Twitter data related to e-cigarettes were collected through the Twitter streaming application programming interface using e-cigarette–related keywords such as “e-cigarette” and “vaping” [44]. From November 19, 2019, to June 10, 2022, we collected 15,787,239 English tweets related to e-cigarettes. For each tweet, we collected as much information as possible, such as tweet ID, created date, text, retweets, favorites, user ID, user name, user tweets, followers count, friends count, and hashtags. In this data set, we identified 3,617,766 unique Twitter user accounts.

To identify e-cigarette influencers on Twitter, we defined the following selection criteria based on previous studies [32-34]: (1) social media influencers must have at least 1000 followers, (2) the follower and following ratio for social media influencers must be at least 2:1, (3) social media e-cigarette influencers have posted at least 10 e-cigarette–related posts within the past year, and (4) social media e-cigarette influencers should have posted at least one promotional tweet related to e-cigarettes within the past year.

We iteratively checked the users and tweets in our database against the above criteria and filtered out the Twitter accounts that did not match the criteria. We used keyword matching to determine the promotional tweets related to e-cigarettes [44]. Toward this end, we created a similar list that contained the most common keywords related to promotional tweets and we compared all of the tweets of the candidate influencers against the keywords in this list. If any words in the tweets matched any of the keywords in the list, we categorized the tweets as promotional tweets. At the end of the process, we collected 33 accounts that completely satisfied the above requirements and were not from the e-cigarette industry, vape shops, or retailers.

Based on the tweet content related to e-cigarettes, through a group discussion of all four authors (RZ, ZX, QT, and DL), we split the 33 identified influencers into three broad influencer categories. The first category is “vape advocates,” who provide educational content about e-cigarettes and aim to persuade smokers to switch from traditional cigarettes to e-cigarettes owing to multiple health benefits. The second category is “vape reviewers,” who tweet about their reviews of certain e-cigarette products/devices. The third category is “other,” encompassing Twitter accounts owned by other vaping-related groups that are not affiliated with any vaping brands, such as vaper social network platforms or even vape expos. The tweets from all 33 influencers were carefully examined, and the influencers’ categories were assigned by the coders based on the contents of their tweets and account information.

Since we were particularly interested in e-cigarette–related influencer social networks, we decided to group the followers of each e-cigarette influencer into those posting about e-cigarettes and those not posting about e-cigarettes, defined as those who posted at least five e-cigarette–related tweets in the past year and those who had not posted any e-cigarette–related tweets in the past 3 years, respectively. To maximize the differences between these two follower groups, we did not include followers of e-cigarette influencers who posted 1-4 e-cigarette–related tweets in the analysis. Considering the large number of followers for each influencer, we randomly sampled 100 followers who post about e-cigarettes and 100 followers who do not post about e-cigarettes from each of the 33 e-cigarette influencers for further social network analysis.

For each e-cigarette influencer, we constructed two social networks for each follower type. We then compared key measures of these networks, including the average clustering coefficient, in-degree centrality, out-degree centrality, closeness centrality, and betweenness centrality, for each social network. The clustering coefficient of a network measures the degree to which nodes in the networks tend to cluster or form groups; a higher clustering coefficient suggests that nodes within the network are more likely to be connected, which can impact the overall structure and function of the network. In-degree centrality measures the number of incoming edges or connections of a node in a directed network, representing the popularity, influence, or attention received by one node from other nodes in the network. Similar to degree centrality, out-degree centrality focuses on the number of outgoing edges or connections from a node in a directed network, representing the extent to which a node reaches out to other nodes and influences them. Betweenness centrality measures the extent to which a node lies on the shortest paths between other pairs of nodes in a network, whereas closeness centrality measures how close a node is to all other nodes in a network. Both betweenness centrality and closeness centrality have significant control over the network’s communication dynamics and are crucial for maintaining efficient information flow.

Two-sample t tests and generalized linear models were used to compare the social network metrics between followers who do and do not post about e-cigarettes using R statistical analysis software (R Core Team, 2017) with the significance level set at 5%.

We performed topic modeling analysis using the latent Dirichlet allocation (LDA) model on all e-cigarette–related tweets posted by the 33 influencers [45]. Further, we identified 100 unique sampled e-cigarette followers for each of the 33 influencers and collected 3.48 million e-cigarette–related tweets posted by these followers in the past 12 months. After standard data cleaning procedures were applied to all tweets, such as removing hashtags, URLs, and emojis [46], we used the LDA model to obtain the most popular topics [45]. We chose the number of topics based on the coherence score and the intertopic distance.

