Linguistic Methodologies to Surveil the Leading Causes of Mortality: Scoping Review of Twitter for Public Health Data

Introduction

Background

Tens of millions of words, pictures, and videos are recorded on the internet via social media every day by people worldwide [1]. Some of the most popular social media platforms, such as Facebook, Instagram, Twitter, and Snapchat [2], enable social media users to interact in various innovative ways, such as sharing pictures, videos, and live streams (ie, real-time web-based video streams allowing participant interaction) [3]. Social media platforms such as Twitter have accumulated immense amounts of linguistic data from their users, with which researchers can identify patterns in individual qualities, characteristics, and social practices. Twitter is a popular web-based social networking and bulletin platform used by millions of people and organizations to post and discover information while interacting with messages known as tweets [4]. Twitter activity is initiated and extended by posting tweets, reposting important messages (ie, retweeting), and attracting the attention of other Twitter users (ie, followers) to an account [5]. Each day, approximately 500 million tweets are posted, conveying a diversity of topics and views [6] from >300 million active accounts worldwide [4,5]. This information, in turn, can help detect human behaviors and predict related medical conditions [1,7].

In the last decade, the internet and mobile technologies have become prevalent with increasing social media use by teenagers and adults each day [6]. In a recent study, almost 90% of 1060 adolescents aged between 13 and 17 years reported using social media, with >70% of them having a profile on multiple platforms [8]. Approximately two-thirds (65%) of adults report daily use of at least one social media platform [9]. Owing to their popularity, social media platforms represent an emerging innovative way to provide valuable information about the lives of individuals, including their health information [6]. According to Coppersmith et al [10], social media is a channel that showcases people’s language and behavior in an unbiased form, enabling researchers to diagnose certain conditions effectively.

Public health researchers use big data methodologies to mine Twitter for health-related data [4] at the individual and community levels as the platform offers unrestricted access to public tweets, unlike Facebook and Google Plus data, which are restricted and proprietary [11]. Researchers have analyzed Twitter data to gain an understanding of health-related conditions and used the data as an epidemiological surveillance source [12]. Unlike traditional surveys and tracking networks, which can be expensive and take extended periods to deliver salient information, Twitter can be less expensive for assessing health concerns and provide a faster response to and an in-depth understanding of public health threats [12,13].

Objectives

Twitter users provide a foundation for new information by openly sharing user-generated, real-time information that is not available to their health care clinicians [12], for example, the unapproved use of medications [14]. Twitter may be a reliable and useful social media platform for public health researchers to remotely monitor vexing public health issues associated with risk behaviors and disease epidemics [15]. It may also be useful in providing potentially valuable insights for researchers who can harness Twitter’s internet-based features to provide health communication and promotion campaigns [16]. In recent years, there have been several review articles published on the benefits of using Twitter as a data source [17,18]; little work has focused on the novel applications of statistical language analyses used to study human health and behavior or on a comprehensive review on public health surveillance of leading causes of mortality related to emerging disease epidemics, chronic diseases, and risk behaviors. Therefore, the primary focus of this scoping review was to comprehensively review the literature that leverages Twitter as a source in public health research for analyzing users’ language to conduct public health surveillance of high-mortality diseases and study physical and mental health conditions. Thus, this scoping review will (1) outline the inclusion and exclusion criteria for identifying and selecting the literature that will be discussed; (2) provide a review of the literature on the 2 themes that emerged supporting Twitter being used as a public health data source; and (3) conclude with an overview of the efficacy of Twitter data and discuss the limitations, implications, and needs regarding future research in this area.

Methods

Overview

A scoping review methodology was used to map relevant research that has leveraged Twitter as a data source in public health research to analyze, determine, and monitor a variety of public health concerns. Using the definition presented by Arksey and O’Malley [19], a scoping review aims to “rapidly” map fundamental conceptual aspects and foundations in the relevant area and identify the primary sources, evidence, and research gaps in a comprehensive format. As Twitter has become widely used as a data source, we decided to focus on comprehensively identifying the relevant research area and summarizing and reporting research findings as per the first and third goals of a scoping review by Arksey and O’Malley [19].

