Original Twitter posts (tweets) were collected between June 1, 2020, and December 31, 2021, from Twitter’s Streaming Application Program Interface using IQOS- and HTP-related keywords and hashtags (eg, “iqos,” “heatedtobacco,” and “heatnotburn”; see Section S1 in Multimedia Appendix 1 for the full list). The initial time period for data collection originally was determined immediately before and 1 year after the US MRTP authorization was made in July 2020. During early data collection in fall 2021, a new regulatory event (ruling of IQOS’s patent infringement against BAT’s Glo and subsequent US import ban) emerged, and the research team decided to expand the data collection window to include when the ruling went into effect in November 2021 (see Figure S1 in Multimedia Appendix 1).

Similar to previous studies [25,29], we excluded all retweets to prevent overrepresenting repeated identical statements. Because of limitations in analytical capacity, we limited our analysis to English-language tweets. Geocodes (indicating where the account is located) were included in the metadata, but because people can opt out of including such information, only a small number of tweets included location information, and therefore, geocodes were excluded from analyses.

The initial search yielded 65,333 unique public tweets collected over 19 months; after removing undiscernible or ambiguous content (eg, non-English or emoji-only tweets), we had 10,423 English tweets. We selected a random subset (n=4138) of them (see Figure S1 in Multimedia Appendix 1 for a flow diagram of the tweet selection process) to facilitate manual content analysis.

We developed a codebook informed by prior research [25,26] and refined it during the initial coding and training process (see Tables 1 and 2 for codebook definitions and tweet distribution; see Figure S2 in Multimedia Appendix 1 for more details about the codebook structure and coding procedure). Previous studies coded emerging themes among tweets, including news, health claims, marketing, personal testimonial, policy, or other [25], and examined overall and across-time distributions of neutral-, positive-, or negative-sentiment tweets [26]. Major updates in this codebook include clarifying mutually exclusive categories of tweets (“personal testimonial,” “news/information,” or “direct marketing/retail”) to better reflect the nature of individual tweets and allow separate analyses within each category. We aimed to better focus on personal testimonials that would more closely reflect population perceptions. In addition, 2 main overarching discussion topics, health and policy, were additionally coded irrespective of whether an individual tweet was categorized as personal testimonial, news/information, or direct marketing/retail.

Tweets were categorized into four mutually exclusive categories: (1) personal testimonials, (2) news/information, (3) direct marketing/retail, or (4) noise (irrelevant or not understandable). We focused our analysis on the 2796 tweets that fit into categories 1-3.

The unit of analysis was the text of the tweet; images were not included in the analyses. When coding the tweets, we only considered information that was explicitly mentioned—that is, content that may have been included in the linked websites but not discussed in the tweets was not considered. Also, many tweets include a long list of hashtags at the end, with many hashtags only loosely related to the actual tweet content or adding any new information. Therefore, we only considered hashtags that were part of the actual tweet sentences (eg, hashtags considered: “Globally, #HTPs help millions of smokers and are a safer alternative”; not considered: “HTPs: next generation of smoke-free alternatives for teens #heatedtobacco #tobacco #prevention”) [30].

The tweets within each category were coded into specific subthemes (not mutually exclusive), using similar approaches from previous research [25,26] as well as via iterative discussion during the coder meetings. Among personal testimonials, six subthemes emerged: (1-a) description of flavor or taste in the context of a personal recommendation, thoughts, or review; (1-b) personal experience using HTPs either favorably or unfavorably (eg, feeling of smoke and taste); (1-c) information seeking about HTPs (eg, troubleshooting and asking about product availability); (1-d) comparison to combustible tobacco; (1-e) comparison to electronic nicotine delivery systems (ENDS); or (1-f) HTP marketing strategies. Personal testimonial tweets were also coded for their valence: positive, neutral, or negative. News/information tweets were coded for four subthemes: (2-a) comparison to combustible tobacco, (2-b) comparison to ENDS, (2-c) tobacco marketing strategies, or (2-d) referring to scientific evidence. Direct marketing/retail tweets were coded for two subthemes: (3-a) flavors and (3-b) product features.

All tweets were further coded for 2 not mutually exclusive topics: health (mentioning health risks and/or benefits of using HTPs; subthemes included comparison to combustible tobacco, comparison to ENDS, and/or cessation) and/or policy (discussing HTP-related tobacco policy or regulation; subthemes included mentions of MRTP authorization, other US-based policy, and/or policy outside of the United States). Tweets were also coded for specific HTP brands (not mutually exclusive): whether they mentioned IQOS, other HTP brands (eg, Ploom and Glo), or referred to general product category.

