Currently, the architecture of Twitter supports different user actions in response to a tweet (ie, replies, retweets, and favorites), each of them holding a specific meaning in terms of communication capabilities. Notably, replies represent the specific response to a sent tweet, retweets stand for the reposting of tweets (which is useful to quickly share and promote content), and favorites indicate that the content of a tweet is highly appreciated by the community and can be seen as a user tweet bookmark.

Figure 1 exemplifies the aforementioned communication modes. On the one hand, replies and retweets (ie, user relations) allow identifying users who support or discuss any sent message. For example, users B and C are interested in a tweet published by user A. This allows to identify how the information is spread and which are the most influential users. On the other hand, the number of retweets and favorites (ie, tweet interactions) helps to measure the relevance of the new content to the community. For example, user D likes the original tweet sent by user A and saves it for future reference.

Figure 2 depicts the workflow implemented in this study to retrieve, process, and analyze BD-related tweets, which consisted of 2 fundamental phases: (1) data collection and filtering and (2) corpus processing and analysis.

From February 1, 2018, to August 31, 2018, tweet data were retrieved via the Twitter application programming interface (API). Tweet contents and associated information were processed: whenever possible, users were geolocated, their gender was determined, and they were identified as organization or patient; tweet contents were cleaned (hashtags and mentions to user accounts were removed) and prepared for further text processing; and, entity recognition and sentiment analysis were applied.

Data collection accounted for tweets containing terms as Inflammatory bowel disease, Irritable bowel disease, Irritable colon, Ulcerative colitis, Ileocolitis, Ileitis, Crohn, Granulomatous, and Jejunoileitis. The Java library Twitter4J [22] was used to perform such collection. From all the retrieved tweets, only those written in English were considered to ensure the consistency of further examination. A total of 4.10% (1055/25,689) tweets written in other languages, such as French, Spanish, or Italian, were eliminated. The final dataset comprised 24,634 unique tweets written in English by 13,295 different users.

All the tweets were automatically labeled by tweet creator. Data present in user profiles were collected and further validated (see details in the next subsections). Specifically, user characterization involved gender determination, differentiation of organizations and patients, and geolocation.

Unfortunately, it was not possible to determine the age of the users because of the low precision of current prediction models and because several studies point that most Twitter users fall into a small range of years [23].

To analyze the content of the generated corpus, it was essential to be able to properly recognize the relevant (topic related) terms mentioned in the tweets. For this purpose, several text preprocessing techniques were applied and then an in-house–developed named entity recognizer supported the annotation of terms pertaining to the selected BD-related semantic categories.

The Results section was structured following the logical order of the proposed questions. Figure 3 illustrates how particular research questions were answered by a certain analysis and identifies the main insights provided by each of these analyses. Most notably, results were structured such that the basic characterization of the BD-related tweets in terms of user activity and topics of interest were presented first and, then, the interrelation of topics in the conversations was detailed along with the visibility of external sources of information.

The ability to identify the most active users (ie, users that post more tweets) and topic-specific communities (eg, gluten-centered community) is pivotal to gain a better understanding about how to adapt or finetune the communication (ie, reach a broader audience or focus on a specific community) as well as discover influencers (ie, vessels of information distribution, within or across the communities). The analysis of user activity (different time windows may apply to different communities) is interesting as a means to plan the best time to post a new tweet (ie, when to expect major user attention).

The semantic annotation of tweet contents goes a step forward, providing knowledge about the topics being discussed (ie, individually and in combination). Likewise, the identification of tweets linking to external information sources is relevant to bring forward the visibility these sources are receiving through Twitter.

A health-related stakeholder is a typical example of someone that can benefit from the insights provided by the overall study. Say that the aim is to plan an awareness campaign about BD and food habits. Likely, this stakeholder wants to study the target audience in terms of topics that are attracting more attention and tweeting habits. The campaign may be planned to receive short-term attention (eg, announcing the launch of a novel drug or a new food supplement) or promote awareness throughout a longer period of time (eg, promote healthier food habits).