Search Strategy

A literature search strategy following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extended guidelines for scoping reviews was used to search specific keywords (Textbox 1) on Twitter and public health on 5 databases: Web of Science, PubMed, CINAHL, PsycINFO, and Google Scholar. In our research, we selected commonly used search terms based on a thorough review of the literature and related studies in the field. These terms were then used as the basis for our search without expanding them to minimize the risk of bias and ensure the accuracy of the results. By carefully selecting and focusing on specific terms, we were able to limit the volume of articles retrieved, effectively narrow down the search results, and ensure that the articles obtained were relevant and of high quality. Furthermore, to ensure the comprehensive nature of our review, we thoroughly examined multiple reference lists [17,18,20] to identify any relevant articles that may have been missed in our initial search strategy. This process of checking reference lists helped ensure that all relevant studies were included in our review. In addition, we evaluated and determined that any articles not identified by our initial search strategy would not considerably affect the conclusions drawn from the review based on identifying a substantial amount (N=4986) of relevant studies.

Study Selection

The search strategy was performed to identify relevant articles that met the following criteria: (1) peer-reviewed empirical articles representing original research from 2008 to 2021, (2) provision of research methodology and findings, (3) use of Twitter to answer research questions, and (4) being available in English (Textbox 2). We defined public health research as research that contributes to the definition of public health used by Winslow [21]: “the science and the art of preventing disease, prolonging life, and promoting health through the organized efforts and informed choices of society, organizations, public and private communities, and individuals.” The research articles were evaluated based on 4 quality criteria: the presence of a precisely defined purpose, description of techniques for mining Twitter data, discussion of methodology, and discussion of limitations [4]. After a database search leading to the identification of 4986 articles, all articles were screened by the first author, which resulted in the exclusion of 3410 (68.39%) articles screened by keywords, title and abstract, and inclusion criteria. The remaining 1576 articles were screened at the full text level to determine inclusion, resulting in the exclusion of 1295 (82.17%) articles. Of the remaining 281 articles, 38 (13.5%) were reviewed by the first 2 authors, followed by an agreed consensus on the quality inclusion criteria, and were subsequently included in the scoping review.

Data Extraction

For each article, data were extracted related to (1) author, year, and location of study; (2) study objective; (3) study design; (4) study methods; (5) sample size; and (6) a succinct observation of the limitations and implications.

Results

Overview

The search strategy successfully identified >1576 peer-reviewed articles. Of these, 38 (Table 1) met the eligibility criteria (see Figure 1 for the PRISMA flowchart). This section presents the literature that used Twitter as a data source for (1) language analysis to identify health threats and physical and mental health understandings about people and societies and (2) public health surveillance related to the leading causes of mortality representing 3 primary categories (respiratory infections: 10/38, 26%; cardiovascular diseases: 3/38, 8%; and COVID-19: 4/38, 11%). Articles included results on analyzing the contents of tweets to study influenza rates; heart disease; cardiac arrest; COVID-19; well-being; substance use; and mental health conditions such as attention-deficit/hyperactivity disorder (ADHD), posttraumatic stress disorder (PTSD), and depression.

‎

Twitter for Public Health Surveillance: Leading Causes of Mortality

Overview

Twitter analysis has the potential to be precise, cost-effective, and practical for keeping track of people’s behaviors, health outcomes, emergencies, and disease outbreaks. The process of public health surveillance systematically collects and analyzes health-related data for disseminating information about public health practices and evaluating implemented methods [56]. Surveillance offers early warnings for approaching calamity, records for monitoring and evaluation, and evidence of disease transmission [56]. Although conventional clinical databases were previously the main available surveillance frameworks, digital technology innovations have changed the conduct of public health surveillance [57]. Researchers have found that digital surveillance methods provide advantages (eg, timeliness and cost) over the more conventional public health surveillance systems [58]. They supplanted passive surveillance methods (eg, physician case reports to local health departments) and active methods (eg, local health department activity searches for other cases) [58]. In this section, we discuss the literature that demonstrates how Twitter data were used for real-time surveilling of chronic illnesses, disease epidemics, and problematic behaviors to support and improve public health.

