Assessment of Antipsychotic Medications on Social Media: Machine Learning Study

Background: Antipsychotic medications are the first-line treatment for schizophrenia. However, non-adherence is frequent despite its negative impact on the course of the illness. In response, we aimed to investigate social media posts about antipsychotics to better understand the online environment in this regard.

Methods: We collected tweets containing mentions of antipsychotic medications posted between January 1st 2019 and October 31st 2020. The content of each tweet and the characteristics of the users were analyzed as well as the number of retweets and likes generated.

Results: Twitter users, especially those identified as patients, showed an interest in antipsychotic medications, mainly focusing on the topics of sexual dysfunction and sedation. Interestingly, paliperidone, despite being among one of the newest antipsychotics, accounted for a low number of tweets and did not generate much interest. Conversely, retweet and like ratios were higher in those tweets asking for or offering help, in those posted by institutions and in those mentioning cognitive complaints. Moreover, health professionals did not have a strong presence in tweet postings, nor did medical institutions. Finally, trivialization was frequently observed.

Conclusion: This analysis of tweets about antipsychotic medications provides insights into experiences and opinions related to this treatment. Twitter user perspectives therefore constitute a valuable input that may help to improve clinicians' knowledge of antipsychotic medications and their communication with patients regarding this treatment.

Introduction

Schizophrenia is a psychiatric disorder characterized by a multidimensional psychopathology that includes positive, negative and mood symptoms, as well as cognitive impairment. Additionally, it is commonly associated with impairments in social and occupational functioning (1). Schizophrenia and psychotic disorders are among the world's leading causes of disability (2). However, evidence-based psychosocial interventions, in conjunction with pharmacotherapy, can help patients achieve recovery (3, 4).

Antipsychotic medications are the first-line treatment for psychosis-related disorders and have been shown to be effective in treating associated symptoms and behaviors (5). First- and second-generation antipsychotic medications are comparable in clinical efficacy, with the exception of clozapine, which demonstrates the best results toward treatment-resistant schizophrenia (6, 7). However, they differ from one another in other characteristics such as dosing, the route of administration, the side effects and the cost (8). All these factors influence the selection of the antipsychotic drug to be used, as well as adherence to a medication regimen, ultimately affecting a medication's effectiveness. For example, weight gain, diabetes, and dyslipidemia are more often associated with second-generation antipsychotics (SGAs), while first-generation antipsychotics (FGAs) generally have a greater risk of extrapyramidal side effects (8, 9).

In patients with a poor response to medication or repeated relapses, non-adherence should be considered (10, 11). Side effects, along with a limited awareness of one's schizophrenia and the need for treatment, are frequently the primary cause of non-adherence, and thus need to be addressed (3). However, many patients hide these issues from their doctor, with the reasons for doing so poorly understood yet nevertheless significant, such as the feeling of shame in disclosing these aspects or the fear of being judged (11).

Traditional research on patients' experiences to treatment has principally relied on surveys or interviews (12–14). However, the analysis of social media posts has emerged as an important tool capable of gathering a more comprehensive understanding of those factors involved in the therapeutic process and in the experience of the disease (15, 16). It also allows for greater insights into the opinions of others apart from just the patients themselves, thereby providing additional data from a wider range of perspectives, including those patients that are reluctant to visit a medical professional (15–17). In addition, social media conversations are generated within a more casual and spontaneous environment than those that take place during a medical appointment; thus, they may be more likely to reflect true beliefs (18–20).

Furthermore, numerous studies have also demonstrated that individuals living with schizophrenia use popular social media platforms at comparable rates to the general population (21). In addition, individuals with mental illness appear to use social media to share their illness experiences or seek advice from others with similar conditions (21). However, less is known about whether people with psychosis talk about antipsychotic medication experiences over social media. Thus, our aims were to (1) investigate the frequency of online communications about antipsychotic medications among Twitter users; (2) characterize the type of users participating in these conversations; (3) determine the main thematic content of Twitter posts and the interest they generated; and (4) analyze references to specific antipsychotic medications.

Methods

Search Strategy and Collection of Twitter Data

In this observational quantitative and qualitative study, we focused on searching for tweets that referred to antipsychotics. We collected all posted tweets using the following list of keywords: amisulpirida, solian, haloperidol, haldol, pimozida, orap, sulpirida, dolmatil, flufenacina, modecate, sulpirirda, dogmatil, clorpromacian, largactil, levomepromazine, sinogan, aripiprazol, abilify, clozapine, leponex, nemea, clozaril, olanzapina, zyprexa, paliperidone, invega, quetipiana, seroquel, risperidone, risperdal, brexpiprazole, rexulti, loxapina, adasuve, lurasidone, latuda, ziprasidone, geodon, and zeldox. We selected keywords according to the most frequently prescribed antipsychotics in Europe and the United States. We have included both generic names and brand names in the selection. We grouped the keywords into 13 categories, with each category corresponding to a different antipsychotic medication, thus allowing for greater ease when comparing one medication with another. Antipsychotic medications were referred to by either their generic name or by their brand name.

The inclusion criteria for tweets were: (1) Being public; (2) Containing any of the previously mentioned keywords; (3) Being posted between January 1st 2019 and October 31st 2020 and; (4) Containing text in English or Spanish. A 22 month period was chosen to avoid any potential bias in the content of the tweets. For instance, we wanted to be sure to prevent content from being affected by the season of the year, a particular event, or a special circumstance (publication of any relevant scientific article related to any of the antipsychotic medications being analyzed, the mention of or a relationship to any medical, psychiatric or pharmacological conferences, etc.). In addition, we obtained the number of retweets and likes each tweet generated as an indicator of user interest on a given topic, the date and time of each tweet, a permanent link to the tweet and each user's profile description. Tweet Binder, the search engine we employed, allows access to 100% of all public tweets that match certain criteria.

Content Analysis Process

All 30,603 retrieved tweets were included in the dataset. First, we randomly selected 1,500 of the tweets to be considered for content analysis. Secondly, we created a codebook based on our research questions, our previous experience in analyzing tweets, and what we determined to be the most common tweet themes (22–25). Third, JG and LA analyzed 300 tweets separately to test the suitability of the codebook. Discrepancies were discussed between the raters and with another two authors (MAAM and MAM), and after revising the codebook the raters then proceeded to manually code the remaining 1,200 tweets.

