The social media platform Twitter allows for broader access to data than other social media platforms. Additionally, the nature of “tweets” (posts from Twitter users) and user profile descriptors is text-based versus image-based (such as content found on Instagram), which allows for more ready analysis of the data. Despite not being the most widely used platform, as is YouTube (81%) or Facebook (69%), Twitter is used by approximately a quarter (23%) of American adults and relatively equally among self-identified men and women and racial groups [5]. A 2021 survey of US adults demonstrated that young adults (18 to 29 years) are the predominant users of social media [5]. However, use by older adults (>65 years) has increased in recent years to 45% of older adults in 2021, which indicates that they use at least 1 social media site [5]. Given the ready availability of the data and wide use of users including individuals with CD, Twitter was chosen as the social media source for this study.

An increasingly common research practice has been to examine affinity spaces found on the popular social media platform Twitter through the use of hashtags (an author’s use of the hash symbol followed by the subject of a message) as a way to categorize and group messages; eg, #celiac and #glutenfree) [20,21]. These hashtags are conceptualized as a type of affinity space to explicate how these organic web-based spaces are used by communities to communicate, share, and find information [20,21]. As an open platform with very few barriers to participation and 330 million monthly active users [22], Twitter encourages the organic development of affinity spaces around topics and events via hashtagged keywords (eg, #celiac and #glutenfree).

The 2 topics most central to this study are “celiac” and “gluten-free.” Information available on Twitter regarding these topics might exhibit different norms in terms of who participates in these affinity spaces and how (eg, someone might want information on a gluten-free diet for non–celiac-related reasons). Hence, the original tweets that were tagged by Twitter users with either the #celiac or #glutenfree hashtag were treated as 2 different affinity spaces rather than 1 collective affinity space. These affinity spaces were then analyzed individually and compared to each other. Moreover, recognizing that many other hashtags might be used synonymously with #glutenfree or #celiac, hashtags akin to either of these terms in their relative affinity spaces also were included (ie, #gluten-free, #glutenfreediet, and #gluten_free, with #glutenfree and including #celiacdisease, #celiacs, #celiacsdisease, #coeliac, and #coeliacdisease with #celiac).

This study collected Twitter user and tweet data using the public Twitter application programming interface. Researchers used custom PHP scripts to collect query results and store them to a MySQL relational database for cleaning and analysis. The data set used in this study consisted of tweets that included any of the hashtags listed above. Given the large number of such tweets, we limited the time frame of our study to 8 months (October 27, 2019, through June 8, 2020), allowing us to have sufficient data for analysis without being influenced too heavily by a single event (eg, the US Thanksgiving holiday season or the onset of the COVID-19 pandemic). Furthermore, because our main emphasis was to understand who was posting to these hashtags, “retweets” (a user who reposts a message created by a different user) were excluded to focus only on original posts and the users who generated them. This resulted in 334,907 and 15,602 original tweets containing #glutenfree and #celiac, respectively, including those from synonymous hashtags for analysis. Table 1 provides an overview of general user and Tweet metadata over the 8-month collection period.

As is standard in analyzing data gathered from Twitter to analyze affinity spaces [20,21], all tweet and author users’ publicly available profile data (eg, Twitter handles and locations) were saved to a database. Descriptive statistics of tweet and author user objects were calculated to determine the method to use to classify users into user types for further analysis. Descriptive statistics revealed that users exhibited a highly positive skew in their posting activities. This behavior was expected given previous studies carried out on Twitter data [23]. Based on the positive skew, van Mierlo’s [24] 90-9-1 Principle was selected to classify users in each affinity space into relative activity groups. Users were classified as follows: superusers (top 1% of users posting content), contributors (next 9% of users contributing content), or lurkers (the remaining 90% of users; see Table 1) [24]. Following the standard for affinity space analysis [20,21], basic language processing techniques were then used to (1) extract keywords from user biographies (eg, “doctor” or “blogger”), (2) identify co-occurring hashtags (eg, “#vegan” or “#recipe”), and (3) identify common domains that users linked to in their tweets (eg, celiac.com). A detailed description of these categories is provided below in the Results section.