Influenza Surveillance

Influenza was the first and most common disease to be examined using Twitter. Several exploratory studies (9/38, 24%) on influenza targeting Twitter users from the United States and the United Kingdom between 2009 and 2012 used primary keyword searches from “influenza” or “flu” [27-29,36] to more definite influenza subtypes such as H1N125 [24,25,29] and symptomatic influenza groups [26,32,35]. A study by Signorini et al [32] searched important terms such as “influenza,” “infection,” and “hospital” to track Twitter users’ concerns during the 2009 H1N1 pandemic. A similar study by Sadilek et al [35] identified the health of Twitter users through geotagging (ie, geographical identification) to potentially predict influenza transmission. Twitter was shown to predict influenza epidemics 2 weeks before traditional surveillance methods used by the Centers for Disease Control and Prevention [32].

Escherichia Coli and Dengue Fever Surveillance

Recent studies used Twitter data to conclude whether a rapid opportunity for the detection of further infectious disease occurrences, such as E coli, cholera, and dengue fever, can be improved. A study by Diaz-Aviles and Stewart [30] analyzed >6 million Twitter posts on health conditions during the 2011 E coli outbreak in Germany. The authors identified almost 500,000 tweets related to E coli and determined that users’ tweets would have detected the epidemic 24 hours earlier than traditional warning surveillance methods. In a study by Gomide et al [31], the authors proposed a dengue fever surveillance system that demonstrated a correlation between the number of dengue cases reported by Brazil’s official statistics and tweets based on the volume, location, time, and public perception of users. In a similar study, Chunara et al [34] found that cholera-related tweets could anticipate earlier estimations of news trends and reported cases 2 weeks earlier than Haiti health authorities reported during the first 3 months of the 2010 Haitian cholera outbreak. Surveillance of disease outbreaks using Twitter data as a tracking system appeared to have timeliness advantages in numerous infectious disease epidemic settings [20].

COVID-19 Surveillance

Twitter activity increased substantially during the COVID-19 pandemic. Recent surveillance studies worldwide used these data to assess the unprecedented impact caused by the disease. The studies examined the rapid dissemination of COVID-19 information and misinformation and news-sharing behaviors [37-39] and identified high Twitter volume as a potential predictor for confirmed cases [38,40]. Specifically, increased volumes of COVID-19–related tweets were observed in large European cities with more concentrated populations [39] following the World Health Organization announcement releasing the name of the nascent disease. In the United States, high volumes of COVID-19–related tweets were analyzed to monitor the trends of information spread and were linked to longitudinal trends in local infection rates [40]. For example, highly concentrated cities such as New York City contained areas such as Manhattan with very high volumes of tweets linked to extremely high infection rates in the early stages of the pandemic [40]. Similarly, but from a global perspective, Singh et al [38] found that confirmed COVID-19 cases were highly associated with the location of COVID-19–related tweets. Their results indicated that users sharing COVID-19–related tweets were mainly from China, followed by other epicenter countries such as the United States, India, Iran, and Italy. The studies also shed light on news-sharing behaviors and the fast dissemination of news information and misinformation via Twitter. South Korean users tweeted or retweeted medical-related news information more often and quickly than nonmedical information, and users who included the word “Coronavirus” were more likely to engage with each other more frequently [37]. COVID-19–related information was widely tweeted, but research also found that users shared a substantial amount of misinformation that might have driven confusion and mistrust in governmental initiatives to combat the pandemic. Although a high volume of incredible information was shared, Singh et al [38] reported that credible information was lateral to incredible information. Overall, the results from these studies suggested that COVID-19–related tweets may help understand the impact of COVID-19 on news-sharing behaviors, public responses, and predictions of potential infectious outbreaks and reproduction when other reliable surveillance measurements are not readily available.

