis and 1 additional category for irrelevant tweets. The training was an iterative process, ensuring that each category was clearly outlined by the examples. The number of coded tweets increased over several runs of the model as we reviewed the categories and coded more tweets.

A total of 10,096 English tweets were collected between December 1, 2019 and April 1, 2020 from a total of 119 providers, of which 29 (24.4%) were female, 59 (49.6%) were male, and 31 (26.0%) were of unknown gender. The average number of tweets per provider was 84.8. The distribution of tweets per country is shown in Figure 3; the majority of tweets were from medical professionals located in the United States.

The distribution of the tweets over the categories identified using qualitative analysis is shown in Figure 4. Overall, the results demonstrate that relevant tweets account for 92% of the collected tweets, and irrelevant tweets account for 8%. Irrelevant tweets are tweets posted by medical professionals that do not discuss COVID-19. For example, tweets that reference websites links, such as “words have never been more well-spoken. #coronavirus #coronavirusoutbreak #covid19” or tweets that make announcements about TV interviews, such as “I gave interview last month where I was asked “Is COVID-19 going to be like Zika, where nobody was really affected in the end?” and “It’s great to see the media interviewing actual experts on epidemics about #COVID19!” The distribution of the relevant tweets was as follows: actions and recommendations (2827/10,096, 28%), fighting misinformation (2019/10,096, 20%), information and knowledge (911/10,096, 9%), health care system (909/10,096, 9%), symptoms and illness (810/10,096, 8%), immunity (707/10,096, 7%), testing (605/10,096, 6%), and infection and transmission (503/10,096, 5%).

Figure 5 shows the volume of tweets over time by category. As shown in the figure, the number of posted tweets increased with time. More tweets were posted as the number of COVID-19 cases increased. The number of tweets posted about actions and recommendations increased noticeably, followed by tweets posted about fighting misinformation.

Overall, the tweets revealed the important topics about the COVID-19 pandemic that medical professionals discussed during the period of the study. Medical professionals provided a wide range of actions and recommendations that must be considered by the government, public health officials, and individuals. These actions and recommendations mainly focused on flattening the curve, quarantine, self-isolation, social distancing, staying at home, and personal self-care. According to Multimedia Appendix 2, tweets related to “must quarantine,” “social distancing,” “flattening,” and “curve” dominated this category. Example tweet:

Medical professionals also discussed concerns related to misinformation and how dangerous and misleading it is to share and report such information. Furthermore, they encouraged the public to seek updates from government officials and trusted sources. According to Multimedia Appendix 3, tweets related to “misinformation” and “disinformation” dominated this category. Medical professionals discussed information related to the virus infection and transmission. Example tweet:

Medical professionals shared their opinions regarding the health care system during the COVID-19 pandemic. They shared information and concerns regarding health care workers and their safety and whether the health care system can accommodate the increasing numbers of patients with COVID-19. According to Multimedia Appendix 4, tweets related to “healthcare,” “hospitals,” “system,” “patients,” and “workers” dominated this category. Example tweet:

Medical professionals posted tweets about general COVID-19 pandemic information and knowledge. This information and knowledge included recent statistics and comparisons between countries and general information about the virus not directly related to any of the other categories. According to Multimedia Appendix 5, tweets related to “information,” “important,” and “knowledge” dominated this category. Example tweet:

Medical professionals also shared their knowledge about the symptoms and illnesses associated with COVID-19. According to the tweets in this category, people infected with COVID-19 can have a range of symptoms, from none to severe. The tweets indicated that some patients with COVID-19 can have no immediate symptoms or no symptoms at all (asymptomatic), most patients seem to have no or mild symptoms, and some patients have pneumonia or breathing issues. According to Multimedia Appendix 6, tweets related to “symptoms,” “asymptomatic,” “illness,” “mild,” and “severe” dominated this category. Example tweet:

Medical professionals shared thoughts and opinions about how our immune systems respond and react to the virus as well as theories related to immunity. According to Multimedia Appendix 7, tweets related to “immunity,” “herd,” “strategy,” and “immune” dominated this category. Example tweet:

The medical professionals also discussed testing in their tweets. These discussions were mainly about testing as the most viable option to control the disease, concerns about testing, and the need to scale up testing and expand testing capabilities. According to Multimedia Appendix 8, tweets related to “testing,” “test,” “kits,” “lab,” “mild,” and “severe” dominated this category. Example tweet:

Medical professionals also discussed information related to virus infection and transmission. According to Multimedia Appendix 9, tweets related to “transmission,” “spread,” “outbreak,” and “infected” dominated this category. Example tweet:

The collected data and analysis show that social media content reveals important topics that medical professionals perceive as relevant to the ongoing discourse about the COVID-19 pandemic. These topics are mainly related to actions and recommendations, fighting misinformation, the health care system, information and knowledge, symptoms and illness, immunity, testing, and infection and transmission. Interestingly, tweets relating to actions and recommendations and concerns about misinformation accounted for more than 50% of relevant tweets, while health care system–related tweets accounted for less than 10% of relevant tweets. This is revealing given that discussion of shortages of medical supplies and limitations of the health care system seems to dominate mainstream media. However, while medical professionals are concerned about the health care system, from their perspective, the importance of actions and recommendations reflects a proactive stance to combat a pandemic that currently has no effective treatment and for which a vaccine will not be available for a long time. Tweets in this category peaked around mid-March, coinciding with the ongoing effort to curtail the pandemic, and included extensive references to approaches such as social distancing, quarantining, and contact tracing. These approaches are considered to be the first response to new infectious diseases [23]. While the volume of tweets has declined since mid-March, there was an uptick toward early April, potentially coinciding with conversations associated with the appropriate timing for “opening” the economy and associated measures that may be needed to keep the pandemic in check.

