Reports and analyses highlight that approximately 60% of health-related tweets contain links to health-related websites [1]. Twitter users frequently share health information related to chronic diseases, mental health, and general wellness topics [2]. Twitter is a popular platform for health-related conversations because it allows users to share their thoughts, experiences, and information in real time [3]. This platform can be particularly useful for sharing information about health events, such as disease outbreaks or public health campaigns. In addition, Twitter can be used to connect with others with similar health concerns or interests and to access information from health care professionals and organizations [4]. Twitter is a social media platform with a character limit, which makes sharing detailed information about health issues difficult. Moreover, the information shared on Twitter may not always be accurate or reliable, as it is not always fact-checked or verified. However, Twitter is a common platform for people to share opinions, discuss health-related topics, and converse with a wide audience. According to a survey conducted by the Pew Research Center, 21% of Twitter users have used the platform to share information about a health condition [5]. The survey also found that 20% of Twitter users have followed a health organization or medical professional on the platform, and 15% have searched for information about a health condition on Twitter. Users often express their views about health policies, medical breakthroughs, health care services, and public health issues [6]. Health professionals, researchers, advocacy groups, and patients actively participate in these discussions, contributing diverse perspectives and sharing evidence-based information. This open and rapid exchange of ideas allows health information dissemination and facilitates conversations that can influence public opinion and policy decisions [7].

Sharing health information on Twitter raises privacy concerns as it involves sharing personal and sensitive data on a public platform [8]. Information privacy refers to individuals’ control over collecting, using, and disclosing their personal information. Regarding sharing personal health information, web-based information privacy refers to protecting sensitive health data from unauthorized access, secondary use, or disclosure [9]. It involves ensuring that individuals can make informed decisions about how their health information is shared and used and that appropriate safeguards are in place to protect the confidentiality and security of this information. The potential privacy risks associated with sharing health information on Twitter can be grouped into 3 reasons. First, potential identification and disclosure of personal information—sharing health information on Twitter can inadvertently lead to disclosing personally identifiable information. A study found that anonymized data from health-related tweets could be reidentified to reveal the identity of users [10]. Researchers were able to reconstruct personal health stories and connect them to specific individuals, highlighting the potential privacy risks involved. Second, data mining and analytics—third parties can analyze and use health-related tweets for various purposes, including targeted advertising or creating consumer profiles. Researchers analyzed tweets related to mental health and found that the content could be used to predict users’ self-reported diagnoses, medication use, and other personal information [11]. This demonstrates the potential for extracting sensitive health-related data from Twitter. In addition, a study used deep neural networks to identify personal health experience tweets, highlighting the potential for using Twitter as a data source for health surveillance studies [12]. Third, public disclosure of sensitive health information—sharing health information on Twitter might inadvertently expose individuals to public scrutiny and judgment. A study examined tweets related to mental health and found that users often disclosed personal experiences, symptoms, and treatments [13]. Although this sharing can provide support, it can also expose individuals to potential stigma, discrimination, or unwanted attention.

Previous studies suggest that individuals may be comfortable with sharing general information that is not sensitive on social media [14]. However, people may not be likely to share personal information, especially health-related data, owing to privacy concerns [15]. According to previous studies, privacy concerns can arise from companies’ information collection and use policies in the age of medical big data [16] and web-based social interactions that may threaten information privacy [17]. Twitter is reported as an important data set for vendors, researchers, and medical companies to collect health-related information [18]. Many medical companies collect health information and patient experiences from Twitter for big data analysis to find patterns for public health management [19]. Although big data collection and data mining techniques could help generate intelligence for monitoring public health issues, they can cause privacy concerns. Reports highlight that many Twitter users have experienced invasion of privacy owing to companies’ collection, sharing, and analytics practices that use information from their tweets, including private health information [20].

Although there are various studies of vendor-related privacy concerns [21] and peer-related privacy concerns [22], little is known about whether these 2 aspects of privacy concerns may collectively influence information-sharing behaviors. As privacy violations can be related to peers (such as inappropriate comments and unauthorized retweeting) and companies (sharing personal information with third parties), more studies are required to examine whether information-sharing disclosure can be affected equally by vendor-related and peer-related privacy concerns. In this study, we aimed to determine whether both aspects of privacy concerns (ie, concern for information privacy [CFIP] and peer privacy concern [PrPC]) can mutually change health-related information-sharing decisions or whether one’s effects can dominate or overshadow the impact of the other. For instance, whether the nature of the relationships and contexts with 2 different information trustees (ie, vendors and peers) can influence information dissemination behavior. Thus, we argue that both aspects of privacy concerns should be considered in a model to better characterize information privacy on social media. Investigating the importance of privacy concerns related to companies and vendors (such as Twitter analytics) and web-based peers (such as retweeting a focal user’s health information without permission) in case of disclosing public and private health information on Twitter would be the main contribution that makes this study different from others.