Cardiovascular Disease Surveillance

Of the 38 studies, 3 (8%) exploratory studies targeting cardiovascular disease focused on analyzing Twitter data to determine mortality rates and the public’s understanding and communication about cardiovascular disease. As stated previously, Twitter linguistics were used to predict heart disease mortality rates based on community-level language patterns reflecting negative emotions that emerged as risk factors and positive emotions that emerged as protective factors [13]. Twitter posts that contained text related to heart disease included information about risk factors, management, and awareness that could enhance cardiovascular disease surveillance and market to precise users in need of medical assistance [4,23]. Sinnenberg et al [4] examined 2500 heart-related tweets and topics, successfully tracked health status in real time, and characterized the public understanding and communication about heart disease. In summary, the studies showed that Twitter data could be a valuable tool for surveilling mortality rates and monitoring real-time changes in discussions about heart disease. In addition to using Twitter data as a surveillance tool for cardiovascular disease, many scholars found effective ways to monitor substance use behaviors.

Substance Use Surveillance

Twitter has been a promising platform for surveying substance use such as alcohol, tobacco, cannabis, and other drugs as Twitter users often publicly share and display their substance use activities via substance use–related content. The studies used Twitter to examine emerging trends in alcohol [44,52,55] and cannabis use [47,48,51,54], track alcohol [46,49,52] and opioid abuse [53] using geocoded data, study the effect of proalcohol and procannabis content on use among young adults [50], and identify the beliefs and behaviors of youth alcohol and tobacco use patterns [45]. Traditional surveillance methods could not provide real-time, population-based surveillance for substance use, but mining tweets provided real-time surveillance that served as a population-based approach to investigating substance use trends and trajectories [49].

Summary

In this section, studies leveraging Twitter data centered on understanding how diseases are dispersed through a population and predicting emerging public health trends by tracking infection rates, health status (eg, cardiovascular disease), and consumption (eg, substance use) in real time. The studies looked beyond a better insight into disease spread and health trends. However, they also gained deep insights into monitoring the public health state of disease occurrence, symptoms, and outbreaks while characterizing the public understanding and communication about a disease. In summary, Twitter data can benefit public health actions in disease and behavioral surveillance, showing positive results at the individual and community levels. In the following section, we will review the literature that used language analyses and approaches to process and analyze Twitter data to identify and understand physical and mental health conditions as mental health problems can affect physical health conditions and are commonly associated with the risk of chronic diseases and substance use.

Language Analysis to Study Physical and Mental Health

Before exploring research on the relationship between physical health, mental health, and Twitter language, it is vital to understand how the linguistic analysis of large data sets is carried out. A variety of language analyses were used descriptively to identify and monitor emerging health threats (eg, influenza outbreaks) [26]. Moreover, these analyses provided health and psychological insights into people and societies to make predictions about a variety of health-related outcomes [1]. To conduct these analyses, researchers used an automatic computer process to extract the relative frequencies of single words, phrases (2 or 3 consecutive words), and topics across millions of users’ tweets [6,13]. Of the 38 articles included in this review, 8 (21%) analyzed tweets’ language about a specific health topic to characterize public discourse on Twitter. Within this group, 1 subcategory focused on sentiment analysis—the process of deriving opinions, feelings, and subjectivity in texts [59], such as whether users’ tweets convey positive, negative, or neutral sentiments [60].

The tremendous amount of personal information shared on Twitter can be examined to provide physical health data [15]. For example, Bosley et al [23] and Eichstaedt et al [13] used Twitter to gain insights into cardiovascular disease based on users’ language. Both characterized >112,000 tweets about cardiovascular issues and resuscitation by identifying keywords such as “cardiac arrest,” “heart disease,” and “cardiopulmonary resuscitation.” The tweets were characterized by content, dissemination, and temporal trends (days and times with a high volume of tweets on a topic). By examining 50,000 tweets, Eichstaedt et al [13] used a cross-sectional regression model based on Twitter language to predict mortality rates from atherosclerotic heart disease. Bosley et al [23] identified 62,163 tweets using 7 search keywords in a 38-day time frame. They found that 25% of users’ tweets contained resuscitation and cardiac arrest information. Twitter users shared linguistic information related to cardiac arrest and resuscitation, including risk factors, symptoms, training, education, screenings, and other information. The findings of these 2 studies indicated that Twitter language data can be mined to detect death rates, identify heart-related content, and better understand how heart health–related information is shared and discussed.