User relations, that is, mentions and retweets, were represented in a network to study communication interplay, namely, to identify target communities and influential users (ie, individuals and/or institutions with high audiences). Figure 4 depicts this network such that the nodes denote the users (ie, account name), the node size is based on the node in-degree (ie, the users that received most communication are represented by bigger nodes), and the edges account for the number of mentions and retweets. The 4 different background areas in the figure denote the communities, mostly characterized by account descriptions: gluten-, nutritional-, BD-, and food-related relations. In this study, only communities with more than 5 connected users were considered.

In this network, the out-degree of the nodes (ie, users initiating communication with other users) is much lower than the in-degree of the nodes (ie, users receiving significant tweets). The user accounts @CrohnsColitisUK, @CrohnsColitisFn, and @HealioGastro, which belong to disease-specific organizations and have primarily informative/educational goals, were among the accounts receiving more communication (ie, higher in-degree). This conveys the rationale that individuals typically prefer to ask for health information to trusted organizations [6,53,54]. Conversely, accounts of BD patients (eg, @colitisandme) and experts in bowel conditions or nutrition (eg, @IBDMD and @charlie_lees) were among those showing highest out-degree.

Public lists of BD-related influencers [55] supported the identification of organization accounts, such as @CrohnsColitisUK, @HealioGastro, and @ACCUCatalunya, and personal accounts, such as @colitisandme, @IBDMD, and @EdwardLoftus2, that are typically reached for medical advice.

Figure 5 shows an example of a tweet exchange: the user @Crohnoid, a Crohn disease patient, asks the user @ibddoctor, an expert in BD, about 2 possible diagnostic techniques, that is, magnetic resonance imaging and computed tomography; @ibddoctor explained the advantages and disadvantages of these techniques, and another user @SandraZelinsky also intervened, pointing out limited access.

Knowledge about how the community is demographically distributed is important to carry out public information campaigns and study the impact of government and institutional actions in different demographic areas. In this analysis, 59.98% of the users (7975/13,295) were geographically distributed all around the world.

Figure 6 shows the geographical distribution of the BD communities. The size of the nodes represents the number of users located in each country (ie, the bigger the circle, the higher is the number of users) and colors identify the continents (ie, blue for America, purple for Europe, brown for Africa, green for Asia, and red for Australia).

In general, most of the users were located in the United States, the United Kingdom, Canada, and Australia, which is consistent with current knowledge about the prevalence of these conditions [56]. The highest reported prevalence is in Europe (with the highest prevalence of ulcerative colitis in Norway and of Crohn disease in Germany) and North America (with the highest prevalence of ulcerative colitis in the United States and Crohn disease in Canada). The prevalence of IBD exceeded 0.3% in North America, Oceania, and many countries in Europe.

The number of users from India, Bangladesh, and the Philippines was also noticeable and may be explained by the rising incidence of BD conditions in newly industrialized countries in Africa, Asia, and South America.

Understanding what the user community is talking about was important to identify the topics that received more attention and to be able to better align new information/communication strategies with the interests of the community. Table 1 presents the top 25 most mentioned terms along with their corresponding semantic category (ie, Food and Diet, Disease, Treatment, Symptom, and Drug) and sorted by the number of including tweets. Explicit mentions to IBS and IBD (eg, Inflammatory Bowel Disease or Ulcerative Colitis) and noncontent–bearing, generalist terms (eg, Disease or Food), were at the top of term mentions but were not listed in the table.

The #Tweets column indicates the number of tweets in which a term is mentioned, the #Favorites column indicates the number of times that a tweet containing a term was selected as a favorite, the #Retweets column indicates the number of times that a tweet containing a term was retweeted, and the #Hashtags column indicates the number of times that a term was used in hashtags, including term variations (eg, anxiety disorder like #anxiety).