This argument is built based on 4 reasons. First, although the main interactions on social media are mainly peer oriented, vendors can still collect a lot of personal data (such as health information) without authorization and use it for unconsented purposes [23]. There have been instances of social media platforms, including Twitter, being used by organizations for health-related data mining and analysis [24]. Twitter data can provide valuable insights into health-related trends for health care organizations through analytics [25]. Researchers and companies (such as pharmaceutical manufacturers) have used Twitter data to track and analyze health-related trends, including disease outbreaks, medication use, and public health concerns [26]. For example, a study found that Twitter data could be used to track the spread of influenza and predict outbreaks [27]. Another study uncovered that Twitter data could be used to monitor adverse drug reactions and identify potential safety concerns [28]. Pharmaceutical companies have also used Twitter data to monitor medication use and patient experiences. For example, a study reported that Twitter data could be used to monitor patient experiences with antidepressant medications [29]. Moreover, it is common for organizations and vendors, including those in the health care industry, to monitor social media platforms to gather insights about consumer opinions, preferences, and trends [30]. Twitter, as a popular social media platform, has been used for these purposes [31]. Health care organizations and vendors may collect health-related information, such as discussions about medical conditions, treatment experiences, and patient preferences, from public Twitter profiles [19]. These insights can be valuable for marketing and market research purposes.

Second, although the primary purpose of peer-to-peer (P2P) interactions on social media is to maintain social connections with peers and there are no explicit business-to-consumer interactions, companies can still use social media analytics to investigate published health information. Companies can leverage various analytics tools to gather and find meaning and patterns in data collected from social channels to support business decisions (such as predicting the risk factors to manage public health) [32]. Third, individuals share a variety of information (such as about lifestyle, health status, chronic issues, and medication) on social media, which can be more sensitive than conventional e-commerce information (such as transaction records). Disclosing a wide range of information across social media platforms could raise concerns about whether companies and peers misuse the shared data (eg, health information). Fourth, Web 2.0, a fundamental technology supporting social media, mainly focuses on bilateral relationships between peers. However, it does not remove traffic between companies and social media users. Thus, peers can comment on conversations about a user’s health condition and share others’ personal health information on their own channels. In contrast, companies can use big data analytical tools to collect, use, or share users’ personal health data with third parties.

The main objective of this study was to investigate the concept of information privacy concerns in the context of social media based on both vendor-related and peer-related aspects. To do so, we used the survey research methodology and Twitter as the empirical context. We also relied on theories discussing 2 dimensions of information privacy (ie, CFIP and PrPC) as the theoretical foundation of our proposed model. In this study, CFIP, emphasizing both consumer perspectives and company responsibilities, represented privacy concerns related to web-based practices of companies and vendors (such as collection and sharing of self-shared information), and PrPC referred to privacy concerns about losing control over digital communications and web-based interactions with peers. Thus, we suggest that information privacy on social media can be multidimensional, focusing on privacy violations associated with companies’ (vendors’) practices and sharing behaviors of peers. The integration of CFIP and PrPC can comprehensively present the entirety of privacy concerns about web-based health information. This study contributes to both theory and practice. We shed more light on information privacy conceptualization in the context of social media. This study also provides an interactive outlook and practical recommendations for handling privacy issues by explaining how web-based vendors and peers may cause privacy violations when dealing with health information (general and specific) shared over the web.

Although tweets are publicly accessible by default, users likely expect some degree of privacy and control over their personal health information shared on the platform. Previous literature has found that even when posting content publicly on social media, individuals still have privacy interests and concerns about how their data might be used or accessed [61]. General health information shared publicly on Twitter, such as mentions of hospitals, physicians, and common diseases, is not considered protected or private. However, more specific personal health details, such as past medical history, allergies, medications, and current symptoms, could reveal private information about an individual’s health status. Although these details may be shared publicly by default on Twitter, users likely still have privacy concerns about this content being widely disseminated or used without their consent.

The concepts of CFIP and PrPC capture these types of privacy concerns. Although users are voluntarily sharing health information publicly on Twitter, they may still desire control over how these data are accessed and used. CFIP reflects concerns about using or sharing personal health data by third parties such as researchers or companies without the user’s knowledge or permission. Even if users willingly post health information publicly, they may still desire control over how that data are collected, analyzed, or shared by entities such as researchers, pharmacies, insurance companies, and so on. PrPC represents concerns about controlling boundaries around health disclosures and limiting exposure to certain audiences, such as employers or insurers, who could misuse the information. Users must balance sharing personal details with managing social risks if the information reaches unintended viewers such as employers, family members, or friends. Thus, although Twitter data are technically public, users are likely to have nuanced privacy interests surrounding their health disclosures. Therefore, concepts such as CFIP and PrPC are useful for quantifying expectations regarding control, anonymity, and audience boundaries that persist even when posting health care–related content openly over the web.

