An exploratory content and sentiment analysis was undertaken. All data were collected and reported according to the terms and conditions of Twitter, which state that content posted by individuals is publicly available to syndicate, broadcast, distribute, retweet, promote, or publish, excluding private information (eg, home addresses or identity documents) [18]. Use of tweets by individuals outside of Twitter can be carried out with no compensation paid to the individual tweeter, as use of Twitter is agreed upon as sufficient compensation [18]. The Monash University Human Research Ethics Committee (Project ID 24354) approved this project.

Publicly available tweets posted in English or with English translation (automated through Twitter), with the hashtags shown in Table 1 were included.

Hashtags were selected based on trialing various combinations through Twitter’s search function. The highest number of tweets retrieved were for the hashtags #coronavirus, #covid19, #arthritis, and #spoonie (the latter term coined by people living with chronic illness to describe various methods of pain management [7]). Hashtags were also searched as words; for example, where #arthritis was searched, arthritis without a hashtag was searched as well. This ensured that specific arthritis types mentioned in tweets without hashtags (eg, rheumatoid arthritis or psoriatic arthritis) were included. Hashtags and words were combined through the advanced search function on Twitter using the domains Hashtags, Words, Language, and Dates (Figure 1 [19]). Tweets were excluded if they originated from organizations, news outlets, or health professionals rather than individuals in order to focus on the personal perspective.

Tweets were retrospectively extracted from March 20 to April 20, 2020. The search strategy and search results are included in Multimedia Appendix 1. The search timeframe was chosen to align with the early phase of the COVID-19 pandemic and the period when many developed countries (eg, the United States, the United Kingdom, Italy, Australia) announced enforceable physical distancing or isolation measures [20-23]. The desktop version of the Twitter website (versus the mobile app) was used for data collection for ethical purposes with only publicly available tweets extracted, rather than through a private login. In addition to the tweets themselves, accompanying data fields were extracted from each tweet using a customized template. Extracted data fields included (where possible): Twitter profile blurb, gender of tweeter, country of tweeter, number of likes, number of retweets, number of replies, hashtags used, number of hashtags, and use of accompanying photos. Data were stored in a Microsoft Excel spreadsheet (v16.0, Microsoft Corp).

To address the research aims, two data analysis techniques were utilized: summative content analysis and sentiment analysis. Content analysis was used to characterize the textual contents of tweets related to arthritis and COVID-19. Content analysis is exploratory; it aims to quantify and describe unknown phenomena [24]. During the content analysis process, the primary researcher (DB) read each tweet and categorized the tweets into a representative theme and subtheme related to a topic of importance for people with arthritis during the study period.

Summative content analysis occurred through a process of coding, which involved counting and comparisons of Twitter content, followed by interpretation of the underlying context [25]. To begin with, the first 10 tweets were analyzed and allocated a summary code. The code represents the theme of a tweet (eg, “health care experiences”). As additional tweets were examined, they were given one of the original codes or allocated a new code based on new content. This process was repeated until each tweet was coded and themed. Once these original themes were finalized, they were recoded for additional context, and a second researcher (DA) checked the coding. For example, “health care experiences” was coded for similarities in people’s health care experiences, such as difficulties accessing medications. Given Twitter’s character limits, each tweet only contained one theme. The frequency of original themes and subthemes was counted to indicate importance [26].

Sentiment analysis enables an examination of written and spoken words for positive and/or negative emotion. When applied to health care or social media research, sentiment analysis facilitates interpretation of textual information about patient experience from a person-centered perspective [27]. Once tweets were coded and categorized into themes, sentiment analysis was employed to assess the emotion associated with the theme using Glaser and Strauss’s [28] 6 codes for sentiment analysis, a common framework used for Twitter-based research [29,30]:

The presence of emojis, which are shorthand facial expression symbols that are frequently used to facilitate communication of mood and emotion, in tweets were also analyzed. To provide information regarding the emotional content of the tweets, Emoji Sentiment Ranking, as outlined by Kralj Novak et al [31], was applied. Tweets containing an emoji were categorized into one of three sentiment scores: (1) negative, (2) neutral, and (3) positive. Together, the content and sentiment analyses provide a proxy indicator of the topics of interest for, and perceived emotions of, people with arthritis during the COVID-19 pandemic.