The volume of retweets is useful to understand how the information flows among users, whereas the number of favorites can be understood as a metric to measure the usefulness of the tweets containing the term [57,58]. In addition, the number of times a term appears in hashtags is interesting to understand how the user wishes the tweet to be indexed (to be easily found by others with similar interests as well as to bring attention to certain topics). This can be understood as a metric to measure the impact of the term in the community [59].

As illustrated in Table 1 (and in the topological analysis of the co-occurrence network of terms in Multimedia Appendix 1), the most mentioned terms related to Disease (35.64%, 11,688/32,794) and Food and diet (25.43%, 8342/32,794) categories, followed by terms from Symptom (17.71%, 5811/32,794), Treatment (14.83%, 4864/32,794), and Drug (6.37%, 2089/32,794) categories, respectively.

Diarrhea (10.33%, 1208/11,688 of the mentions to disease terms) and constipation (8.94%, 1046/11,688 of the mentions to disease terms) were the most mentioned disease-related terms. This was somewhat expected considering that many IBD and IBS patients have diarrhea as a side effect of the disease and many others claim to have problems of constipation [60]. Interestingly, constipation is one of the terms most often included in hashtags, which indicates that the topic is meaningful and timely to the community. Other diseases typically associated with BD conditions were also discussed, notably cancer (3.37%, 395/11,688 of the mentions to disease terms), anxiety disorder (3.04%, 356/11,688 of the mentions to disease terms), depression (2.49%, 292/11,688 of the mentions to disease terms), arthritis (1.71%, 200/11,688 of the mentions to disease terms), and asthma (1.46%, 171/11,688 of the mentions to disease terms). Chronic inflammation is known to be a major risk factor for the development of gastrointestinal malignancies and is often associated with inflammatory arthritis. Notably, as the population of patients with IBD grows older, with longer periods of chronic inflammation and longer exposure to immunosuppression, there is an increased risk of developing cancer [61] and arthritis [62]. Moreover, studies show that the rates of anxiety and depression tend to be higher among patients with Crohn disease or ulcerative colitis compared with those with other diseases as well as the general population [63,64]. Finally, interest in discussing asthma is justified by the association between this disease and early- and late-onset ulcerative colitis, particularly because of shared environmental risk factors [65]. Posts discussing BD and other diseases often show high retweet rates, and it is noticeable that depression is currently at the center of attention (ie, those tweets have the greatest number of favorites).

Symptoms typically associated with the different stages of the disease such as bloating (4.40%, 256/5811 of the mentions to symptom terms), flatulence (3.20%, 186/5811 of the mentions to symptom terms), and abdominal pain (2.97%, 173/5811 of the mentions to symptom terms) were also discussed with considerable frequency and the containing tweets were among the most retweeted, that is, people within this community find it relevant to broadcast information related with the symptomatology of BD conditions (eg, symptom-disease evidence and symptom relief therapies) [60].

The BD community is also sharing information about diet-based treatments and shows particular interest in gluten-free dietary interventions [66] and probiotics [67] (both with a 7.32%, 611/8342 and 544/8342 of the mentions to food and diet terms). Although the characteristics of gluten sensitivity in IBD remain unclear, gluten is known to generate peptides that can alter intestinal permeability and affect the immune system [68]. Recent studies investigated the effects of a wheat gluten–containing diet on the evolution of sodium dextran sulfate–induced colitis [69] and evaluated the usefulness of a low fermentable oligosaccharides, disaccharides, monosaccharides, and polyols diet on patients with IBS, nonactive IBD, and celiac disease compared with a gluten-free diet [70]. Likewise, some prebiotics, such as germinated barley foodstuff, psyllium, or oligofructose-enriched inulin, might provide some benefit in patients with active ulcerative colitis or ulcerative colitis in remission [71]. Other studies suggest that the VSL#3 probiotic may be effective in inducing remission in active ulcerative colitis [72]. The high number of favorites, retweets, and hashtags reflect the interest of the BD community in knowing more about these dietary interventions and discuss the pros and cons of different diets and foods.