Tweets were categorized as classifiable or unclassifiable. We considered a tweet as non-classifiable when its content did not provide enough information or if it was written in a language other than English or Spanish. Each of the tweets considered classifiable was rated according to the area of clinical interest mentioned or discussed in the text of the tweet, such as quality of life, mood and anxiety, sedation, metabolic disturbances and extrapyramidal symptoms, sexual dysfunction, and cognitive complaints. In addition, tweets with non-medical content were rated as commercial, economic, ask/offer help, trivialization, or non-specific. We classified as trivialization those tweets that included mockery or joking, associated treatment with undesirable attributes or associated treatment with grossly inaccurate stereotypes. Moreover, the following characteristics were also analyzed: if the tweet included a link to a health care provider (either a hospital, health institution, university, or pharmaceutical company), if it mentioned a scientific article or specific aspects of posology, or if it expressed a personal opinion, mentioned a famous person or stated the use of an antipsychotic to treat a particular disorder. Finally, users were classified into 5 categories: patient, relative or friend of a patient, health professional, health institution, and Twitter-user interaction (when at least two different users participated in a conversation but were classified into different categories). We determined the nature of the users that posted tweets according to the information available: content of the tweet (use of pronouns, disclosure of personal or family experiences, professional information provided, etc.), user profile description or Twitter handle (this section was especially informative in the case of health institutions since the majority refer to their social media accounts in their profile description or their Twitter handle). In those cases in which the nature of the user was not possible to know, they were considered as indeterminate.

The coding categories used were not mutually exclusive. In the case of finding content that was repeated exactly or posted almost identically in different tweets, those postings were classified in the same way as the first tweet encountered.

Multilingual Machine Learning Classifier

The goal of the initial tagging of 1,500 tweets was to provide data to train, test, and validate Machine Learning classifiers so that any extracted tweet classifications could be inferred. To train the classifiers, the transformer multilingual model xlm-roberta neural net was used in conjunction with a “k-train” library to deploy it (26, 27). The following additional features were generated to improve understanding of the selected set: the number of tokens that the sentence contained, the total length of the tweet in terms of characters, the language of the tweet and the extracted hashtags from the tweet. Additionally, on order to improve the Machine Learning classifier performance, we generated a clean text that took any mentions (@) and hyperlinks out of the tweet so that it became more readable.

Out of the 1,500 manually labeled tweets, we reserved 10% to use as a blind set for model validation so that the setup we employed for training the classifier constituted 80% training and 20% validation. Initially, training was done on the classifiable feature and then out of those tweets that were labeled classifiable, the remaining classifications were trained. The weighted average F1 score of the training validation against the blind dataset was above 0.80 in all cases except for the User and Interest categorization, which was slightly lower.

These analyses were performed using Python 3.7 and the libraries “pandas,” “numpy,” “json,” and “k-train.”

Ethical Considerations

This study received the approval of the University of Alcala Research Ethics Committee and was compliant with the research ethics principles of the Declaration of Helsinki (7th revision, 2013). However, this study did not directly involve human subjects nor include any intervention but instead used only publicly available tweets. Nevertheless, we have taken care not to reveal any usernames and to avoid citing any tweets that could reveal them.

Statistical Analysis

The frequency distribution (percentage) of tweets, retweets, and likes according to certain characteristics of each tweet, such as the content included or the antipsychotic drug mentioned, were displayed across several figures and tables. Because the sample of tweets corresponds to all tweets from the selected period, and are not just a representative sample of these tweets, the causal inference p-values do not apply. The accuracy of the different tweet distributions obtained from the Multilingual Machine Learning models is reflected by the weighted average F1 score (a combination of precision and recall; the closer the score is to 1, the less possibility of classification error). As well, retweet-to-tweet and like-to-tweets ratios according to user type, the area of clinical interest and non-medical aspects were calculated.

According to the content of the tweet, we used simple logistic regressions to calculate the probability of retweeting or liking a tweet. These results were presented as an odds ratio (OR) with 95% Confidence Intervals (CI). These analyses were conducted with the software packages STATA v16 (StataCorp) and MS Excel.

Results

Sedation and Sexual Dysfunction Are the Most Common Areas of Clinical Interest in Antipsychotic Related Tweets

For this work, we collected 30,603 tweets including the keywords antipsychotic medications from January 1st 2019 to October 31st 2020. According to the inclusion criteria of the codebook, a total of 22,092 tweets were considered classifiable.

First, we analyzed the content of those tweets referencing the specific areas of clinical interest studied (Figure 1A), with 5,415 out of the 22,092 tweets analyzed featuring related content. There was a significant difference in distribution (P < 0.05) of the number of tweets between the different categories studied, the predominant majority being related to sexual dysfunction and sedation. Additionally, there was a lower frequency of tweets related to cognitive complaints and mood or anxiety. Those related to metabolic disturbances or extrapyramidal symptoms represented the minority of tweets.

FIGURE 1

Figure 1. Different percentages (%) of tweets according to areas of clinical interest (A) and non-medical aspects (B).

We also analyzed the content of those tweets including no medical aspects of antipsychotic medications (8,517 tweets) and we classified them according to four categories: commercial activities, economic issues, ask for or offer help and trivialization) (Figure 1B). We found a significant distinction in the number of tweets between these four categories with the highest frequency of postings related to drug trivialization followed by those asking for or offering help (P < 0.05). Those related to commercial activities and economic issues represented the minority of tweets.

Further investigating certain tweet characteristics, we found that 18.87% of the 22,092 tweets studied included a link to a health care provider (either a hospital, health institution, university, or pharmaceutical company), 10.55% referenced a scientific article and 10.44% were related to specific aspects of posology. Moreover, 14.1% of the tweets included a personal opinion while 6.4% mentioned a famous person. Lastly, 17.7% mentioned the use of antipsychotics for treating a particular psychiatric disorder.

Patients Are the Most Active Twitter Users With a Differential Pattern of Areas of Interest in Antipsychotic Related Conversations

From investigating the types of users that posted tweets, 7,188 tweets were posted by users identified as patients (72.1%), health institutions (17.2%), patients' friends or relatives (7.0%), and health care professionals (3.7%). Seven thousand six hundred and fifty-five tweets were interactions between two or more Twitter users. From the remaining 4,465 tweets, the users were considered unclassifiable.

Subsequently, we investigated areas of clinical interest in those tweets posted by different types of users, finding significant differences between them (P < 0.05) (Figure 2). Patients, for example, focused the content of their tweets on the specific clinical aspects of antipsychotic medications, specifically sexual dysfunction and sedation. In contrast, almost all of the tweets posted by health institutions concerned non-specific issues. However, the distribution of postings related to areas of clinical interest stemming from patients' friends or relatives, health professionals and Twitter interactions was similar, with a predominance centered on cognitive complaints.

FIGURE 2

Figure 2. Areas of clinical interest related to antipsychotic medications by the type of user posting. Percentages (%) were calculated with respect to the total number of tweets posted by each type of user.