This study was reviewed and approved by the Research Subjects Review Board of the Office for Human Subject Protection at the University of Rochester (STUDY00006570). This study is a secondary analysis of publicly available Twitter data. All study data have been deidentified before analysis.

Between November 19, 2019, and June 10, 2022, we identified 3,617,766 unique Twitter user accounts that posted at least one e-cigarette–related tweet. Among these, 126 Twitter accounts were determined to be e-cigarette influencer candidates based on our criteria. Among them, 33 Twitter accounts were ultimately characterized as e-cigarette influencers, while the remaining 93 were vape shop or company accounts.

Among the 33 e-cigarette influencers, 10 belonged to the category of “vape advocates,” 19 were “vape reviewers,” and four were classified in the “other” category. “Vape advocates” (14,748 followers on average) had more followers than found in the “vape reviewers” (7426 followers on average) and “other” (4683 followers on average) categories. By comparison, “vape reviewers” had a higher percentage of followers that also post about e-cigarettes (1152/7426, 15.5%) than found for the influencer categories “vape advocates” (1357/14,748, 9.2%) or “other” (398/4683, 8.5%). e-Cigarette influencers categorized as “vape reviewers” and “vape advocates” had almost the same proportion of e-cigarette–related tweets at 53.1% (8205/15,454) and 53.2% (8,668/16,295), respectively. Influencers from the “other” category had a slightly higher proportion of e-cigarette–related tweets at 55.7% (3035/5449).

Further examination of all 19,908 e-cigarette–related tweets posted by the 33 e-cigarette influencers identified three major topics discussed (Table 1). The most popular topic was “advocating against vaping policy,” followed by “vaping helps to quit smoking” and “vaping product promotion.”

The followers of the 33 identified e-cigarette Twitter influencers were grouped into those who previously did or did not post e-cigarette–related tweets. Considering the large number of followers (range 1097-49,215) for each influencer, we randomly selected 100 followers from each group (with or without e-cigarette–related posts of their own) for each influencer to create individual social networks based on their interactions (how they follow each other). To better illustrate the density differences between the social networks of the two types of followers, we randomly selected three e-cigarette influencers to demonstrate their sampled followers’ social networks. As shown in Figure 1, for any of the three representative influencers, the social network of followers who post about e-cigarettes had more connections (edges) than that of followers who do not post about e-cigarettes.

As shown in Table 2, the average clustering coefficient for the networks of followers who post about e-cigarettes was significantly higher than that of the networks of followers who do not post about e-cigarettes. Therefore, the networks of followers posting about e-cigarettes have more connections than those of followers not posting about e-cigarettes for these 33 e-cigarette influencers. In addition, other social network metrics were significantly higher for the followers who post about e-cigarettes compared to those of followers who do not post about e-cigarettes, including in-degree centrality, closeness centrality, betweenness centrality, and out-of-degree centrality (Table 2). Therefore, the social networks of followers of e-cigarette influencers who also post about e-cigarettes have much more incoming and outgoing connections than those of the social networks of followers who do not post about e-cigarettes.

In addition, we compared the number of followers and their tweets in the social networks of the two e-cigarette influencer follower groups. The average number of follower accounts for followers who post about e-cigarettes was significantly higher than that for followers who do not post about e-cigarettes (322 vs 201; P<.001; see Multimedia Appendix 1). Furthermore, the average number of tweets posted in the past 36 months by followers who post about e-cigarettes was higher than that of followers who do not post about e-cigarettes (93 vs 43; P<.001; see Multimedia Appendix 2). Considering the higher number of followers and more tweets in the networks of followers who post about e-cigarettes, we decided to fit a generalized linear model to account for these differences. After controlling for the number of followers and tweets, compared to that of the networks of followers who do not post about e-cigarettes, the clustering coefficient of the network of followers who post about e-cigarettes was significantly higher (β=.26, P<.001).

Social network measures of the followers who post about e-cigarettes differed between vape advocate and vape reviewer influencers (Table 3). The followers of vape reviewers had higher clustering coefficients than those of vape advocates, which indicates that the e-cigarette–posting followers of vape reviewers have tighter connections than those of vape advocates. However, the network of followers of vape advocates had significantly higher in-degree centrality and closeness centrality compared to those of the networks of followers of vape reviewers. This finding indicates that, on average, the e-cigarette followers of vape advocates have better information flow between each other and a greater amount of incoming information compared to those of followers of vape reviewers. We did not observe a significant difference in the betweenness centrality and out-of-degree centrality between the followers of vape reviewers and vape advocates.