In general, drugs and nondiet-based treatments are less mentioned than the other categories. However, medical cannabis (36.38%, 760/2089 of the mentions to drug terms) and hypnotherapy (3.57%, 174/4864 of the mentions to treatment terms) raise some attention as an alternative, nonconventional treatments. Although these results are still inconclusive, recent studies show improvements in some BD-related symptoms. Medical cannabis is being tested for the treatment of gastrointestinal disorders such as abdominal pain and diarrhea. Experimental tests show that single ingredients from cannabis, such as tetrahydrocannabinol and cannabidiol, are responsible for these effects [73,74]. Conversely, the use of hypnotherapy is being validated in the treatment of gastrointestinal symptoms such as reducing fasting distal colonic motility or reducing systemic and rectal mucosal inflammatory responses [75]. Although, the results of using these alternative therapies remain unclear. As it stands, users show greater interest for cannabis-related posts, as denoted by the high number of tweet favorites and retweets as well as the inclusion of hashtags. In addition, users find these tweets interesting as they choose to disseminate them among their followers and make the topic indexation easier.

A closer look into the tweets posted by BD patients was relevant to better understand how patients feel about and deal with BD symptoms, associated diseases, changes in habits, and medication. This analysis considered only the tweets posted by user accounts identified as patients (58.63%, 7795/13,295 of the total of user accounts). Moreover, it was possible to identify 2800 females (35.92%, 2800/7795 of the user accounts identified as patients) and 3860 males (49.51%, 3860/7795 of the user accounts identified as patients). Typically, the 11,098 tweets (45.05%, 11,098/24,634 of the total of tweets) posted by patients expressed a negative sentiment (51.99%, 5770/11,098), but there were also positive tweets (31.99%, 3551/11,098) and some tweets with a neutral sentiment (13.99%, 1553/11,098).

The high amount of negative opinions is explained by the fact that patients are known to use social platforms to vent out their emotions, namely, their frustrations when it comes to diseases that do not have a cure, such as in the case of IBD [76,77]. Table 2 describes the tweets of patients in terms of the recognized semantics terms and the tweet sentiment.

An important part of the tweets expressing positive emotions was related to gluten-free diets and probiotic supplements. In this line, there were more positive tweets posted by females than males (57.98%, 2274/3922 of the female tweets against 51.99%, 2390/4597 of the male tweets). However, in general, the tweets containing mentions to foods and diets (Figure 7) showed a similar distribution in terms of negative and positive sentiments. The main reason was that the effects of certain food components (eg, fiber, iron, and magnesium) were perceived differently, depending on the state of the disease and other personal conditions of the patient.

Disease and symptom (Figure 7) were the semantic categories with the highest number of mentions in negative tweets. These tweets discussed the bad side of BD conditions, that is, symptoms such as pain, fatigue, and migraines along with the co-occurrence of other conditions, namely, depression, diarrhea, and fibromyalgia [64,78]. In contrast, several tweets containing mentions of drugs (Figure 7) showed positive emotions, namely, the tweets highlighting the benefits of medical cannabis for the treatment of different chronic diseases.

As a means to look into these tweets from another perspective, Figure 8 depicts a subgraph of the semantic co-occurrence network reconstructed from the patient tweets (see details on the topological analysis of the complete network in Multimedia Appendix 2). This subgraph shows the relations between diseases and drugs. The size of the nodes is based on the node degree (ie, bigger nodes represent the terms mentioned in more tweets), whereas the edge size expresses the strength of co-occurrence (ie, thicker edges represent a higher number of term-term occurrences). Red nodes represent drugs and yellow nodes represent diseases, whereas the edge color stands for the tweet sentiment (ie, red indicates a majority of negative tweets, green indicates a majority of negative tweets, and black indicates neutral sentiment).