Tweets Asking for or Offering Help, Referring to Cognitive Impairment or Being Posted by Health Institutions Generated the Greatest Interest

We investigated the interest generated by those tweets related to antipsychotic medications by quantifying the number of retweets and likes generated by each tweet. In doing so, we found that the probabilities of a tweet being retweeted or liked were distinct between the different categories of users. Tweets posted by health institutions accumulated the highest median of retweets per tweet, which was three times higher than that observed in tweets posted by health care professionals or patients' relatives and friends (Table 1). Not insignificantly, those tweets posted by patients obtained the lowest median of retweets per tweet. By contrast, when we analyzed the probability of tweets receiving likes, we found a noticeable similarity among the different types of users. Nevertheless, tweets posted by health professionals obtained twice as many likes as those posted by patients, health institutions and patients' relatives (Table 1).

TABLE 1

Table 1. Retweet-to-tweet and like-to-tweet ratios per type of user.

Furthermore, we investigated the number of likes and retweets generated by each area of clinical interest (Table 2). Tweets related to cognitive complaints were clearly those with the highest ratios of being retweeted and liked, followed by those referencing mood and anxiety and sedation. Finally, those tweets concerned with quality of life, metabolic disturbances and extrapyramidal symptoms, and sexual dysfunction had a similar average number of retweets and likes per tweet. Additionally, we investigated the probability of a tweet of being retweeted or liked depending on non-medical aspects. We found that tweets asking for or offering help were clearly retweeted more than those either referring to commercial or economic issues and trivialization toward a certain drug. That being said, tweets related to drug trivialization still received the most likes (Table 3).

TABLE 2

Table 2. Retweet-to-tweet and like-to-tweet ratios per area of clinical interest.

TABLE 3

Table 3. Retweet-to-tweet and like-to-tweet ratios per non-medical aspects.

Finally, we assessed which characteristics were more associated with engagement (Figure 3). We found that those tweets including a link to a health care provider had almost three times greater odds of being retweeted than those tweets that did not include such a link. As well, tweets that referred to a psychiatric diagnosis had a greater probability of being retweeted than those tweets that did not. On the other hand, tweets that mentioned a famous person or specific aspects of posology, or that expressed a personal opinion, had lower probabilities of being retweeted. By contrast, the probabilities of a tweet being liked maintained a different pattern even though most of the associated characteristics being analyzed had a modest effect on the probability of a tweet being retweeted or liked.

FIGURE 3

Figure 3. ORs and 95% CIs for retweeting (A) and liking (B) a tweet according to specific characteristics.

The Frequency of Tweets Related to Antipsychotic Medications Was Heterogeneous

Of the 22,092 tweets included in our research, we investigated those specifically related to each of the 13 antipsychotic medications studied. The number of mentions of each drug followed a heterogeneous pattern of distribution, ranging from 14.3% for olanzapine to 1.2% for sulpiride (Figure 4). Of note, aripiprazole and paliperidone, which are among the newest antipsychotic medications, only accounted for 15.0% of the total number of tweets.

FIGURE 4

Figure 4. Distribution of tweets by antipsychotic medications.

We also analyzed the type of user who posted tweets related to antipsychotic medications. For example, the percentage of tweets posted by users identified as patients was greater in lurasidone (57.0%), aripiprazole (56.6%), quetiapine (51.7%), brexpiprazole (43.6%), and paliperidone (41.0%) (Figure 5). The percentage of tweets posted by users identified as friends and relatives of patients or by health professionals was low overall, with the exception of clozapine (10.0 and 6.6%, respectively). Tweets posted by health institutions held higher percentage totals when related to sulpiride (59.6%) and levomepromazine (54.5%) but constituted a total below 10% when referencing all other drugs. Lastly, Twitter user interaction represented the category with the highest percentage of tweets related to haloperidol (58.0%), chlorpromazine (52.2%), risperidone (43.6%), or olanzapine (41.8%).

FIGURE 5

Figure 5. Tweets related to each antipsychotic medication by the type of user posting. Percentages (%) were calculated with respect to the total number of tweets generated on each antipsychotic medication.

Most Recent Medications Did Not Attract Much Attention Among Twitter Users

We measured the number of retweets and likes generated by those tweets related to each of the antipsychotic medications studied. We found that the retweet-to-tweet and like-to-tweet ratios between the different drugs were distinct (Figure 6). For instance, levomepromazine was the drug with the highest retweet-to-tweet ratio (13.3), whereas haloperidol and aripiprazole were the drugs with the highest like-to-tweet ratios (9.8 and 9.4, respectively). Moreover, we found that brexpiprazole, lurasidone, paliperidone, quetiapine, and ziprasidone generated a retweet-to-tweet ratio below 1.

FIGURE 6

Figure 6. Retweet-to-tweet and like-to-tweets ratios per each antipsychotic medication.

We also analyzed the distribution of tweets referencing an antipsychotic medication according to areas of clinical interest (Figure 7). Almost 60% of the tweets referring to quality of life mentioned aripiprazole or lurasidone, whereas none of the tweets mentioning sulpiride, paliperidone, levomepromazine, clozapine, chlorpromazine, or brexpiprazole addressed issues related to quality of life. In regards to mood and anxiety, we found that this type of content was minimal in tweets mentioning chlorpromazine, clozapine, haloperidol, levomepromazine, paliperidone, or sulpiride. A similar pattern was observed in tweets discussing issues related to sedation. Interestingly, tweets mentioning metabolic disturbances and extrapyramidal symptoms had a similar pattern of distribution to those tweets discussing sexual dysfunction, with aripiprazole, lurasidone, and quetiapine accumulating most of the tweets. Finally, those tweets referencing issues related to cognitive impairment were also distributed heterogeneously among the different drugs; in this case, haloperidol, olanzapine, and risperidone received the majority of tweets.

FIGURE 7

Figure 7. Tweets related to each antipsychotic medication by areas of clinical interest. Percentages (%) were calculated with respect to the total number of tweets generated on each antipsychotic medication.

Discussion

In this study, we have found that Twitter users show a great interest in antipsychotic medications focused on sexual dysfunction and sedation, especially those users identifying themselves as patients. Tweets asking for or offering help and those posted by institutions obtained the highest retweet-to-tweet ratio, whereas tweets discussing cognitive complaints obtained the highest like-to-tweet ratio. Moreover, health professionals did not command a strong presence in driving Twitter conversations with regards to antipsychotics, nor did medical institutions. In addition, tweets referring to the trivialization of certain medications was frequently observed. Interestingly, paliperidone, despite being among one of the most recent drugs launched on the market, nevertheless has failed to generate much interest among Twitter users, accounting for only a small number of tweets.