Since we found that the followers who post about e-cigarettes were more likely to follow each other than the followers who do not post about e-cigarettes for the 33 e-cigarettes influencers, we further wanted to understand the types of information that these followers were sharing about e-cigarettes. As shown in Table 4, we identified two major topics from e-cigarette–related tweets posted on e-cigarette social networks. The relatively more popular topic was promoting e-cigarette products or vaping activity (74,572/133,881, 55.7%), followed by discussing the ability of vaping to help with smoking cessation and harm reduction (59,309/133,881, 44.3%).

In this study, with the identification of 33 influencers related to e-cigarettes on Twitter, we investigated the social network differences between followers of these influencers who do and do not post about e-cigarettes on Twitter. We showed that the social networks of followers who post about e-cigarettes have significantly more connections and higher density than those of followers who do not post about e-cigarettes, as indicated by the higher network clustering coefficient. The social networks of followers who post about e-cigarettes also had higher closeness centrality and betweenness centrality than those of followers who do not post about e-cigarettes, which suggests that the social networks of the former group are more efficient in terms of information dissemination, coordination, and influence. There were two main topics shared in the social networks of the followers who post about e-cigarettes, including vaping helping with smoking cessation and harm reduction and promoting e-cigarette products and vaping activity.

With thousands or even millions of followers, social media influencers can have a significant influence on their followers, especially with respect to behavior changes. Therefore, it is crucial to understand how these influencers might engage with their followers and how they influence their followers through social network analysis. First, it is important to understand who these influencers are. A previous study investigated the influence of e-cigarette influencers on their followers on Instagram, showing that only 2 of 55 e-cigarette influencers used the word “influencer” in their descriptions, instead opting to use words such as “public figure,” “brand ambassador,” “promoter,” “video creator,” “artist,” “photographer,” “blogger,” “model,” or “fitness lover” to describe themselves [32]. In this study, we characterized e-cigarette influencers into different categories, including vape reviewers, vape advocates, and others. We did not observe any obvious difference in the hashtags used in their e-cigarette–related tweets, and most of them were related to vaping (Multimedia Appendix 3). Our topic modeling results of the tweets posted by the 33 influencers showed they were advocating against vaping policy, supporting vaping to help smoking cessation, and promoting vaping products on Twitter. In addition, based on the social network measures, we showed that influencers that advocate for vaping can spread their information to more general Twitter users, as indicated by high in-degree and closeness centrality values. However, influencers that review certain e-cigarette products attract more followers who post about e-cigarettes themselves, as indicated by a high clustering coefficient. Together, these results show that the vape reviewers’ followers tend to have more connections with each other, while the vape advocates’ followers tend to have more efficient information spread among each other.

To understand how social networks might be associated with user behavior, we grouped the influencers’ followers into those who post and do not post about e-cigarettes. Our results showed that the network of followers who also post about e-cigarettes had a higher clustering coefficient than that of the network of followers who do not post about e-cigarettes, suggesting that followers who post about e-cigarettes form a denser network (more connections) than followers who do not post about e-cigarettes. While we do not know if this tight connection among followers who post about e-cigarettes is due to their shared interest in e-cigarettes, the messages about e-cigarettes are much easier to share among these followers. In addition, besides the connection between the e-cigarette influencers and followers who also post about e-cigarettes, we found connections between the e-cigarette influencers themselves. For the 33 identified e-cigarette influencers, the number of followers who were influencers ranged from 0 to 17 (Multimedia Appendix 3). Therefore, users with similar interests tend to form a more closed network or community on social media. A previous study showed that social networks with active posting of alcohol-related messages were significantly associated with more frequent alcohol use [47]. It is plausible to speculate that the denser network for followers who post about e-cigarettes might indicate a greater influence on vaping behaviors, such as being more likely to vape and more frequent vaping.