The model focuses on health information and Twitter (as the research context). There are a few critical differences in the privacy concerns around health information compared with other types of information. First, health information is considered to be very sensitive and private. It can reveal details about medical conditions, treatments, prescriptions, family history, and so on. Other types of information, such as social media posts or shopping habits, are generally not as sensitive. Second, health information has strict legal protections such as Health Insurance Portability and Accountability Act in the United States and General Data Protection Regulation in the European Union. These laws place restrictions on how health data can be collected, shared, and used. Other information does not have the same level of legal safeguards. Third, health information could potentially be used to discriminate against people in areas such as employment, insurance, and so on. This type of discrimination is legally prohibited, but the risk remains owing to the sensitive nature of the data. Other data, such as social media posts, have less potential for this type of discrimination. Finally, breach of health information is considered very serious, given the sensitivity of the data. Strong security protections are needed, and breaches can carry heavy penalties. Breaches of other types of data may not have the same level of severity.

Regarding privacy on social media, there are some key characteristics of the concerns around Twitter compared with other platforms. First, most Twitter content is public by default, whereas other platforms such as Facebook allow more privacy controls. This can raise concerns about a lack of control over dissemination. Second, tweets are often archived and searchable indefinitely; therefore, there are concerns about permanent availability even for “deleted” content. Other platforms may have more ephemeral sharing. Third, the open nature of Twitter makes it easy for tweets to spread rapidly and become viral compared with platforms such as Instagram, where sharing can be more controlled. This raises concerns about loss of context and lack of containment. Finally, the ability to create anonymous accounts on Twitter is greater than that on platforms such as Facebook that require real identities. This raises concerns about harmful speech, misinformation, and so on.

We proposed the following research framework for disclosing general and specific health information on Twitter by integrating 2 aspects of information privacy concerns (Figure 1). As several studies may have found empirical evidence for the hypotheses proposed in this study, we need to clarify what is new in our study. First, this study integrated both aspects of privacy concerns for the first time in a model. Previous studies examined either privacy concerns related to companies’ practices with web-based information (CFIP) [62] or concerns related to the web-based behaviors of peers (PrPC) [63]. However, as mentioned in the previous section, individuals may be concerned about disseminating their health information on Twitter because companies’ collection practices and web-based peers’ behaviors could violate their privacy. In this study, we wanted to examine whether both aspects of privacy concerns (ie, CFIP and PrPC) can collectively change health-related information-sharing decisions or whether one can dominate the other. For instance, whether the nature of the relationships and contexts with 2 different information trustees (ie, vendors and peers) can shape information dissemination behavior. Second, as Twitter is considered as a rich database for collecting individual health-related information to examine sentiments and manage public health [64] and reports highlight that individuals may be concerned about web-based interactions with peers [65], Twitter would be the best research context to meet the goals of this study. Third, this study distinguished between general and specific health information. Thus, we could offer more insights about privacy concern levels and disclosure behaviors related to the 2 types of health information on Twitter. These 3 reasons can make our study different from previous studies in the privacy literature.

In addition, we controlled for several variables such as age, gender, education, Twitter experience, privacy violation experience, and misrepresentation of identity on Twitter. According to previous studies in the privacy concern domain, some demographics, such as age [66], gender [67], and education level [68], can affect people’s intention to disclose information on social media. Moreover, the impacts of these variables have been examined in previous studies investigating individuals’ perceptions about sharing eHealth-related information [69,70]. The effects of these variables are often controlled in previous studies in the field of information privacy threats [71]. Thus, we assumed that individuals of different ages, genders, and educational levels engage in various disclosure behaviors because they have diverse backgrounds, individual characteristics, and personal differences. Therefore, we considered these demographics to be control variables in the proposed research model.

Moreover, the effects of misrepresentation of identity, experience with technology, and privacy violation experiences are controlled in previous studies examining relationships between privacy concerns and self-disclosure [22,42]. Thus, it is believed that individuals with different privacy violation experiences, previous identity misrepresentation, and experiences with Twitter are more likely to demonstrate various disclosure behaviors. Therefore, we treated these experience-related variables as control variables in our model.

We administered a web-based survey questionnaire to achieve the defined objectives and test the proposed model and research hypotheses. The survey consists of 4 sections. In the first part, the purpose of the study is described clearly, and a qualifying question is used to select respondents. The question for filtering respondents attempts to screen individuals with a Twitter account. Thus, individuals without a Twitter account are excluded from data collection and analysis. In the second section, respondents are asked to express their perceptions about privacy concerns associated with companies and third parties, peer-related privacy concerns, and health information dissemination behaviors. In the third section, demographic questions (ie, age, gender, education, income, and race) are asked. Finally, the last section focuses on personal privacy experiences (ie, Twitter experience, privacy violation experience, and misrepresentation of identity).