Furthermore, the medical treatment of patients with chronic diseases has become an increasing challenge to society, with several factors involved in patient outcomes (28). These are the efficacy and side effects of medications, access to medical information and social considerations toward a particular disease and its treatment, all of which serve to influence patients' attitudes in regards to treatment (29, 30). Thus, identifying both patients' and the public's areas of concern about diseases and their treatment in general, along with mental health conditions in particular, is extremely relevant for improving clinical outcomes (31). Correspondingly, the analysis of social media platforms such as Twitter is a recognized tool to explore societal opinions of chronic health conditions (32). Unfortunately, schizophrenia and psychosis-related disorders are frequently mischaracterized pejoratively over Twitter (33–36). In fact, a previous study found that school shooting was among the hashtags most frequently associated with psychosis, which shows that there are still many people who mistakenly associate psychosis with aggressiveness or violence (31). This association of negative hashtags does not occur in other physical illnesses such as breast cancer or diabetes, nor in other mental illnesses such as Hikikomori (37). Furthermore, in this study, we have found that antipsychotic medications are often trivialized in tweets. Collectively, Twitter data shows that patients don't just suffer from psychosis as antipsychotic treatments are also the targets of negative feelings and judgments by Twitter users, further demonstrating the persistence of social stigma toward schizophrenia and its treatment (38–41). This experience of stigmatizing attitudes against people with schizophrenia may as a result be internalized, leading to “self-stigma” that contributes to poor patient adherence to treatment (42, 43).

A comparative analysis of the prevalence of psychosis and the number of tweets posted related to antipsychotic medications, contrasted with those referencing treatment received for other chronic diseases, supports the notion of considerable interest in antipsychotic medications among the Twitter community (44). Interestingly, our results show that patients with schizophrenia are the most common users posting content about antipsychotic medications. These results contrast with the findings reported in research describing other medical diseases mentioned on Twitter (45). Several reasons may explain this higher use of Twitter by people living with schizophrenia. First, the anonymity provided by this social media platform may favor use by persons suffering from stigmatization. Secondly, the difficulty in personal communication that these patients exhibit is mitigated through their online interactions. Third, easy accessibility to Twitter and facility in posting content may also be involved in the use of this platform by people living with schizophrenia who otherwise might not possess the willingness and confidence to engage with others in person.

In addition, social media appears to be especially valuable for monitoring the treatment experiences of individuals with psychotic related disorders who frequently use online platforms to engage with people facing similar struggles (46, 47). Our data shows that patients posted tweets related to specific areas of clinical interest stemming from antipsychotic treatment, namely sedation, sexual dysfunction, mood, and anxiety. In fact, sexual dysfunction is common with antipsychotic treatment although its frequency varies among studies. For example, a 2011 meta-analysis found that around half of those patients taking clozapine, olanzapine, and risperidone reported sexual side effects, while those using aripiprazole, quetiapine, and ziprasidone experienced lower rates (48). However, other clinical trials have reported even lower frequencies of sexual dysfunction. These differences may be due in part to patients' reluctance to report sexual side effects during in-person interviews (49). In contrast, from our Twitter data, sexual dysfunction was one of the most discussed topics, which reinforces the notion of the platform as a resource that, as a result of its particular characteristics, can better serve to facilitate patients' disclosure of sensitive, potentially embarrassing information. The meta-analysis indicated above also found that among patients who experienced sexual dysfunction secondary to an antipsychotic medication, switching to aripiprazole was found to be the most common solution (50). Interestingly, aripiprazole was the antipsychotic that accounted for the most tweets related to sexual dysfunction.

Patients with schizophrenia have higher rates of depressive and anxiety disorders than the general population (51). Noticeably, this issue has been discussed more frequently by the patients themselves than by relatives or friends, health professionals and health institutions. This is probably due to the fact that patients are the ones most aware of these particular symptoms. On the other hand, we found that patients posted less about cognitive complaints than did their relatives or friends, suggesting that patients suffering from cognitive deterioration are often less aware of their condition and less prone to using Twitter. These findings also point to patients' psychopathologies influencing the type of content they publish over social media (52, 53).

Furthermore, the relatively small number of tweets concerning metabolic disturbances posted by patients with schizophrenia is worth highlighting. Psychotic disorders are often associated with altered glucose homeostasis, insulin resistance, hyperlipidemia, and hypertension (54, 55). This is notable because resultant cardiovascular diseases are largely responsible for the shorter life expectancy of people with schizophrenia, which is reduced by more than a decade when compared with that of the general population (54, 55). Our observation of the limited interest in metabolic disturbances shown by patients is consistent with previous studies reporting that people with schizophrenia often possess other risk factors for cardiovascular disease, in addition to the use of antipsychotic medications, including a sedentary lifestyle, a poor diet and smoking (56, 57). Thus, our results emphasize an insufficient awareness of healthy habits by these patients (58). Additionally, it is troubling that neither health professionals nor health institutions have dedicated much attention to the promotion of cardiovascular health, despite metabolic disturbances being responsible for the excessive mortality of these patients (59).

Among the Twitter community, we also investigated the interest generated by tweets related to specific antipsychotic medications. We used the number of retweets and likes generated by each tweet as our standard of measure (37). Our results showed that, with regards to user type, health institutions obtained the highest retweet-to-tweet ratio. Being that they are usually very influential in the medical field, health institutions tend to have a greater number of followers than do personal accounts (60). This trend reflects the importance of having health institutions and professionals involved in health conversations since patient education is the first step in treatment effectiveness, a fact that Twitter users appear to appreciate. In regard to content, those tweets asking for or offering help were the ones most retweeted, further reinforcing the notion of Twitter serving primarily as a community of users (61). Finally, concerning specific medications, the ones that generated the most interest were aripiprazole, levomepromazine, and haloperidol. These results perhaps may be explained by the fact that Twitter is an international platform, as evidenced by the high use of aripiprazole in Europe and the United States, while levomepromazine and haloperidol are commonly employed in other parts of the world for the treatment of schizophrenia. In fact, a majority of tweets, most notably from South America, asked for assistance in acquiring these medications. Probably for their connection to treatment, levomepromazine, and haloperidol were among the two most retweeted and liked medications.

Previous research has suggested that it may be in the best interest of health care providers and the pharmaceutical industry to focus on disease diagnosis and treatment (62, 63). Additionally, it has been noted that some companies have been especially active in promoting the benefits of their products over social media (64, 65). In fact, the food and beverage industries have been increasingly promoting their brands over social media platforms, as well as utilizing posts to advertise them (66). However, according to our results, only a small percentage of the tweets studied related to antipsychotic medications promoted commercial activities. Moreover, paliperidone, despite being one of the latest drugs launched on the market, accounted for only a small number of tweets. As well, the content that obtained the highest retweet-to-tweet ratio was that related to either the asking for or offering of help. These results suggest that Twitter is frequented by users participating in conversations related to antipsychotic medications for the primary purpose of seeking out information and support rather than looking for commercial opportunities.