Even in social media studies, it is imperative to protect participant anonymity [32]. To avoid reverse identification of participants based on their tweets (which can be found through internet searches), tweets analyzed in this study are not quoted verbatim. Instead, all data are expressed in aggregate form through descriptive statistics and qualitative syntheses.

This study aimed to identify proxy topics of importance for people with arthritis during COVID-19 by characterizing the textual content and sentiment of tweets, and exploring the emotional context of tweets by people with arthritis during the early phase of the pandemic. We anticipated that this novel approach would ascertain contemporary topics of importance for immunocompromised and isolated individuals. Content analysis revealed seven themes relating to health care experiences, and sentiment analysis revealed that the majority of tweets contained negative emotion, particularly around medication shortages, increased arthritis symptoms, and the physical and mental toll of physical distancing or living in isolation. Our findings provide a starting point for understanding the impacts of COVID-19 on vulnerable arthritis populations and provide some insights for physicians and researchers regarding current concerns that may inform tailored care.

More than one-third of tweets discussed health care experiences, primarily focusing on the reduced availability of hydroxychloroquine. Individuals were highly anxious about hydroxychloroquine shortages after preliminary research found that the drug might act as a potential preventive measure, or possible cure, for COVID-19 and these early findings were highly publicized around the world. The original published article has since been retracted after researchers were unable to verify the reliability of their results [33]. Regardless, President Trump tweeted his support for hydroxychloroquine as “one of the biggest game changers in the history of medicine” [34], causing panic about medication shortages, particularly from those with systemic lupus erythematosus for which it is the first-line therapy [35]. Due to off-label prescriptions and hoarding practices, difficulties accessing hydroxychloroquine have been reported globally [36]. Hydroxychloroquine shortages pose a threat to the health and safety of people with inflammatory arthritis, with reports that many will experience flare-ups and may develop irreversible organ damage without their regular dose [35]. Our findings highlight the need for accurate information about treatments and their effectiveness and the critical role that clinicians play in dispelling myths and inaccuracies during times of rapidly changing information. A growing body of literature describes the potential benefits of using Twitter in clinical settings, and reports ways that clinicians are using platforms such as Twitter to communicate health information to the broader population [37-39]. In the context of infectious diseases in particular, evidence suggests that Twitter is beneficial to translate real-time clinical information [40,41]. Use of Twitter, being a component of mobile health (mHealth), also empowers patients to more positively perceive their abilities to manage chronic illness [42]. This presents an opportunity for clinicians and professional societies to use social media platforms such as Twitter to overcome evidence dissemination methods (eg, peer-reviewed articles and care guidelines) that are traditionally slow. Together, clinicians and patients can contribute to care in adults with arthritis, encouraging positive health outcomes throughout, and beyond, the pandemic.

Several tweets contained very personal narratives that highlighted individual fears of contracting COVID-19, and the challenges associated with being vulnerable to infection. Individuals described perceived barriers to accessing care, citing discomfort caused by wearing medical-grade personal protective equipment. Tweets contained accounts of neck pain attributed to wearing a mask; a documented side effect in other vulnerable populations [43]. Fitting a mask is also a dexterous task that some people with arthritis struggled to perform, and as a result, encountered abuse from others when buying groceries or going for a walk. The long-term psychosocial impact of this stigma should receive consideration in future research. Clinicians should consider COVID-19–related functional concerns for physically impaired patients, and incorporate new aspects of information-seeking into clinical consultations. For example, asking patients about their degree of difficulty with COVID-19–related functional tasks would help to elicit relevant functional challenges faced by people with arthritis and inform the provision of tailored patient education. Asking these questions also contributes to factors beyond patient education and care, such as facilitating access to nonacute symptom or pain management services such as allied health. These health services still need to be maintained during and after the pandemic, and for musculoskeletal health in particular, there is emerging commentary around the physical and psychosocial impacts of inhibited access [44,45].