Questions to measure each construct were adapted from validated instruments available in the existing literature. Slight changes in the wording were made to fit the context of this study. We adapted items to measure CFIP (as a second-order construct with 4 dimensions) from the study by Stewart and Segars [40]. Following Zhang et al [22], we also conceptualized and measured PrPC as a second-order reflective construct with 4 dimensions. Previously defined scales to measure general and specific health information disclosure were adapted from the study by Hsu et al [72]. Respondents rated all the measuring items included in the survey using a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Multimedia Appendix 1 shows the questions used in the web-based survey.

Data were collected in April 2022 by uploading the questionnaire to Amazon’s Mechanical Turk (MTurk). MTurk is a crowdsourcing platform that enables researchers to access data from potential target samples to conduct a study. MTurk has been recognized as an acceptable web-based means for collecting individual-level data. Literature about health care analytics shows a growing number of studies using MTurk for health-related research [73]. Previous studies highlight that MTurk can measure individual perceptions in various domains, such as social media [74]. As the target population of this study was US citizens who use Twitter for web-based interactions, we limited the respondents’ location to the United States. Moreover, 2 attention-check questions were used to remove participants who chose answers without correctly replying to reverse-coded filler items [75]. The filtering questions were as follows: (1) It does not bother me that my peers may try to influence me through comments on my health-related postings on Twitter and (2) I am not concerned that I have little control over who can start a health-related conversation with me on Twitter. We received 364 questionnaires and excluded 35 (9.6%) that were either incomplete or failed the response quality questions, resulting in 329 (90.4%) valid and usable responses. The average response time to complete the questionnaire was 12 minutes. The descriptive statistics for demographics were performed using SPSS (version 26; IBM). The research model was tested using AMOS (version 26; IBM) within the structural equation model framework.

The institutional review board of Florida International University reviewed and approved the study (approval 112755). According to the institutional review board approval, written informed consent to participate in the study was obtained from all participants. Moreover, the data collected in this study were anonymous. We considered US $1 as an incentive for each respondent to participate in the study.

We used confirmatory factor analysis to assess convergent and discriminant validity. Table 1 shows the results of the convergent validity test. The standardized factor loadings for all constructs exceeded 0.7, which is the acceptable range for factor loadings [76]. The composite reliability values and Cronbach α values were above the recommended value of .7, demonstrating the adequate reliability of the constructs [77]. All the values of average variance extracted (AVE) exceeded 0.5, which is the cutoff value [78]. These measures indicated the acceptability of the measurement model’s convergent validity.

Table 2 shows the discriminant validity of the constructs. All diagonal values (square roots of the AVEs) were >0.7 and greater than off-diagonal values (correlations) between any pair of constructs [79]. Thus, the discriminant validity requirements were satisfied for the research model.

Moreover, we checked the convergent and discriminant validity of the second-order constructs. The composite reliability, Cronbach α, and AVE values for CFIP were 0.91, .88, and 0.64, respectively, and these measures for PrPC were 0.94, .89, and 0.72, respectively. The correlation between the second-order variables (eg, CFIP and PrPC) was 0.58. Finally, the square roots of the AVEs for both constructs were >0.7 and higher than the correlations between the constructs. These results confirm an acceptable convergent and discriminant validity for both second-order constructs in the model.

Table 3 shows the participants’ characteristics. The descriptive statistics demonstrate that respondents were fairly distributed across gender, where 52.9% (174/329) were men and 47.1% (155/329) were women. The age range was positively skewed, indicating that most participants were young, with a range between 25 and 34 years (155/329, 47.1%) being high, followed by the range between 35 and 44 years (102/329, 31%). Approximately half (178/329, 54.1%) of the respondents had undergraduate or graduate education levels, which aligns with previous studies highlighting that people with high education levels tend to search more often for web-based health information [80]. The annual income was fairly distributed, with income between US $60,000 and US $79,999 showing a high range (135/329, 41%) among the provided categories. Most respondents were White (174/329, 52.9%), followed by Hispanic and African American individuals. Approximately half (174/329, 52.9%) of the respondents reported using Twitter for 4 to 6 years. Overall, 52% (171/329) of the respondents indicated that they had a privacy violation experience at least once (for instance, their account was hacked), and 38.9% (128/329) mentioned that they tried to use a fake account on Twitter (at least once).

When implementing second-order variables in a measurement model, there are 2 common approaches: the repeated items approach and the 2-step approach [81]. This study used a repeated items approach to measure reflective second-order constructs. In the repeated items approach, the indicators used to measure the second-order construct are included in the measurement model twice: once as indicators of the second-order construct and once as indicators of the corresponding first-order constructs [82]. This approach allows for a direct assessment of both the second-order and underlying first-order constructs in a single measurement model. The repeated items approach provides a holistic view of the measurement model by simultaneously assessing the second-order construct and its underlying dimensions [83]. Using the repeated items approach provides a more integrated perspective about how CFIP and PrPC are influenced by their respective first-order constructs. It allows for a direct examination of the relationships between the second-order construct and its underlying factors.