Recognizing the significance of social networks on social media in disseminating e-cigarette–related messages, tobacco companies and vape shops have used social media influencers to promote their e-cigarette products. A previous study showed that, on average, each e-cigarette influencer on Instagram was sponsored by more than 10 different e-cigarette brands such as Voopotech, Innokin, and Geekvape [32]. Moreover, despite the FDA’s efforts to reduce e-cigarette advertisements targeting adolescents on social media, only 25% of the sampled e-cigarette influencers indicated age restrictions to access their e-cigarette promotional posts on Instagram [48]. These influencers on Instagram are very impactful, and their promotional posts through interconnected social networks can reach tens of thousands of adolescents. While more attention has focused on the influence of social media influencers on promoting e-cigarettes, it is important to understand how these influencers influence their followers through social networks. In this study, we showed that, on average, the followers who post about e-cigarettes have more connections in the sampled 100-node network compared to those of followers who do not post about e-cigarettes. The dense network might help disseminate e-cigarette–related promoting messages to more users in the social network, which can be a potentially effective strategy to effectively communicate health education messages with the public for future digital tobacco education campaigns.

Provaping messages are dominant on social media [49-51]. One study demonstrated that while the FDA’s sponsored antivaping hashtag “TheRealCost” only appeared 50 times a month on Instagram, provaping hashtags such as “e-juice” or “e-liquid” appeared more than 1000 times [52]. Another study showed that despite the FDA’s requirement to add warning labels on e-cigarette advertisements on social media, most e-cigarette promotional images have no warning labels [53]. Our study showed that all e-cigarette–related influencers were promoting vaping as a smoking cessation tool and various vaping products on Twitter/X. Besides posting provaping messages, vaping advocates also posted messages advocating against tobacco regulatory policies, as shown in the major topics discussed by the e-cigarette influencers on Twitter/X. The vaping advocates directly contribute highly negative comments to FDA communication efforts, such as the FDA flavor enforcement policy and the proposed rules on menthol cigarettes [30,54-57]. These antiregulatory posts on social media might have a negative effect on the effectiveness of the FDA regulatory efforts. In addition, we noticed that some tweets from e-cigarette influencers are spreading misinformation about e-cigarettes, such as vaping being harmless, which is consistent with findings from other social media studies [58-60]. The spreading of this misinformation about e-cigarettes on social media, especially within social networks or communities, might undermine public understanding of the health risks associated with e-cigarettes, thereby promoting the initiation or continuation of vaping.

In our study, two topics from the networks of followers who post about e-cigarettes were vaping helping with smoking cessation and harm reduction and promoting e-cigarette products and vaping activity. The most popular hashtag we found in this social network was “WeVapeWeVote” (Multimedia Appendix 3), an organization fighting to protect the rights of adults to access vapor devices and other smokeless alternatives. The Hashtag “Vapefam,” used by someone who views the vaping community as an extended family, has also been mentioned more than 65,000 times in our data set. It is reasonable to speculate that these dominant provaping messages on social media might push back against current efforts of vaping regulation policies (such as the FDA’s flavor enforcement policy) that aim to reduce the current vaping epidemic in youth, which should draw more attention from public health authorities. Further, more educational warning messages about the health risks of e-cigarette use should be encouraged and promoted on social media, which future tobacco education campaigns should consider. In contrast, as shown in Multimedia Appendix 3, the top 10 hashtags in tweets posted by the followers who do not post about e-cigarettes were more likely to be related to health, such as #health and #selfimprovement.

There are several limitations of this study. First, the definition of followers who post about e-cigarettes was based on Twitter user accounts that have posted at least 10 e-cigarette–related tweets within the last 12 months. However, it is uncertain whether these posters are also e-cigarette users. In future studies, we will determine if they are actual e-cigarette users through use of a human-guided deep-learning model. Second, other factors (such as the demographics of Twitter users) might partially influence the average clustering coefficients, which need to be controlled in the future. In this study, we did not investigate the impact of social networks on user behaviors, which can be further measured in the future through a longitudinal study. Finally, the social network is dynamically evolving. Therefore, it is important to monitor the social network longitudinally to study how it evolves and influences user behaviors.

Through social network analysis, we showed that the social networks of followers of e-cigarette Twitter influencers who themselves also post about e-cigarettes have a denser connection compared to that of followers who do not post about e-cigarettes, which helps us better understand how e-cigarette influencers influence their followers through social networks. More importantly, social networks on social media might provide a novel and effective intervention approach for future health education campaigns to disseminate antivaping messages. For example, considering the dense social networks among e-cigarette social media users, public health advocates can join these social networks by following either the influencer or their followers who post about e-cigarettes, which would enable vaping prevention messages to be disseminated quickly in the networks and to specifically target potential or current e-cigarette users.