Ethics approval was obtained or determined to not be necessary by all author institutions owing to the public nature of the data.

The purpose of this study was to combine methods from computer science and the behavioral sciences to begin to examine internet-based CD educational sources. As part of this initial investigation, this study describes information about CD and the GFD disseminated on the social media platform Twitter. With increasing use of social media as an educational resource and source of support for populations of individuals with chronic illness [28-30], it is crucial to understand the nature of information on platforms such as Twitter. Our findings emphasize the prominence of posts on both CD and the GFD, which appear to come from users focused on promotion of themselves (eg, identifying as vegan) or a business (eg, endorsing a restaurant) rather than from more traditional sources of information such as medical professionals or nonprofit organizations [19]. This supports previous findings regarding the hesitancy of medical providers to engage in social media as a form of medical education [6]. It also raises concerns about the quality of information individuals are receiving about CD and the GFD, as individuals with CD require the GFD for medical purposes [19]. This is likely not unique to CD as concerns have been raised in the field of food allergies [30]. We propose the need for a social media presence focused on providing high-quality, up-to-date, fact-checked information to users, particularly for those within the CD or other gluten-related diseases.

Based on our findings, there is an opportunity and arguably a demand for increased presence on social media and internet-based platforms among medical and nonprofit experts in CD to provide high-quality information to consumers. This has been executed among populations of individuals with other diseases, such as inflammatory bowel disease (IBD). For example, ImproveCareNow [31] is a community of clinicians, researchers, parents, and patients of children and youths with IBD. The main goal of this organization is to provide a platform to help this community learn about “more reliable, proactive IBD care” [31]. Their social media campaign involves accounts on various platforms, including a blog, Facebook, Twitter, and YouTube. The content posted on these platforms is monitored by the organization.

Guidelines have been developed by several organizations to help inform medical providers on social media best practices, including the Association for Healthcare Social Media [32]. The use of guidelines can best inform medical providers on the use of social media as a source of patient education. Other groups are working to develop competencies including advocacy and communication responsibilities that specialists in various areas of health might develop in helping to educate certain patient populations [33].

There are several additional considerations for this study in analyzing publicly available Twitter data. First, we collected our sample of data during a relatively narrow (8-month) time period, which may not account for natural variations across seasons and events (eg, holidays and major scientific or medical conferences). The activity and nature of posts may have changed as the COVID-19 pandemic has continued. Second, our analysis did not examine co-occurring words within individual user accounts. For instance, it is possible that one account may note being a “vegan,” “blogger,” and “mom.” Future research could collect more detailed information about active members of social media to better understand “influencers” in this area. Furthermore, this study should be understood in light of the typical Twitter user. Twitter is used by about a quarter of American adults, both men and women of various racial groups, but we recognize that social media users may be younger and not necessarily representative of all ages and demographics [5]. Future work might examine the role of social media use in educating different subgroups of the population.

Additionally, we used established but relatively new methods of automated extraction and categorization of data rather than human coding, though we used human observation and judgment during the process of cleaning and synthesizing the data. This relied on algorithms based on anticipated data and did not allow for inductive reasoning by the human eye. Such an approach allowed the study team to rely on objective data rather than potential biases or a priori assumptions of individual experts [34]. Future studies may strengthen knowledge on this topic through expansion of data collection across a longer time span and further evaluation of the nature of users as well as the sentiments and accuracy of content within tweets.

To our knowledge, this was the first study evaluating Twitter data using the topics #celiac and #glutenfree. Given the popularity and broad use of social media, this is an important starting point for this research that generates several new hypotheses and research questions. Our findings emphasize the large volume of information communicated on social media. We suggest that platforms such as Twitter pose risks of spreading biased or inaccurate information to the public, particularly when the sources of information come from entities who may be influenced by commercial conflicts of interest.

Social media represents an immense opportunity to achieve open and clear dialogue between health care professionals and the public, which could be a major facilitator of future research and patient education about CD and the GFD.