Both CFIP and PrPC are conceptualized as second-order reflective constructs, consistent with existing literature. A reflectively measured construct shares a common theme across subdimensions; the dimensions are expected to be highly correlated [84]. Table 2 shows that, consistent with reflective measurement, the 4 dimensions of CFIP (ie, collection, unauthorized secondary use, improper access, and errors) are highly correlated with each other. As expected, the 4 dimensions of PrPC (ie, psychological, communication, virtual territory, and peer-related information privacy concerns) are also highly correlated. Results show that the 4 dimensions of CFIP, as first-order factors, load significantly on the second-order construct. The loadings were 0.91 for collection, 0.80 for unauthorized secondary use, 0.95 for improper access, and 0.88 for errors. Therefore, the interaction among 4 dimensions reflects CFIP, which shares a common theme of losing control over information privacy owing to companies’ sharing behaviors. Furthermore, the 4 dimensions of PrPC also load significantly on the second-order construct. The loadings were 0.90 for psychological privacy concerns, 0.92 for communication privacy concerns, 0.86 for virtual territory privacy concerns, and 0.95 for peer-related information privacy concerns. Thus, interactions among these 4 dimensions represent PrPC, which exhibits a shared theme of privacy concerns about losing personal control owing to web-based peer behaviors.

Consistent with privacy literature, we controlled variables such as age, gender, education, years of experience, privacy violation experience, and misrepresentation of identity in the structural model [42]. Findings demonstrate that when the control variables are present, the coefficients and R² change significantly. Specifically, when age (β=−.12; P=.01, education (β=.19; P=.003), and privacy violation experience (β=−.58; P=.008) are present in the model, they significantly influence health information disclosure. Thus, the findings confirm that young people with high education levels who have not experienced privacy violations are more likely to disclose health information on Twitter. However, no effects of gender, years of experience, and misrepresentation of identity were found on health information–sharing behaviors. We used the structural equation model technique to analyze the factors affecting health information disclosure on Twitter. The results of model fit indexes exhibit a good fit with the goodness-of-fit indexes (χ²₃₅₃=2.2; goodness-of-fit index=0.84; adjusted goodness-of-fit index=0.81; comparative fit index=0.90; normed fit index=0.91; incremental fit index=0.90; standardized root mean square residual=0.03; and root mean square error of approximation=0.04) where all indexes meet their recommended cutoff values [85]. Table 4 depicts the summary of path analysis for 4 hypotheses (ie, H1A, H1B, H2A, and H2B).

Figure 2 shows the standardized path coefficients of the structural model. Support is not found for H1A, which proposes that CFIP significantly influences general health information disclosure on Twitter (β=−.07; P=0.16). In contrast, the findings support H1B by confirming that CFIP significantly attenuates sharing behaviors when disclosing specific health information on Twitter (β=−.43; P<.001). H2A, which posits that PrPC would directly affect the disclosure of general health information on Twitter, is supported (β=−.38; P<.001). The analysis also exhibits that PrPC negatively shapes the sharing of specific health information on Twitter (β=−.72; P<.001), and this significant relationship supports H2B.

Regarding H1C and H2C, an alternative model was created for each hypothesis, and the 2 relationships in that hypothesis were constrained [86]. Next, a 2-tailed t test was used to compare the difference between the alternative and the original model. H1C posits a significant difference between the impact of CFIP on general and specific information–sharing behaviors. As the t value was significant (t₁₆₅=3.45; P<.001), we confirm that CFIP imposes a more negative effect on specific health information dissemination than on sharing general health information on Twitter. In addition, H2C proposes that people may disclose more general health information than specific health information owing to peer-related privacy concerns. The t value was significant (t₁₆₅=4.72; P<.001); thus, the effect of PrPC was more prominent in specific health information sharing than in disclosing general health information on Twitter.

Finally, the model explains 41% of the variance in general health information disclosure and 67% in specific health information sharing on Twitter. The R² scores suggest that the 2 aspects of information privacy concerns (ie, concerns about the web-based practices of companies and peers’ behaviors) can provide reliable explanatory power to predict the variance in sharing general and specific health information.

Information sharing is one of the most important objectives of social media. People use Twitter for conversation, and information sharing can initiate a web-based exchange of ideas about an issue. As health information is more sensitive than other types of personal information, disclosing such data can raise privacy concerns. Most previous studies have mainly focused on privacy concerns related to companies and vendors as they may collect and use individuals’ personal information for other purposes or may not properly protect the collected information [87]. Few studies have also explained privacy concerns related to the web-based behaviors of peers [22]. However, previous literature did not consider both sides of information privacy concerns in a model in the context of social media. Moreover, disclosure behaviors on social media can be contingent upon the type of health information owing to sensitivity levels. Few studies have examined the sharing behaviors based on the unique characteristics of general and specific health information [86].