Although medical cannabis (and its components such as cannabidiol) was the most discussed drug, it was also possible to track down patient discussions about commercial drugs such as l-glutamine, lactulose, loperamide, and Plantago seed. For example, patients reported the positive effect of l-glutamine on diarrhea and muscular atrophy, but they also expressed their concern with glutamate leading to the occurrence of anxiety disorders. Glutamine is used to protect the mucous membrane of the esophagus and intestines and can boost immune cell activity in the gut [79,80]. Conversely, glutamine helps to create gamma-aminobutyric acid, a neurotransmitter that can stable the mind but can also produce glutamate, an excitatory neurotransmitter that can overstimulate the brain [81].

Patients discussed the use of laxatives in treating constipation, but not all of the mentioned laxatives were recommended. Notably, lactulose is a sugar that cannot be digested in the gut and thus, tends to cause or aggravate IBS symptoms, such as gas, bloating, discomfort, and cramping [82]. In turn, loperamide was suggested as an effective astringent to treat diarrhea and constipation. Indeed, a previous study reported a significant improvement in stool frequency and consistency [83].

Finally, patients were interested in the beneficial healing properties of medicinal plants such as Plantago (in various forms, such as roasted seeds, decoction, or syrup), namely, anti-inflammatory, laxative, and astringent properties [84].

Table 3 describes the number of tweets and the volume of tweet interactions (ie, retweets and favorites) from February 1, 2018, to August 31, 2018. It was interesting to identify the periods of time when users are more likely to tweet and, in particular, how specific events, such as the IBD day, a scientific conference, or an informational campaign, could affect such activity.

In particular, the celebration of the World IBD Day (on May 19, 2018), which is a worldwide event to raise awareness about BD conditions and to urge governments and health care professionals to take action and show support to the sufferers, motivated an increase in tweet interactions, that is, retweets and favorites (ie, 15,820 and 18,158 tweet interactions in May and June 2018, respectively). The average number of retweets during these months increased by 174% (3314/1902) compared with the average number of retweets during the rest of the year. Regarding tweet favorites, the increment was still noticeable (329%, 13,585/4130) compared with the average number of favorites during the rest of the year.

The hyperlinks shared via tweet provided useful information about current health research and development initiatives (public and private), health promotion actions, and other events that are promoted via Twitter as means to reach out to and engage more people. Tables 4 and 5 summarize the results obtained by grouping the URLs into 4 categories: (1) informative or awareness campaigns about BD; (2) scientific sites, articles, and conferences, usually covering new treatments; (3) health Web pages related to a specific disease; and (4) commercial sites. The #Tweets column indicates the number of tweets in which the URL is mentioned, the #Retweets column indicates the number of times that a tweet containing the URL was retweeted, and the #Favorites column indicates the number of times that a tweet containing the URL was selected as a favorite.

Only a small number of external links included in tweets posted by organizations got highly retweeted and labeled as favorite. Notably, the Guts4life Web page (a portal about IBD) got the highest number of retweets and favorites. This sort of analysis can be of aid in identifying the external sources that the community finds most useful/interesting, especially considering that most of them are related to pages describing BD symptomatology and potential treatments.

Regarding the external links included in the tweets shared by patients, their information flow was in the BD community at the same level compared with that of links posted by organizations (an average of 49 retweets against an average of 49 retweets, respectively). In turn, the links shared by patients had a lower average number of favorites compared with those shared by organizations (ie, an average of 44 favorites against an average of 92 favorites, respectively).

Looking into the linked contents, most of the resources were related to posted articles that belonged to highly prestigious journals, namely, Nature and British Medical Journal, and reported recent research in BD topics (with no particular focus). The presence of commercial links to pages selling stoner- and other drug-related products that do not require a medical prescription was also noteworthy.