The language analysis of tweets may also be beneficial for understanding other health-related conditions. Influenza was the most frequently analyzed disease using Twitter data [12] by evaluating and characterizing the language used. Twitter users frequently freely tweeted private content; tweets similar to “sick with the flu” and “I got the flu” are typical posts [22]. Although realizing that 1 user has the seasonal influenza virus might not be particularly intriguing, millions of such tweets can be revealing, for instance, to follow the influenza rate in the United States [26]. For example, Culotta [26] mined Twitter discussions as a way to validate Twitter’s role as a method of warning practitioners and health officials about influenza epidemics in the United States. More recent influenza epidemics such as H1N1 were successfully analyzed to evaluate the general public’s awareness of public health advice [25]. In 3% (1/38) of the studies, researchers analyzed 2 million tweets containing the key terms “swine flu” and “H1N1” using Infovigil, an “infoveillance” system, and found elevated levels of “precise knowledge among the public” [25]. The study recommended that Twitter mining may be an innovative method for health officials to “measure public awareness of their campaigns and respond to shifting concerns in real time” [25]. Another influenza study by Ritterman et al [24] stated that analysis of tweets about influenza-related keywords could enhance prediction model precision by giving advance notices of severe occurrences, for example, the H1N1 epidemic.

Analyzing Twitter messages can also have a more substantial impact on public health informatics than monitoring influenza rates [22]. A recent study by Guntuku et al [6] focused on adult behaviors with self-reported diagnoses of ADHD as described by users in their Twitter posts. The authors computationally analyzed 1.3 million tweets from nearly 1400 users to gain insights into the type of discussions and information shared and “how their language is correlated with users’ characteristics, personality, and temporal orientation” [6]. By analyzing users’ tweets, they found that Twitter users with ADHD were more active on the platform, less friendly, and posted more profanity. According to the researchers, the posting behaviors of Twitter users with self-proclaimed ADHD were, in fact, consistent with the symptoms of the disorder.

In addition to providing data for physical health conditions, researchers analyzed Twitter language to predict and determine emotional well-being [24,43] as well as various mental illnesses [10,41]. Psychological conditions, particularly depression [42] and PTSD [10], constitute a growing problem that affects millions of Americans. Unlike other health research using social media that relies on exact words or phrases related to a disease or health concern, research on mental illness has leveraged variations in language to suggest behavior changes. For example, a change in word use or frequency of posts signaled changes in a person’s mental health [10]. Similar to how other researchers extracted tweets to study influenza rates and predict heart disease, researchers used the same automatic computer process to extract the relative frequencies of words and phrases related to mental illness. However, another innovative process called crowdsourcing collected data by soliciting contributions from a large group of people, known as crowd workers, to examine mental health disorders on Twitter [41].

Twitter language is a potential source for measuring and predicting mental health conditions in people who use the platform. de Choudhury et al [41] identified users with major depression using crowdsourcing data on the presence of depression. They found that users with depression exhibited lower social media activity, more negative emotions, higher self-attentional focus, and more expressions related to religion. Similarly, Coppersmith et al [10] demonstrated that some Twitter users with PTSD were not difficult to identify when mining for Twitter posts that communicated definitive diagnoses rather than relying on conventional PTSD measurements. Coppersmith et al [10] used positive and negative PTSD language sentiments to train classifiers, which are language models that analyze probabilities that the same underlying process generates words or a thread of characters as the training data. The words were then divided into positive and negative classes. Using this technique, Coppersmith et al [10] captured a fundamental difference between Twitter users with and without PTSD from their Twitter language. The results revealed that users with PTSD posted substantially more Twitter messages using more negative emotions, anger, anxiety, and death-related words. On the basis of these 2 studies, analyses of users’ language on Twitter were deemed useful for detecting and identifying mental health conditions in individuals [41].