Furthermore, misinformation is a major concern for medical professionals; this was addressed in many tweets, which emphasized how this misleading information could lead to infection of more people with the virus. In a sense, some tweets suggested that the spread of misinformation was equally as disconcerting as the spread of COVID-19. Although the actual process by which such infection and spread could occur due to misleading information is not clear, there is ongoing effort by government and public health organizations such as the Centers for Disease Control and Prevention (CDC) and WHO to disseminate credible information about the state of the pandemic. This effort is imperative to develop interventions to fight misinformation in cases where high quality information may literally be a life-and-death concern [24]. As of April 1, 2020, this topic remained second with respect to tweet volume, indicating continued concern. Implications for public health include the need to expand the reach of credible information about various aspects of the virus, including symptoms, treatment, testing, vaccination, and progression. It is also important to increase public awareness about the duty to share information wisely and the importance of seeking information from trusted sources such as medical professionals.

Concerns were also shared about health care systems and health care workers. Medical professionals expressed concern that the increase in the number of cases will lead to collapse of the health care system and that the shortage of medical personal protective equipment (PPE) will increase the likelihood that health care workers will be infected. According to the CDC [25], it is critical to make every effort to protect the essential national workforce of health care providers, both at work and in the community. Also, CDC data show that PPE shortages are posing challenges to the health care system because of the COVID-19 pandemic [26]. Examples of suggested measures include scaling up existing facilities, provisioning field hospitals, and directing resources to support ailing health care infrastructure in a timely and proactive manner. Interestingly, the number of tweets showed a declining trend toward the end of the analysis period, potentially reflecting the global effort to ramp up health care infrastructure and the adaption of health care providers to the fledging health crisis.

In addition, medical professionals shared their knowledge and information about the symptoms associated with COVID-19; they stated that a person with COVID-19 can show a wide range of symptoms and it is even possible that they will show no symptoms at all. This is aligned with the existing literature, where accumulating evidence is indicating that a substantial fraction of people infected with COVID-19 are asymptomatic [27]. These posts reflect less than 20% of all relevant posts; however, they reflect another opportunity for medical professionals to influence the course of the pandemic by helping to disseminate credible and accurate information about the disease. This role can also extend to proactively debunking misinformation. While the volume of tweets showed a declining trend after mid-March, it is encouraging to note that there was an uptick toward the end of the analysis period.

Issues related to how the human immune system acts and reacts with the virus were also discussed. Most notably, medical professionals were concerned about the fact that some countries are considering herd immunity as an option to address the COVID-19 pandemic. This option is criticized because it is practically impossible to perfectly tune actual interventions without exceeding or undershooting the capacity of the health care system [28]. Interestingly, immunity shows a consistent upward trend during the analysis period among all topics. This trend is likely to continue as health care professionals, policymakers, and communities attempt to determine the role of the immune system (particularly post-infection) in suppressing the likelihood of future infections and how these findings can impact future courses of action.

Medical professionals tweeted about testing as the most urgent and efficient option to control the spread of COVID-19 while there is no effective treatment or vaccine. Given the magnitude of the COVID-19 pandemic, effective testing can reduce or prevent the need for much greater intrusions [29]. Suggested measures included increasing testing, surveillance, and detection as much as possible, adopting drive-through testing, and rapid scaleup of diagnostic testing outside of hospitals.

Finally, information about how the virus infects people, transmits from person to person, and spreads in communities was shared by different medical professionals. In this category, there was a significant emphasis on the role of the public in stemming the transmission of the infection. This information included the need to practice social distancing, basic personal hygiene, and self-quarantining after potential exposure to COVID-19.

This research has a number of limitations due to its reliance on social media. For example, despite the breadth of tweets collected, not all medical professionals use Twitter, and those who do use Twitter use a significant amount of discretion with respect to their level of engagement with the platform. There are also temporal and geographic dimensions [5,6] that are not necessarily captured. The findings from the analysis could be improved through additional refinement of the defined categories and by focusing on specific categories (eg, actions and recommendations). In addition, the data collection could be complemented with surveys of medical professionals with more focused and specific questions to better understand their specific concerns and experience.

In this research, we analyzed tweets by medical professionals on social media to understand topics, insights, and information about the COVID-19 pandemic outbreak. Using a mixed methods approach that blended social media analytics and qualitative analysis, this research revealed trending themes and topics of concern by medical professionals about the novel coronavirus. While this health crisis is still unfolding, this study provides a unique perspective of medical professionals during the early stages of the pandemic outside of China. At this stage, a sizeable volume of tweets pertained to proactive actions to combat the virus and to recognition of the scale of the spread of misinformation as well as its adverse effects on the ongoing effort to fight the pandemic. Other issues characterizing this stage included concern about the current status of the health care system, the dissemination of information about the disease, the role of testing to better assess the scope of the crisis and properly target mitigation efforts, and the potential response of the human immune system.