Limitations

This study has some limitations. First, since Twitter users tend to be younger than the general population, our findings may not pertain to older age ranges (67, 68). Indeed, younger patients in the earlier stages of psychotic disorders are generally more active, suffer less cognitive impairment and are more likely to engage over social media. Secondly, we were unable to examine how clinical characteristics such as symptom severity, illness duration or cognitive dysfunction found in individuals with psychosis influenced the content of their social media posts due to a lack of psychiatric evaluation. Third, the codebook design and text analysis we used imply a degree of subjectivity. However, this methodology is consistent with previous medical research studies on Twitter and could be applied to assorted topics by various authors (69–71). Fourth, among the list of keywords we included both generic names and brand names but those tweets that included spelling mistakes might have been left out.

Conclusions

Our results highlight the potential to leverage social media for a better understanding of patients suffering from schizophrenia and their treatment in a manner that is more comprehensive, one that not only includes health care providers and health institutions but also family members, friends, and other social media users. Although guidelines recommend assessing the efficacy of, adherence to and tolerability toward treatment during medical consultations, in the treatment of a patient with schizophrenia, one's individual environment beyond the confines of a doctor's office is extremely important, especially as it can frequently extend to include other influences, many of which are found on social media and are worth analyzing due to the broadened perspective they can offer.

Data Availability Statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics Statement

The studies involving human participants were reviewed and approved by the Research Ethics Committee of the University of Alcala (OE 14_2020). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

Author Contributions

MAA-M, JQ, and MÁ-M: conceptualization, validation, and resources. MAA-M, CD-V, JS-V, LA, RR-J, JQ, and MÁ-M: methodology and data curation. MAA-M, CD-V, JS-V, LA, JG, RS-B, FM, MO, GL, RR-J, JQ, and MÁ-M: formal analysis, investigation writing—original draft preparation, and writing—review and editing. JQ and MAA-M: supervision. MÁ-M: project administration and funding acquisition. All authors have read and agreed to the published version of the manuscript.

Funding

This work was partially supported by grants from the Fondo de Investigación de la Seguridad Social, the Instituto de Salud Carlos III (PI18/01726) (Spain), the Programa de Actividades de I+D de la Comunidad de Madrid en Biomedicina (B2017/BMD-3804), Madrid (Spain), and Helekulani SL.

Conflict of Interest

JS-V was employed by IBM. RR-J has been a consultant for, spoken at events of, or received grants from the Instituto de Salud Carlos III, the Fondo de Investigación Sanitaria (FIS), the Centro de Investigación Biomédica en Red de Salud Mental (CIBERSAM), the Madrid Regional Government (S2010/BMD-2422 AGES, S2017/BMD-3740), JanssenCilag, Lundbeck, Otsuka, Pfizer, Ferrer, Juste, Takeda, Exeltis, Casen-Recordati, and Angelini.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher's Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

1. Jiménez-López E, Sánchez-Morla EM, López-Villarreal A, Aparicio AI, Martínez-Vizcaíno V, Vieta E, et al. Neurocognition and functional outcome in patients with psychotic, non-psychotic bipolar I disorder, and schizophrenia. A five-year follow-up. Eur Psychiatry. (2019) 56:60–8. doi: 10.1016/j.eurpsy.2018.11.008

PubMed Abstract | CrossRef Full Text | Google Scholar

2. Global Burden of Disease Study 2013 Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet. (2015) 386:743–800. doi: 10.1016/S0140-6736(15)60692-4

PubMed Abstract | CrossRef Full Text | Google Scholar

3. Davarinejad O, Mohammadi Majd T, Golmohammadi F, Mohammadi P, Radmehr F, Alikhani M, et al. Identification of risk factors to predict the occurrences of relapses in individuals with schizophrenia spectrum disorder in Iran. Int J Environ Res Public Health. (2021) 18:546. doi: 10.3390/ijerph18020546

PubMed Abstract | CrossRef Full Text | Google Scholar

4. Leendertse JCP, Wierdsma AI, van den Berg D, Ruissen AM, Slade M, Castelein S, et al. Personal recovery in people with a psychotic disorder: a systematic review and meta-analysis of associated factors. Front Psychiatry. (2021) 12:622628. doi: 10.3389/fpsyt.2021.622628

PubMed Abstract | CrossRef Full Text | Google Scholar

5. Howes OD, Whitehurst T, Shatalina E, Townsend L, Onwordi EC, Mak TLA, et al. The clinical significance of duration of untreated psychosis: an umbrella review and random-effects meta-analysis. World Psychiatry. (2021) 20:75–95. doi: 10.1002/wps.20822

PubMed Abstract | CrossRef Full Text | Google Scholar

6. Okhuijsen-Pfeifer C, Huijsman EAH, Hasan A, Sommer IEC, Leucht S, Kahn RS, et al. Clozapine as a first- or second-line treatment in schizophrenia: a systematic review and meta-analysis. Acta Psychiatr Scand. (2018) 138:281–8. doi: 10.1111/acps.12954

PubMed Abstract | CrossRef Full Text | Google Scholar

7. Shah P, Iwata Y, Plitman E, Brown EE, Caravaggio F, Kim J, et al. The impact of delay in clozapine initiation on treatment outcomes in patients with treatment-resistant schizophrenia: a systematic review. Psychiatry Res. (2018) 268:114–22. doi: 10.1016/j.psychres.2018.06.070

PubMed Abstract | CrossRef Full Text | Google Scholar

8. Huhn M, Nikolakopoulou A, Schneider-Thoma J, Krause M, Samara M, Peter N, et al. Comparative efficacy and tolerability of 32 oral antipsychotics for the acute treatment of adults with multi-episode schizophrenia: a systematic review and network meta-analysis. Lancet. (2019) 394:939–51. doi: 10.1016/S0140-6736(19)31135-3

PubMed Abstract | CrossRef Full Text | Google Scholar

9. Carli M, Kolachalam S, Longoni B, Pintaudi A, Baldini M, Aringhieri S, et al. Atypical antipsychotics and metabolic syndrome: from molecular mechanisms to clinical differences. Pharmaceuticals. (2021) 14:238. doi: 10.3390/ph14030238

PubMed Abstract | CrossRef Full Text | Google Scholar

10. Fernández-Miranda JJ, Díaz-Fernández S, De Berardis D, López-Muñoz F. Paliperidone palmitate every three months (PP3M) 2-year treatment compliance, effectiveness and satisfaction compared with paliperidone palmitate-monthly (PP1M) in people with severe schizophrenia. J Clin Med. (2021) 10:1408. doi: 10.3390/jcm10071408