Although both antecedents (CFIP and PrPC) have been examined separately in previous studies, this study’s findings could propose scientific novelty. The study differs from previous research in this field because we integrated 2 aspects of privacy concerns (eg, related to companies and peers) to investigate the disclosure of general and specific health information on Twitter. In this study, we examined whether both aspects of information privacy concerns can jointly influence sharing decisions related to health-related information or whether the effect of one aspect can be overshadowed by the other; for instance, whether the nature of the relationships and contexts with 2 different information trustees (ie, vendors and peers) can influence people to share their health information on Twitter. The findings indicate that privacy concerns related to companies play a more significant role in predicting specific health information than in predicting general health information. Privacy concerns related to companies’ practices reflect the collection and misuse of health information by vendors, such as concerns about using health information for data mining and research purposes [88]. Our findings demonstrate that Twitter users are more concerned about vendors and companies using or sharing their personal health information than general health information. Thus, the dimensions of CFIP (ie, collection, unauthorized secondary use, improper access, and errors) become salient only when people want to disclose specific health information (such as information about their chronic diseases, signs, and symptoms or personal health status). Sharing public health information about hospitals, medical costs, and medications is not significantly affected by concerns about how companies may use such information. A plausible justification is that general health information cannot reflect any personal information associated with an individual, and even if it is used for data mining or big data analysis, it will not violate the user’s privacy needs. Consistent with previous studies [89], individuals may deliberately want to share general health information on social media to increase public awareness and knowledge about a medical situation, such as COVID-19 symptoms and vaccination. Regardless of information accuracy or misinformation, users may engage in sharing their general medical knowledge and public information about treatment options with almost no or minor privacy concerns related to companies and vendors’ collection and use practices.

Our results also show that peer-related privacy concerns can significantly shape both general and specific health information sharing on Twitter. Although Twitter is not the same as web-based health communities designed to share health information, many individuals use tweets to share personal and public health information [90]. Web-based interactions with peers may affect their sharing behaviors as they may feel unable to control who can see, comment on, or exchange the health information they share on Twitter. The dimensions of PrPC (ie, psychological, communication, virtual territory, and peer-related information privacy concerns) are important factors in predicting how users may disclose public and personal health information. However, peer-related privacy concerns are more intense for sharing personal than general health information. This finding indicates that when a user wants to share public information about a disease (for instance, cancer, COVID-19, or HIV), they are still concerned about how peers relate such general information to their profile. This concern becomes more salient when the user decides to reveal personal health information about that disease, such as what treatments or medications they are using daily or what medical procedures they will undergo in the future. Previous studies associate sharing personal health information related to physical health problems or mental disorders on web-based P2P networks with stigma [91]. The more sensitive the health information, the more stigma is attached to sharing such information with peers. Thus, being judged by peers (close friends and other web-based users) because of sharing personal health information may prevent them from disclosing that content on Twitter.

Although results show significant impacts of both aspects of privacy concerns on sharing specific health information on Twitter, peer-related privacy concerns are leading factors in shaping personal health information disclosure, more so than privacy concerns associated with companies and third parties. This result confirms the importance of web-based interactions with peers on social media and how to deal with their sharing behaviors, such as commenting or tagging [60]. This finding implies the critical effects of Twitter friends and the circle of people who can see and share tweets about private health information. Individuals may be first concerned about the Twitter circle and how peers would react to the shared personal information about health status and then become worried about how many companies may access such data and how they would use or share them. Thus, secondary dissemination of personal health information by web-based peers through liking, reposting, retweeting, or commenting on posts is more challenging to the maintenance of privacy controls than secondary use of data or unauthorized access to such private information by companies and vendors. Our finding that peer-related privacy concerns have a strong impact on health information sharing compared with privacy concerns associated with companies and third parties offers a counternarrative to prevalent assumptions in digital privacy research. This could be attributed to Twitter’s highly interactive and public nature, which might accentuate peer-related concerns. Previous studies, mainly those conducted in the context of web-based shopping or general social media use, might have overestimated the role of concerns associated with companies and third parties owing to the commercial and private nature of these web-based activities.

This study may offer some theoretical contributions. First, our findings have implications for information privacy research in social media by integrating the existing privacy concern perspectives. This study can open up the discussions through which privacy needs related to companies, third parties, and peer-related aspects can be addressed. Then, this comprehensive mechanism may strongly affect users’ sharing behavior patterns. Second, this study distinguishes the differences between sharing public and private health information on Twitter. Although disclosing specific health information may help share personal experiences related to various medical situations that could be useful for peers, it is more challenging than disseminating general health information. The findings demonstrate how company-related and peer-related privacy concerns could shape the 2 types of information-sharing behaviors. Third, this study investigates the effects of information privacy mechanisms on health information sharing in the context of Twitter. The findings can promote discussions about health information disclosure in other P2P networks such as other social media platforms, virtual worlds, or Metaverse. Fourth, exploring the determinants of information sharing regarding different types of privacy concerns can expand our current understanding of knowledge acquisition. As sharing both general and specific health information on social media can contribute to people’s medical knowledge, addressing the barriers to specific health information sharing and removing the privacy challenges can significantly help the procedures of medical knowledge acquisition from web-based interactions with peers.