Tweets detailed marked increases in general physical symptom burden; a concerning prospect as COVID-19 has impacted face to face consultations, with indications that disruptions to traditional service models will likely persist for some time. Evidence about the utility of telemedicine to manage pain is emerging, with consideration of barriers to implementation, and potential inequity in access [46], although health systems have been generally slow to implement this approach at scale. While the included tweets provide preliminary information about the growing symptom burden during the pandemic, the collection of systematic patient-reported outcomes data is needed to ensure that health care services are meeting the needs of people with arthritis during and after the COVID-19 pandemic.

A proportion of tweets related to social connection, that is, people reaching out to likeminded peers with arthritis through potentially informative blogs. Most links provided were to official (eg, CreakyJoints) or unofficial (eg, online communities and forums) blogs and provided information on how to manage physical and mental health in isolation, a range of arthritis-appropriate exercises, and existing evidence on the association between arthritis and risk of COVID-19 infection. Access to resources that are relevant, credible, and trustworthy appears to have been challenging for people with arthritis throughout the pandemic [47], and combined with high levels of misinformation online [48,49], this may account for the recent growth of platforms such as Twitter for sourcing information and advice.

Before the COVID-19 pandemic, people with arthritis primarily used social media for self-expression and positive messaging [50,51]. Our sentiment analysis (enhanced by classifying emojis to further characterize common emotions) demonstrates that the role of Twitter has evolved throughout the current pandemic to act as a space for people to share symptoms; to reach out to peers, organizations, and health professionals for information; and to create a sympathetic community of care. This is advantageous as it fosters connection between individuals with shared experiences but conversely may enable proliferation of misinformation [52]. Already, Twitter has been shown to inform clinical practice by capturing the experiences of patients with multiple sclerosis during the pandemic [53]. Understanding the COVID-19–related concerns of people with arthritis is also key to providing person-centered care and reducing distress during these rapidly changing times.

The observational exploratory nature of this study enabled us to examine topics of importance for individuals with arthritis through a person-centered lens, without ethical issues or compromising the well-being of immunocompromised patients during the pandemic. Social media research is still in its infancy, and this novel method of data collection demonstrates the concerns of people with arthritis during a time of peak anxiety. There is some indication that tweeters were representative of the general inflammatory arthritis population; the majority were female and the most common arthritis type identified was rheumatoid arthritis. Nearly half (44.0%) of tweeters were based in the United States, which currently leads the world in COVID-19 cases and deaths [54].

We also acknowledge the research limitations. It is important to note that only 15% of adults regularly use Twitter, and that younger adults and minority communities tend to be more highly represented on Twitter than the general population [55], although minority communities have been significantly impacted by COVID-19 [56,57]. Our results should therefore be interpreted as representing a small subset of people with arthritis, and not all people with the disease. Data were extracted rather than collected directly from people with arthritis and critics of social media research purport that posts or tweets are often curated and may not be reflective of reality. We have attempted to minimize this potential bias by conducting sentiment analysis, which helped us determine the emotional tone associated with Twitter content. Regardless, sentiment analysis has its limitations: populations and individuals are constantly stimulated by their political and socioeconomic surroundings and individual demographics, which can influence the content and sentiment of people’s tweets [58,59]. We also recognize the potential limitations of our search strategy (eg, we did not search for hashtags related to specific symptoms, such as pain or function, that are not unique to arthritis) that may have impacted the number of retrieved tweets. While we had a modest sample size of tweets due to our focused study aims, sentiment analysis has previously been conducted in studies with comparable sample sizes of tweets (n=260 and n=200) [60,61]. We were unable to determine the specific diagnostic category for over half the people tweeting; while a small number of tweets were from people with osteoarthritis, it is possible that more may be represented within the “diagnosis not specified” category. We were only able to analyze tweets in English, and these largely came from high-income, developed countries. Tweets in other languages and those from people in low-and-middle income countries may provide further insights, especially where the prevalence of COVID-19 infection is high [62].

This study highlights the spectrum of concerns facing people with arthritis during the COVID-19 pandemic. By exploring the content and sentiment of recent tweets, we found that individuals with arthritis conditions experience marked anxiety about medication shortages and increased physical symptom burden, and are seeking connection with and information from peers. These findings can be used to raise awareness of key issues relevant to people with arthritis during the pandemic, and to guide clinicians to tailor care that addresses the specific concerns and needs of their patients during the pandemic.