PubMed Abstract | CrossRef Full Text | Google Scholar

11. Yu W, Tong J, Sun X, Chen F, Zhang J, Pei Y, et al. Analysis of medication adherence and its influencing factors in patients with schizophrenia in the Chinese institutional environment. Int J Environ Res Public Health. (2021) 18:4746. doi: 10.3390/ijerph18094746

PubMed Abstract | CrossRef Full Text | Google Scholar

12. Mohammed MA, Moles RJ, Chen TF. Medication-related burden and patients' lived experience with medicine: a systematic review and metasynthesis of qualitative studies. BMJ Open. (2016) 6:e010035. doi: 10.1136/bmjopen-2015-010035

PubMed Abstract | CrossRef Full Text | Google Scholar

13. Haslam C. Patients' experiences of medication for anxiety and depression: effects on working life. Fam Pract. (2004) 21:204–12. doi: 10.1093/fampra/cmh218

PubMed Abstract | CrossRef Full Text | Google Scholar

14. Roe D, Goldblatt H, Baloush-Klienman V, Swarbrick M, Davidson L. Why and how people decide to stop taking prescribed psychiatric medication: exploring the subjective process of choice. Psychiatr Rehabil J. (2009) 33:38–46. doi: 10.2975/33.1.2009.38.46

PubMed Abstract | CrossRef Full Text | Google Scholar

15. Golder S, Bach M, O'Connor K, Gross R, Hennessy S, Gonzalez Hernandez G. Public perspectives on anti-diabetic drugs: exploratory analysis of twitter posts. JMIR Diabetes. (2021) 6:e24681. doi: 10.2196/24681

PubMed Abstract | CrossRef Full Text | Google Scholar

16. Farrar M, Lundt L, Franey E, Yonan C. Patient perspective of tardive dyskinesia: results from a social media listening study. BMC Psychiatry. (2021) 21:94. doi: 10.1186/s12888-021-03074-9

PubMed Abstract | CrossRef Full Text | Google Scholar

17. Golder S, O'Connor K, Hennessy S, Gross R, Gonzalez-Hernandez G. Assessment of beliefs and attitudes about statins posted on Twitter. JAMA Netw Open. (2020) 3:e208953. doi: 10.1001/jamanetworkopen.2020.8953

PubMed Abstract | CrossRef Full Text | Google Scholar

18. Branley DB, Covey J. Pro-ana versus pro-recovery: a content analytic comparison of social media users' communication about eating disorders on Twitter and Tumblr. Front Psychol. (2017) 8:1356. doi: 10.3389/fpsyg.2017.01356

PubMed Abstract | CrossRef Full Text | Google Scholar

19. Emmert-Streib F, Yli-Harja O, Dehmer M. Utilizing social media data for psychoanalysis to study human personality. Front Psychol. (2019) 10:2596. doi: 10.3389/fpsyg.2019.02596

PubMed Abstract | CrossRef Full Text | Google Scholar

20. Alvarez-Mon MA, de Anta L, Llavero-Valero M, Lahera G, Ortega MA, Soutullo C, et al. Areas of interest and attitudes towards the pharmacological treatment of attention deficit hyperactivity disorder: thematic and quantitative analysis using Twitter. J Clin Med. (2021) 10:2668. doi: 10.3390/jcm10122668

PubMed Abstract | CrossRef Full Text | Google Scholar

21. Hswen Y, Naslund JA, Brownstein JS, Hawkins JB. Monitoring online discussions about suicide among Twitter users with schizophrenia: exploratory study. JMIR Ment Heal. (2018) 5:e11483. doi: 10.2196/11483

PubMed Abstract | CrossRef Full Text | Google Scholar

22. Viguria I, Alvarez-Mon MA, Llavero-Valero M. Asunsolo del Barco A, Ortuño F, Alvarez-Mon M. Eating disorder awareness campaigns: thematic and quantitative analysis using Twitter. J Med Internet Res. (2020) 22:e17626. doi: 10.2196/17626

PubMed Abstract | CrossRef Full Text | Google Scholar

23. Abbasi-Perez A, Alvarez-Mon MA, Donat-Vargas C, Ortega MA, Monserrat J, Perez-Gomez A, et al. Analysis of tweets containing information related to rheumatological diseases on Twitter. J Environ Res Public Health. (2021) 18:9094. doi: 10.3390/ijerph18179094

PubMed Abstract | CrossRef Full Text | Google Scholar

24. Alvarez-Mon MA, Asunsolo Del Barco A, Lahera G, Quintero J, Ferre F, Pereira-Sanchez V, et al. Increasing interest of mass communication media and the general public in the distribution of tweets about mental disorders: observational study. J Med Internet Res. (2018) 20:e205. doi: 10.2196/jmir.9582

PubMed Abstract | CrossRef Full Text | Google Scholar

25. Alvarez-Mon MA, Donat-Vargas C, Llavero-Valero M, Gea A, Alvarez-Mon M, Martinez-Gonzalez MA, et al. Analysis of media outlets on women's health: thematic and quantitative analyses using Twitter. Front Public Health. (2021) 9:644284. doi: 10.3389/fpubh.2021.644284

PubMed Abstract | CrossRef Full Text | Google Scholar

26. Maiya AS. ktrain: a low-code library for augmented machine learning. arXiv. New York, NY. (2020). https://arxiv.org/abs/2004.10703arXiv:2004.10703.

Google Scholar

27. Conneau A, Khandelwal K, Goyal N, Chaudhary V, Wenzek G, Guzmán F, et al. Unsupervised cross-lingual representation learning at scale. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics (2020). p. 8440–51. doi: 10.18653/v1/2020.acl-main.747

CrossRef Full Text | Google Scholar

28. Ross Middleton KM, Patidar SM, Perri MG. The impact of extended care on the long-term maintenance of weight loss: a systematic review and meta-analysis. Obes Rev JAMA. (2012) 13:509–17. doi: 10.1111/j.1467-789X.2011.00972.x

PubMed Abstract | CrossRef Full Text | Google Scholar

29. Nanna MG, Navar AM, Zakroysky P, Xiang Q, Goldberg AC, Robinson J, et al. Association of patient perceptions of cardiovascular risk and beliefs on statin drugs with racial differences in statin use: insights from the patient and provider assessment of lipid management registry. JAMA Cardiol. (2018) 3:739–48. doi: 10.1001/jamacardio.2018.1511

PubMed Abstract | CrossRef Full Text | Google Scholar

30. Bradley CK, Wang TY Li S, Robinson JG, Roger VL, Goldberg AC, et al. Patient-reported reasons for declining or discontinuing statin therapy: insights from the PALM registry. J Am Heart Assoc. (2019) 8:e011765. doi: 10.1161/JAHA.118.011765

PubMed Abstract | CrossRef Full Text | Google Scholar

31. Alvarez-Mon MA, Llavero-Valero M, Sánchez-Bayona R, Pereira-Sanchez V, Vallejo-Valdivielso M, Monserrat J, et al. Areas of interest and stigmatic attitudes of the general public in five relevant medical conditions: thematic and quantitative analysis using Twitter. J Med Internet Res. (2019) 21:e14110. doi: 10.2196/14110

PubMed Abstract | CrossRef Full Text | Google Scholar

32. Sinnenberg L, Buttenheim AM, Padrez K, Mancheno C, Ungar L, Merchant RM. Twitter as a tool for health research: a systematic review. Am J Public Health. (2017) 107:e1–8. doi: 10.2105/AJPH.2016.303512

PubMed Abstract | CrossRef Full Text | Google Scholar

33. Athanasopoulou C, Sakellari E. “Schizophrenia” on Twitter: content analysis of Greek language tweets. Stud Health Technol Inform. (2016) 226:271–4.