All the studies mentioned in this section analyzed the content of tweets and constructed various statistical language models (eg, PTSD classifiers). Statistical language models are algorithms for learning and capturing the critical statistical characteristics of the distribution of sequences of words in a language. These models use predictors representing expressions of positive and negative sentiment in Twitter language variables [61]. Many studies (18/38, 47%) also constructed predictive models using data mining and probability [62]. In addition, in recent years, deep learning and natural language processing have been widely used and are popular techniques embedded in many analytical models using Twitter data to study public health research. Leveraging these techniques and strategies, researchers used Twitter language for many functional tasks, ranging from providing insights into individual characteristics and personalities to continually monitoring positive and negative sentiments and predicting various health-related outcomes. In summary, the analysis of language on Twitter was practically unlimited. Individuals left digital footprints everywhere, expressing their emotions, behaviors, identities, personalities, and experiences [13].

Limitations of the Existing Studies

A few limitations should be noted about the existing studies. A limitation of many Twitter-based studies was that Twitter users are only a subset of the general population. The Pew Internet and American Life Project [63] found that 31% of Twitter data tend to come from users aged between 18 and 29 years, in contrast to 19% of Twitter users being aged between 30 and 49 years, 9% of users being aged between 50 and 64 years, and <5% of users being aged >65 years. Therefore, there is a requirement for higher granularity in Twitter-based studies, especially for studies that leverage Twitter as a surveillance tool. Twitter analysis as a public health instrument raises ethical concerns as well. Regarding users’ privacy, some apprehensions include the use of the general population’s information without verbal or written consent as well as individuals lacking informed expectations regarding how their data are being used as part of public health research [64]. Twitter has the potential to help facilitate information sharing but might also be problematic in disseminating false information during standard and seasonal health events and crises. The literature also provides limited insights into how to decrease the greater possibilities of inaccurate information being mistakenly or deliberately shared with individuals searching for health information on the internet, creating accountability and authenticity challenges [65].

Discussion

Principal Findings

This scoping review provided a comprehensive synthesis of current research, leveraging Twitter data to analyze users’ generated language to explore and understand physical and mental health conditions and remotely monitor the leading causes of mortality related to emerging disease epidemics and risk behaviors. From the 38 reviewed empirical articles that were published within the past 13 years using Twitter as a data source in public health, two themes were identified: (1) analyzing Twitter language to determine physical and mental health outcomes and (2) using Twitter for public health detection and surveillance. On the basis of the research reviewed in this study, we discovered that Twitter holds several distinct types of valuable evidence for the field of public health in several different conditions. Analyzing language on Twitter can serve as an innovative method to examine health-related conditions such as chronic illnesses, especially for mental illnesses such as depression [41] and PTSD [10]. We identified studies using Twitter data to analyze mental illness, with findings suggesting that people are more open to addressing their mental health issues on social media [41]. Finally, we observed trends in Twitter-based studies proving that analyzing millions of tweets can offer a beneficial understanding of population health, disease tracking, and surveillance and produce rich data that connect with public health metrics and knowledge [22].

We found that several published studies used a variety of sophisticated language analysis techniques and strategies to mine user-generated tweets to help detect a wide range of conditions such as mental illnesses (depression and PTSD) and ADHD symptoms, monitor emerging disease outbreaks, and provide a deeper insights into people’s personalities and behavioral characteristics at a large scale. Furthermore, similar to previous review studies [17,18,20], our findings suggest that, by users leaving digital footprints, studies could analyze Twitter language in a real-time and less time-consuming manner while also opening an unobtrusive window into understanding complex health and behavior conditions. In addition, our findings demonstrated that most surveillance studies (17/38, 45%) primarily focused on infectious diseases, partly because of their global importance. Notably, the studies focused excessively on the observation and surveillance of influenza rates. However, we identified studies with a scope to explore the utility of Twitter data to survey other infectious diseases such as COVID-19 and noninfectious diseases such as cardiovascular disease and substance abuse. Despite our efforts to report on these studies, these areas remain understudied, yet these diseases often represent an enormous health burden among general populations worldwide. Overall, our findings suggest that using language analytical techniques and surveillance via Twitter could have a long-lasting impact on the domain of mental and physical illnesses and health-related and disease surveillance, conveying the larger media lens to the public health field [20]. This effect was imperative in public health crises such as Haiti’s cholera epidemic, where Twitter was used as a data tool for immediately evolving circumstances. As such, we are confident in the thoroughness and accuracy of our findings.