The study contributes significantly to the theoretical understanding of privacy concerns in web-based health information sharing. The evidence that peer-related privacy concerns influence more strongly than those related to companies and third parties highlights a potential oversight in theoretical perspectives. Current theories largely view companies as the predominant source of digital privacy concerns, and this may need re-evaluation. The results extend existing theories by emphasizing the role of peer interactions in privacy concerns, particularly in public and highly interactive web-based environments such as Twitter. This recognition of the social dimension of privacy concerns could be integrated into existing theoretical models to provide a more comprehensive framework for web-based privacy behavior. Furthermore, although our study is specific to Twitter and health information, the insights gained may have broad applicability. The potential role of peer-related privacy concerns could be a valuable area of exploration in other social media contexts and in sharing other types of sensitive information. Thus, our findings open up new avenues for theoretical exploration and suggest a need for further studies to fully understand the complexities of privacy behavior in the digital age.

Unlike other research approaches, such as experiments, observational data, or qualitative interviews to assess privacy concerns and information sharing, our study used a quantitative survey approach. This allowed us to capture data from a large and more diverse sample, providing a more robust and generalizable understanding of privacy concerns in web-based health information sharing. The strength of our quantitative approach lies in its ability to establish clear patterns and relationships among various factors influencing privacy concerns. This enabled us to derive a more comprehensive and systematic understanding of the factors that significantly influence privacy concerns and health information sharing on Twitter. In terms of comparison, our findings offer a novel perspective about the role of peer-related privacy concerns in shaping web-based health information–sharing behaviors. Previous studies have predominantly focused on company-related and third party–related privacy concerns. However, our study highlighted the paramount influence of peer-related privacy concerns, thus suggesting a reorientation of focus in subsequent research efforts in this area. Our study also provides quantifiable evidence about the relative influence of peer-related privacy concerns and such privacy concerns associated with companies and third parties on Twitter users’ health information–sharing behaviors. Such quantifiable insights could serve as valuable benchmarks for future studies seeking to measure and compare similar variables in different contexts or on different platforms. Our survey methodology, coupled with a comparative analysis of the findings, underscores the contribution of our study to the field, offering both nuanced insights and broad trends that enrich our understanding of privacy concerns and health information sharing on social media platforms.

This study also provides several practical and technical implications for promoting privacy protection on Twitter. To promote the sharing of specific health information, it is essential to address privacy concerns related to both companies and peers. However, addressing peer-related privacy concerns is vital to encourage the disclosure of general health information. This is because concerns related to companies and third parties do not significantly predict general health information sharing. Thus, a robust privacy policy cannot be developed regardless of information type. As the 2 types of health information require different ways of satisfying privacy needs, mechanisms and regulations facilitating general and specific health information sharing cannot be the same. Depending on the type of health information, it is essential to customize the ability of Twitter users to control their self-concept and meet different privacy protection requirements. General procedures and privacy policies to regulate the dissemination and use of personal posts are not sufficient to address the information privacy concerns. Twitter should allay users’ privacy concerns about sharing specific health information using advanced technology and management mechanisms. For example, Twitter can enable individuals to restrict access to their shared personal health information. Punitive regulations can be established for inappropriate behaviors (such as retweeting without consent) that may discourage sharing specific health information. All controlling mechanisms and privacy protection functionalities should be easy to understand and use and should not be an additional burden on the users.

As the 2 aspects (company-related and peer-related aspects) of privacy concerns manifest in several dimensions, different features can be developed to address the need for effective protection mechanisms. Twitter can add a new feature to tweets, enabling users to identify the sensitivity of posts related to health information. The content will be recognized as a private post if the sensitivity score (eg, calculated based on a scale ranging from 1 to 10) is more than average. Then, that post is automatically restricted from exposure to everyone, and users can share their thoughts and experiences only with a small crowd. Users can also define terms and conditions for peers who want to retweet sensitive posts. For instance, a “request for share” button can appear for each sensitive post, and peers can only share the posts when they get approval from the focal users. Given our findings, Twitter could introduce a feature that allows users to select the audience for their health-related posts, thereby addressing peer-related privacy concerns. They could also introduce a “Health Information” mode that automatically applies high privacy settings for tweets marked as health related. In addition, given the significant role of knowledge in shaping privacy concerns, Twitter should consider educational campaigns or prompts to inform users about these features and the importance of privacy when sharing health information.