PubMed Abstract | Google Scholar

34. Joseph AJ, Tandon N, Yang LH, Duckworth K, Torous J, Seidman LJ, et al. #Schizophrenia: use and misuse on Twitter. Schizophr Res. (2015) 165:111–5. doi: 10.1016/j.schres.2015.04.009

PubMed Abstract | CrossRef Full Text | Google Scholar

35. Reavley NJ, Pilkington PD. Use of Twitter to monitor attitudes toward depression and schizophrenia: an exploratory study. PeerJ. (2014) 2:e647. doi: 10.7717/peerj.647

PubMed Abstract | CrossRef Full Text | Google Scholar

36. Robinson P, Turk D, Jilka S, Cella M. Measuring attitudes towards mental health using social media: investigating stigma and trivialisation. Soc Psychiatry Psychiatr Epidemiol. (2019) 54:51–8. doi: 10.1007/s00127-018-1571-5

PubMed Abstract | CrossRef Full Text | Google Scholar

37. Pereira-Sanchez V, Alvarez-Mon MA, Asunsolo del Barco A, Alvarez-Mon M, Teo A. Exploring the extent of the Hikikomori phenomenon on twitter: mixed methods study of western language tweets. J Med Internet Res. (2019) 21:e14167. doi: 10.2196/14167

PubMed Abstract | CrossRef Full Text | Google Scholar

38. Hayhow B, Gaillard F, Velakoulis D, Walterfang M. Delayed diagnosis of multiple sclerosis in a patient with schizoaffective disorder: a case of “diagnostic overshadowing”. Aust N Z J Psychiatry. (2015) 49:186–7. doi: 10.1177/0004867414551067

PubMed Abstract | CrossRef Full Text | Google Scholar

39. Corrigan PW, Druss BG, Perlick DA. The impact of mental illness stigma on seeking and participating in mental health care. Psychol Sci Public Interest. (2014) 15:37–70. doi: 10.1177/1529100614531398

PubMed Abstract | CrossRef Full Text | Google Scholar

40. Webber M, Corker E, Hamilton S, Weeks C, Pinfold V, Rose D, et al. Discrimination against people with severe mental illness and their access to social capital: findings from the Viewpoint survey. Epidemiol Psychiatr Sci. (2014) 23:155–65. doi: 10.1017/S2045796013000243

PubMed Abstract | CrossRef Full Text | Google Scholar

41. Lasalvia A, Zoppei S, Van Bortel T, Bonetto C, Cristofalo D, Wahlbeck K, et al. Global pattern of experienced and anticipated discrimination reported by people with major depressive disorder: a cross-sectional survey. Lancet. (2013) 381:55–62. doi: 10.1016/S0140-6736(12)61379-8

PubMed Abstract | CrossRef Full Text | Google Scholar

42. Drapalski AL, Lucksted A, Perrin PB, Aakre JM, Brown CH, DeForge BR, et al. Model of internalized stigma and its effects on people with mental illness. Psychiatr Serv. (2013) 64:264–9. doi: 10.1176/appi.ps.001322012

PubMed Abstract | CrossRef Full Text | Google Scholar

43. Dickerson FB, Sommerville J, Origoni AE, Ringel NB, Parente F. Experiences of stigma among outpatients with schizophrenia. Schizophr Bull. (2002) 28:143–55. doi: 10.1093/oxfordjournals.schbul.a006917

PubMed Abstract | CrossRef Full Text | Google Scholar

44. Ahne A, Orchard F, Tannier X, Perchoux C, Balkau B, Pagoto S, et al. Insulin pricing and other major diabetes-related concerns in the USA: a study of 46 407 tweets between 2017 and 2019. BMJ Open Diabetes Res Care. (2020) 8:e001190. doi: 10.1136/bmjdrc-2020-001190

PubMed Abstract | CrossRef Full Text | Google Scholar

45. Alghamdi A, Abumelha K, Allarakia J, Al-Shehri A. Conversations and misconceptions about chemotherapy in arabic tweets: content analysis. J Med Internet Res. (2020) 22:e13979. doi: 10.2196/13979

PubMed Abstract | CrossRef Full Text | Google Scholar

46. Naslund JA, Grande SW, Aschbrenner KA, Elwyn G. Naturally occurring peer support through social media: the experiences of individuals with severe mental illness using YouTube. PLoS ONE. (2014) 9:e110171. doi: 10.1371/journal.pone.0110171

PubMed Abstract | CrossRef Full Text | Google Scholar

47. Birnbaum ML, Rizvi AF, Correll CU, Kane JM, Confino J. Role of social media and the Internet in pathways to care for adolescents and young adults with psychotic disorders and non-psychotic mood disorders. Early Interv Psychiatry. (2017) 11:290–5. doi: 10.1111/eip.12237

PubMed Abstract | CrossRef Full Text | Google Scholar

48. Serretti A, Chiesa A. A meta-analysis of sexual dysfunction in psychiatric patients taking antipsychotics. Int Clin Psychopharmacol. (2011) 26:130–40. doi: 10.1097/YIC.0b013e328341e434

PubMed Abstract | CrossRef Full Text | Google Scholar

49. de Boer MK, Castelein S, Wiersma D, Schoevers RA, Knegtering H. The facts about sexual (dys)function in schizophrenia: an overview of clinically relevant findings. Schizophr Bull. (2015) 41:674–86. doi: 10.1093/schbul/sbv001

PubMed Abstract | CrossRef Full Text | Google Scholar

50. La Torre A, Conca A, Duffy D, Giupponi G, Pompili M, Grözinger M. Sexual dysfunction related to psychotropic drugs: a critical review part II: antipsychotics. Pharmacopsychiatry. (2013) 46:201–8. doi: 10.1055/s-0033-1347177