Finally, we think that the critical implications of our findings include emphasizing the importance of using Twitter as a supplement to more conventional public health surveillance infrastructures. Twitter can potentially fortify researchers’ ability to collect data in a timely way [65,66] and improve the early identification of potential health threats [67]. In addition, as Twitter expands rapidly worldwide, results from Twitter-based studies can inform health policies related to social media, health communication, and prevention. More strategies regarding physical and mental health promotion that help people connect with reliable information and resources are necessary, and health promotion messages should be reframed to help users feel more comfortable accessing or reaching out to services related to mental health. Finally, public health agencies must continue to use Twitter and other social media platforms (eg, Facebook) as a new form of communication targeting audiences and strengthening health and prevention messages from more traditional media outlets (eg, television, radio, and print) [68-72]. Twitter can concurrently empower the fast and constant capture of the public’s perceptions, notions, and knowledge about health-related topics. It is an inexpensive development for communicating health campaigns and messages. Moreover, Twitter exhibits a considerable potential to modify messages and connect with people in real-time discussions about health and prevention.

Strengths and Limitations

There are several strengths and limitations to this scoping review. First, compared with most review articles examining the benefits of using Twitter as a public health data source, this scoping review explicitly elaborates on the statistical language analyses and techniques implemented and used to study human health and behavior. Another strength is that this review provides a narrower objective than previous review papers by focusing on Twitter’s ability to surveil leading causes of mortality related to emerging disease epidemics, chronic diseases, and risk behaviors, representing 3 categories (respiratory infections, cardiovascular disease, and COVID-19). Finally, we incorporated studies that used Twitter to understand mental health conditions as mental health problems can increase the risk of chronic diseases and risky behaviors. The primary limitation of this scoping review is the restrictions in the search methodology. For example, excluding non–peer-reviewed literature, non–English-language articles, and works in progress and the focus on broader search strings used to identify literature may have prevented relevant studies from being discovered. Despite this limitation, this is common in scoping review studies as they are deliberate to broadly map a topic of interest while successfully achieving a balance between breadth and depth in a prompt time frame [73].

Gaps and Future Research

Despite the existing literature on the potential application of Twitter in various public health settings, a limited amount of public health research has examined the relationships between users’ generated images and videos, specifically food-related and health-related behaviors. Most of the presented literature exclusively focused on using text to analyze and communicate. Visual content (ie, images and videos) has increasingly become the most shared content on social media platforms and may provide new avenues to study social and health-related behaviors. Food is one of the most common visual pieces of content shared on all social media platforms. There has been a strong association between the food people eat and how it affects people’s moods and feelings in many ways [74]. Previous research has found that people eat to relieve stress, depression, anxiety, and other harmful emotional strains [75,76]. To explore this gap, creating a computer-based process that simultaneously analyzes food-related images and text is needed. New techniques may provide insights into how food-related images can determine a positive or negative emotional state.

Conclusions

As an emerging field, Twitter analysis has a promising future in public health communication and surveillance based on the 38 research articles included in this scoping review. Our findings shed light on the benefits of mining Twitter, provide evidence of language analysis applications that have been used to study tweets that can help identify subtle signals in language for understanding physical and mental health conditions and monitor emerging disease outbreaks, and could enable public health researchers interested in big data methodologies related to health outcomes to identify relevant Twitter-based studies. Most notably, our work extends the literature to include focusing on novel applications of language analyses and comprehensively reviewing public health surveillance of leading causes of mortality. In addition, beyond understanding and monitoring human health and behavior, our review shows that the Twitter platform can support public health policy and communication, thus being used to convey a broad scope of time-sensitive health and mental health promotion and prevention information. This innovation will conceivably expand the reach to hard-to-reach populations such as adolescents, who are more inclined to use and interact with Twitter than with more traditional public health media channels. Public health agencies nationwide are critical actors in this domain. It is enlightening to analyze how these agencies use relevant Twitter applications, which probably improve intelligence in communicating with their followers.