Three independent coders coded the tweets using a binary system (yes/no) for each category, topic, and subtheme (see Figure S2 in Multimedia Appendix 1). To establish interrater reliability, 3 coders double-coded a subsample of tweets (total 300, approximately 10% of the analytic sample [19,30,31]) through an iterative procedure, where differences across coders in the initial practice set were discussed in real time and coding definitions were amended and refined. Coders applied the refined codebook to the next smaller set and repeated this process until consensus was reached and inter-rater agreement reached at least 90%. Because of a lack of variance for some codes (ie, too many 0’s vs 1’s), κ could not be reliability calculated; therefore, we used percent agreement instead to assess inter-rater reliability. The first author served as an arbitrator resolving discrepancies between the coders. Then the 3 coders each independently coded a third of the remaining tweets.

After completing the coding process, we conducted a descriptive analysis of the tweets according to categories and themes. We also performed a pairwise co-occurrence analysis to assess connections across themes: for example, we examined the proportion of personal testimonials tweets that also discussed health-related content about HTPs. To visualize the co-occurrence matrix containing the counts of co-occurrences of themes, the igraph package version 1.2.6 in R (R Foundation for Statistical Computing) was used [25]. Examining the co-occurrence of themes, separate analyses were conducted within the mutually exclusive categories, and we present results from the personal testimonials and news/information tweets, excluding direct marketing/retail tweets that were the least frequent.

As this study analyzed publicly accessible information only, no ethics board review has been conducted. All URLs and individual Twitter handles were redacted (eg, “[URL]” rather than reproducing what was included in the tweet) to protect privacy. To further protect privacy, posts used in this paper were paraphrased by one of the coders so that the tweets included cannot be easily traced back to original accounts.

The distribution of tweets across the 19-month period is shown in Figure 1, along with several policy and business events. Compared with June 2020, a 50% increase in the number of tweets was observed in July 2020 after the IQOS MRTP authorization. Other peaks in the number of tweets were observed in February, May, and September 2021, related to events surrounding IQOS patent infringement lawsuits and the US import ban. This trend was mostly driven by news/information and policy-related tweets. Health-related tweets were most frequent in July 2020, immediately after the US FDA’s MRTP authorization of IQOS reduced exposure claims. Personal testimonials were most frequent in February and September 2021, coinciding with legal disputes and the US import ban. Direct marketing/retail tweets were generally lower in number, with a peak observed in January 2021.

Figures 2 and 3 display separate co-occurrence analyses within personal testimonial (Figure 2) and news/information (Figure 3) tweets. Among personal testimonials, neutral tweets mentioning IQOS brand (634/1613, 39.3%) showed the largest co-occurrence, followed by neutral tweets discussing policy (378/1613, 23.4%) and neutral tweets about general HTPs (268/1613, 16.6%). There were slightly more health-related positive testimonials (220/1613, 13.6%) than health-related negative testimonials (144/1613, 8.9%). News/information tweets had the highest co-occurrence with IQOS brand (447/862, 51.9%), policy (407/862, 47.2%), general HTPs (393/862, 45.6%), followed by scientific evidence (225/862, 26.1%), US-based policy (223/862, 25.9%), and health (178/862, 20.6%). The thickness of the lines represents the frequency of pairwise co-occurrences, and the size of the circles represents the frequency of theme occurrences.

This study reviewed tweets related to HTPs between June 2020 (1 month before the US FDA’s MRTP authorization for IQOS 2.4) and December 2021 (1 month after the IQOS US import ban took effect) to observe people’s response to major policy decisions related to HTPs. Social media such as Twitter, Reddit, and TikTok have emerged as a major source of news and other information, especially among young people [32,33]. Tobacco industry has also been actively using social media as a promotion outlet for emerging tobacco products [27,28]. A recent study on Instagram suggests that despite the platform’s attempts at regulating tobacco promotions, paid partnership posts are abound promoting venues and events that feature tobacco use, suggesting limited compliance and/or enforcement [34]. Social media use is associated with greater exposure to tobacco marketing and promotions [35] and more favorable perceptions toward and use of nicotine and tobacco products [36,37]. Coupled with concerns related to the lack of regulations and enforcement due to their global nature and strong reliance on user-created content [38,39], a closer surveillance is warranted for social media discourse surrounding emerging tobacco products.