PubMed Abstract | CrossRef Full Text | Google Scholar

51. Subotnik KL, Nuechterlein KH, Asarnow RF, Fogelson DL, Goldstein MJ, Talovic SA. Depressive symptoms in the early course of schizophrenia: relationship to familial psychiatric illness. Am J Psychiatry. (1997) 154:1551–6. doi: 10.1176/ajp.154.11.1551

PubMed Abstract | CrossRef Full Text | Google Scholar

52. Birnbaum ML, Ernala SK, Rizvi AF, Arenare ER, Van Meter A, De Choudhury M, et al. Detecting relapse in youth with psychotic disorders utilizing patient-generated and patient-contributed digital data from Facebook. NPJ Schizophr. (2019) 5:17. doi: 10.1038/s41537-019-0085-9

PubMed Abstract | CrossRef Full Text | Google Scholar

53. Rezaii N, Walker E, Wolff P. A machine learning approach to predicting psychosis using semantic density and latent content analysis. NPJ Schizophr. (2019) 5:9. doi: 10.1038/s41537-019-0077-9

PubMed Abstract | CrossRef Full Text | Google Scholar

54. Vancampfort D, Wampers M, Mitchell AJ, Correll CU, De Herdt A, Probst M, et al. meta-analysis of cardio-metabolic abnormalities in drug naïve, first-episode and multi-episode patients with schizophrenia versus general population controls. World Psychiatry. (2013) 12:240–50. doi: 10.1002/wps.20069

PubMed Abstract | CrossRef Full Text | Google Scholar

55. Correll CU, Solmi M, Veronese N, Bortolato B, Rosson S, Santonastaso P, et al. Prevalence, incidence and mortality from cardiovascular disease in patients with pooled and specific severe mental illness: a large-scale meta-analysis of 3,211,768 patients and 113,383,368 controls. World Psychiatry. (2017) 16:163–80. doi: 10.1002/wps.20420

PubMed Abstract | CrossRef Full Text | Google Scholar

56. Correll CU, Detraux J, De Lepeleire J, De Hert M. Effects of antipsychotics, antidepressants and mood stabilizers on risk for physical diseases in people with schizophrenia, depression and bipolar disorder. World Psychiatry. (2015) 14:119–36. doi: 10.1002/wps.20204

PubMed Abstract | CrossRef Full Text | Google Scholar

57. Foley DL, Mackinnon A, Morgan VA, Watts GF, McGrath JJ, Castle DJ, et al. Predictors of type 2 diabetes in a nationally representative sample of adults with psychosis. World Psychiatry. (2014) 13:176–83. doi: 10.1002/wps.20130

PubMed Abstract | CrossRef Full Text | Google Scholar

58. Maj M, Os J, De Hert M, Gaebel W, Galderisi S, Green MF, et al. The clinical characterization of the patient with primary psychosis aimed at personalization of management. World Psychiatry. (2021) 20:4–33. doi: 10.1002/wps.20809

PubMed Abstract | CrossRef Full Text | Google Scholar

59. Hennekens CH, Hennekens AR, Hollar D, Casey DE. Schizophrenia and increased risks of cardiovascular disease. Am Heart J. (2005) 150:1115–21. doi: 10.1016/j.ahj.2005.02.007

PubMed Abstract | CrossRef Full Text | Google Scholar

60. Schwenk ES, Jaremko KM, Park BH, Stiegler MA, Gamble JG, Chu LF, et al. I tweet, therefore I learn. Anesth Analg. (2019) 130:330–40. doi: 10.1213/ANE.0000000000004036

PubMed Abstract | CrossRef Full Text | Google Scholar

61. Naslund JA, Aschbrenner KA, McHugo GJ, Unützer J, Marsch LA, Bartels SJ. Exploring opportunities to support mental health care using social media: a survey of social media users with mental illness. Early Interv Psychiatry. (2019) 13:405–13. doi: 10.1111/eip.12496

PubMed Abstract | CrossRef Full Text | Google Scholar

62. Serôdio PM, McKee M, Stuckler D. Coca-Cola - a model of transparency in research partnerships? A network analysis of Coca-Cola's research funding (2008-2016). Public Health Nutr. (2018) 21:1594–607. doi: 10.1017/S136898001700307X

PubMed Abstract | CrossRef Full Text | Google Scholar

63. Bes-Rastrollo M, Schulze MB, Ruiz-Canela M, Martinez-Gonzalez MA. Financial conflicts of interest and reporting bias regarding the association between sugar-sweetened beverages and weight gain: a systematic review of systematic reviews. PLoS Med. (2013) 10:e1001578. doi: 10.1371/journal.pmed.1001578

PubMed Abstract | CrossRef Full Text | Google Scholar

64. Tapsell LC, Neale EP, Satija A, Hu FB. Foods, nutrients, and dietary patterns: interconnections and implications for dietary guidelines. Adv Nutr Adv Nutr. (2016) 7:445–54. doi: 10.3945/an.115.011718

PubMed Abstract | CrossRef Full Text | Google Scholar

65. Jacobs DR, Steffen LM. Nutrients, foods, and dietary patterns as exposures in research: a framework for food synergy. Am J Clin Nutr. (2003) 78:508S−13S. doi: 10.1093/ajcn/78.3.508S

PubMed Abstract | CrossRef Full Text | Google Scholar

66. Bragg MA, Pageot YK, Amico A, Miller AN, Gasbarre A, Rummo PE, et al. Fast food, beverage, and snack brands on social media in the United States: an examination of marketing techniques utilized in 2000 brand posts. Pediatr Obes. (2020) 15:e12606. doi: 10.1111/ijpo.12606

PubMed Abstract | CrossRef Full Text | Google Scholar

67. Smith A, Anderson M. Social Media Use in 2018. Washington, DC: Pew Research Center (2018).

68. Anderson M, Jiang J. Teens, Social Media & Technology. Washington, DC: Pew Research Center (2018).

Google Scholar

69. Ghaznavi J, Taylor LD. Bones, body parts, and sex appeal:an analysis of #thinspiration images on popular social media. Body Image. (2015) 14:54–61. doi: 10.1016/j.bodyim.2015.03.006

PubMed Abstract | CrossRef Full Text | Google Scholar

70. Tufts C, Polsky D, Volpp KG, Groeneveld PW, Ungar L, Merchant RM, et al. Characterizing tweet volume and content about common health conditions across pennsylvania: retrospective analysis. JMIR Public Heal Surveill. (2018) 4:e10834. doi: 10.2196/10834

PubMed Abstract | CrossRef Full Text | Google Scholar

71. Albalawi Y, Nikolov NS, Buckley J. Trustworthy health-related tweets on social media in saudi arabia: tweet metadata analysis. J Med Internet Res. (2019) 21:e14731. doi: 10.2196/14731

PubMed Abstract | CrossRef Full Text | Google Scholar