In May 2022, Elon Musk called for further investigation into the accuracy of spam and fake account estimates, which Twitter announced to be <5%. Fake and spam accounts could lead to undesirable social interactions with peers, unwanted peer-shared information, and an unpleasant web-based social environment. Twitter needs to use new procedures to detect spam and fake accounts and better control the functionality of Twitter bots to provide a more appropriate web-based social platform. This new mechanism could automatically limit the visibility of private posts containing highly sensitive health information to everyone, even to people who users follow. On the basis of the current Twitter privacy policy, people can mention who can reply to a specific tweet. However, it is hard to confirm who can actually see the posts because of bots and recommendation agents.

Social bots use computer algorithms to artificially create content and interact with people on social media [92]. Twitter bots can be manipulative and purposely change people’s attitudes and opinions about a topic [93]. For instance, bots can share posts with peers who are not following a user but usually read posts with health information content. The existence of bots may be useful for sharing general information but can be very harmful because it can increase exposure to private posts with sensitive health information. A plausible recommendation is to add a new category for sensitive content (such as specific health information) besides the photo, graphics interchange format (GIF), and poll categories. Then, users can create a new circle of people who are allowed to see, reply to, or share these sensitive posts. Users can also customize the configuration and limit the possible unwanted interactions by selecting who can see the posts but cannot share them. This small crowd can be saved for future use and can be easily modified later. The advantage of this new category is that people are notified to customize their Twitter circle depending on different content (eg, highly sensitive, semisensitive, and nonsensitive). For instance, a user can select everyone to see and comment on posts containing information about cryptocurrency, high-technology companies, or humanitarian issues. In contrast, the user can select a group of followers to see their thoughts about general health information and choose only a few close friends to see and comment on posts with sensitive health information.

Stringent privacy policies are required to enable Twitter users to limit who (ie, peers) can view, comment on, and share web-based content. People should be able to easily edit with whom they can share health information to exercise control over their personal digital information. Spambots on Twitter should also be controlled, modified, or filtered because they can involve potentially deceptive, harmful, or annoying activities. A more transparent policy is required to detect and deactivate invasive Twitter bots that can automatically like or retweet users’ postings without consent.

Finally, the insights from our study are not only limited to Twitter but also have implications for other social media platforms where users might share health information. Such platforms should recognize the significant role of peer-related privacy concerns and consider introducing similar audience control features. Health professionals and health-related organizations using social media for patient engagement should also be aware of these concerns and take steps to ensure that their communication respects patient privacy. Policy makers should consider our findings when developing regulations for health information sharing on social media to ensure that they address the most significant privacy concerns.

Our study also has some limitations that can be considered as opportunities for future studies. First, a web-based survey through MTurk was used to collect data, which may be biased toward people familiar with crowdsourcing platforms. Future studies can use other data collection and sampling strategies, such as collecting data directly from Twitter. Second, we collected data from 329 Twitter users, which may not be a good representative of Twitter users. Next, studies can increase the sample size to reduce sampling bias and improve the generalizability of the findings. Third, we did not consider the effects of cultural dimensions (such as individualism, uncertainty avoidance, etc) on sharing health information on Twitter. It can be interesting for future studies to explore the effect of culture on disclosing different types of health information on social media. Fourth, our study tests a model to analyze health information sharing from the perspective of privacy concerns. However, there may be other essential variables. More studies are required to examine other factors inhibiting and promoting sharing behaviors on social media, such as reputation, incentives, trust, stigma, and social support. Fifth, this study did not examine the accuracy of the health information shared on Twitter or the risks of misinformation because it is not within the scope of this study. Future studies could expand upon our results to investigate the role of misinformation risks in information-sharing behaviors. Finally, our study focused on Twitter as a study context. We encourage future studies to extend the proposed model to other social media platforms (eg, Facebook, TikTok, and Instagram) where web-based interactions with peers are essential.

This study provides insights into health information sharing on Twitter from a privacy perspective. The findings propose that including CFIP and PrPC constructs can help in better conceptualization of information privacy concerns in the context of social media. The integration of these 2 aspects of information privacy can expand the discussion about internet privacy by addressing the privacy needs associated with the practices of companies, such as collection, unauthorized secondary use, improper access, and errors. It also considers psychological privacy concerns, communication privacy concerns, peers’ sharing behaviors, and territory privacy concerns related to peers in such interpersonal interactions. This interactive approach can provide a more comprehensive analysis of information privacy (related to web-based vendors and web-based peers) and adds a more substantial explanation of privacy needs on social media channels (such as Twitter). Privacy concerns may not always prohibit disclosure behaviors on Twitter; it depends on the type of health information. The findings demonstrate that peer-related privacy concerns are more salient to predicting general and specific health information sharing on Twitter than privacy concerns related to companies and third parties. The results could propose practical contributions by shedding more light on the negative impacts of web-based peer behaviors on losing personal control over digital communications and information access. Privacy policies should focus on companies’ practices, such as sharing users’ information with third parties for big data analytics. We suggest mitigating privacy concerns and promoting health information sharing on Twitter by creating policies that tailor privacy needs to the type of health information shared (ie, general or specific).