The objective of this paper was to characterize and study the BD community on Twitter. To do so, a dataset of tweets related to BD was collected, processed, and analyzed. The dataset covered a consecutive period of 8 months, from February 1, 2018, to August 31, 2018. As a whole, this analysis provided new insights into 6 main questions: Which are the main communities and most influential users?; Where are the main content providers from?; What are the key biomedical and scientific topics under discussion? How are topics interrelated in patient communications?; How do external events influence user activity?; What kind of external sources of information are being promoted?

Health organizations and BD experts (eg, @CrohnsColitisUK and @IBDMD) were the users that received more tweets, typically looking for trusted information about the conditions. Patients shared experiences among themselves or asked for medical advice. Moreover, the most active users were located in the United States and the United Kingdom, which are among the demographic regions with highest BD prevalence.

Most of the tweets talked about BD symptoms, related diseases, foods, and diets. Specifically, diarrhea, constipation, and pain were the symptoms that raised more concern (in general as well as among patients), whereas gluten and probiotics were among the most discussed dietary interventions (including a high number of favorites and retweets). In this line, females showed higher positive emotions about these dietary interventions than males. Medical cannabis was the most commented drug, notably in the tweets that raised the highest number of favorites, and patients actively discussed the beneficial effects of cannabis (and its components) in mitigating common BD symptoms. Regarding more commercial drugs, patients expressed positive emotions with the usage of l-glutamine on diarrhea and muscular atrophy but also reported negative sentiments because of the production of glutamate and its influence on anxiety disorders. Another notable group of drugs in the discussion were the laxatives, for example, the usage of lactulose was associated with negative emotions because it tends to cause or aggravate IBS symptoms, whereas loperamide was noted to be an effective astringent to treat diarrhea and constipation.

Users were more active during and after the World IBD Day, which shows that these types of initiatives are raising public awareness about these diseases as well as indicates that social networks are part of the routine communication of the BD community. The external resources being shared in tweets by organizations aim to draw people’s attention to awareness/informational sites, whereas those shared by patients typically point to more scientific contents (eg, scientific articles on BD) and alternative treatments (such as cannabis).

The most immediate limitation arises from the fact that the capture of raw data was carried out using the free Twitter API, that is, the identification of users that select one tweet as a favorite as well as the number of available tweets is restricted, with no assurance of a random or representative sample [105]. For this reason, it was not possible to perform a more exhaustive analysis of the social BD communities. Thus, a full data retrieval through automated dashboard vendors, or using a paid service of the Twitter API, may provide further insights.

It should also be noted that this study was based on the assumption that the data entered by Twitter users are true. It is not possible to detect if certain data (eg, the user profile picture or the user location) are reliable. This limitation impacts mainly the analysis of the demographic distribution and the conclusions inferred for a particular gender. That being said, the obtained results were in accordance with the common knowledge of these communities.

Language is another aspect of analysis to take into consideration. This study was only focused on tweets written in English. If the applied techniques are extended to support a greater variety of languages, such as Chinese or Spanish, it may provide complementary findings.

Finally, this study was focused on Twitter. However, the set of social networks might be expanded to analyze a richer dataset from a wide variety of sources. Considering other studies in the literature [21,106,107], Facebook and Instagram would be also sources of interest, although public data access is greatly limited.

In this study, tweets related to BD were analyzed to characterize the user community and the exchanged contents. According to the obtained results, it was possible to detect communities and to describe the most discussed topics among these communities. The large and increasing volume of tweets demonstrates that Twitter is becoming a space for online conversation about BD, namely, associated symptoms and alternative treatments. In addition, the location of users indicates that conversations are happening at a global scale and, motivated by this, health-related stakeholders are using the platform to reach out to a larger audience on a daily basis.

In terms of future research, it would be interesting to perform user classification, that is, being able to identify experts, researchers, and companies, as well as patients. Thus, it would be possible to apply different sentiment analysis and TM approaches to the tweets to explore user-specific motivations, questions, and concerns. For example, it would be interesting to discover the opinion of patients about different treatments and specific symptoms.