We found that more than half of the tweets were personal testimonials, expressing thoughts and opinions based on one’s own experience and/or knowledge, which provide a valuable window into public opinion regarding HTPs. Close to half of the personal testimonials were neutral, either not expressing clear valence or mentioning both positive and negative perspectives. Among the non-neutral personal testimonials, consistent with a previous study [25], more tweets expressed a positive than negative valence toward HTPs. In positive-valence tweets, people described IQOS as less harmful, “better,” or “safer” than combustible tobacco products, namely cigarettes, or discussed HTPs as being able to help people who smoke quit smoking. Another study examining IQOS-related tweets between November 2019 and August 2020 (2 months after MRTP authorization) found that overall there were more negative than positive tweets, but also found that positive tweets were more frequently observed around MRTP authorization [26]. This study suggests that this trend continued during the year after MRTP authorization. This is also consistent with PMI’s harm reduction marketing strategies in and outside of the United States [3,40,41], including their global public relations efforts emphasizing the US FDA’s MRTP authorization [16,42], and may reflect explicit and implicit marketing influence. However, it should be noted that the US FDA’s MRTP authorization for IQOS was limited to reduced exposure to harmful chemicals and did not include reduced risks of tobacco-related disease, compared to cigarettes [7]. With insufficient evidence on whether or how much IQOS or other emerging tobacco products may reduce the health risks in the short and long term compared to cigarettes [43], such discussions should be closely monitored and, if necessary, corrected using effective communication campaigns.

This study provides recent examination of social media discourse by collecting recent dataset from Twitter. This is also extremely timely given the impending return of IQOS to the US market after a legal settlement between PMI and BAT, which will resolve ongoing litigations [10]. Compared to an early study [25], tweets that directly market or promote HTPs comprised a substantially lower proportion of the corpus (Barker et al [25]: 32.3%; this study: 11.5%). The observed reduction in direct marketing may be a result of an increase in other categories. For example, personal testimonials may have increased as IQOS is known to more people after MRTP authorization and PMI’s continued promotional effort globally. Or, this may reflect the tobacco industry’s shift to more stealthy marketing where paid influencers post promotional content as personal opinion, as the tobacco industry has been historically doing [28]. In addition, our timeline involved multiple major policy events surrounding IQOS. There may have been more policy-related discussions in both the personal testimonial and news/information categories demonstrated by our co-occurrence analysis, resulting in a relative decrease in the proportion of direct marketing/retail tweets. This finding may also reflect relatively lower direct marketing/retail activities on Twitter than other social media platforms, such as Instagram and YouTube, by IQOS official accounts [27].

Direct marketing/retail tweets were observed most often during January 2021. This could be because the HTP retailers may target those who want to stop smoking combustible cigarettes as part of their new year’s resolutions, suggesting switching to HTPs. Such cessation claims have been observed in early marketing and promotion of e-cigarettes [44-46]. However, there were no specific health or cessation claims observed in direct marketing/retail tweets in our analytic set. Most tweets surrounding cessation claims were personal testimonials, in which individual Twitter users discussed their own experience of smoking cessation using HTPs.

It should be noted that HTP marketing is regulated by many social media platforms [39]. Moreover, the US FDA’s MRTP marketing order requires that all digital marketing for IQOS be allowed only with audience age verification and mandates continuous monitoring and reporting of all promotional activities [47]. However, content on social media and other online channels that originates overseas can easily reach an international audience, including people in the United States. Our search, conducted from the United States, included multiple direct marketing/retail tweets that could be seen without an age verification. Content of some direct marketing/retail tweets suggested that many originated from outside of the United States, such as from the United Arab Emirates or the United Kingdom, with some tweets offering “international delivery.” Therefore, despite the US regulation that restricts shipping of tobacco products [48], it should be closely examined whether retailers operating from overseas may be able to circumvent national retail and promotional restrictions [49].

This study adds to existing information on social media discussions about other emerging tobacco products such as ENDS [29,30,50] by examining how an emerging product (ie, HTPs) is discussed in comparison to ENDS and cigarettes. Our findings are consistent with existing research that people who post on Twitter may misjudge the emerging products’ relative risks to other tobacco products [50], as some posts we reviewed suggested confusion between MRTP authorization and approval, and included frequent mentions of reduced health risks compared to cigarettes. This study also adds to existing social media content analyses of HTPs [25,26] by including tweets from more recent time points that overlap with many meaningful policy events (eg, MRTP authorization and US import ban). Although these events’ direct legal impact is limited to the United States, PMI has been using MRTP authorization in their global public relations efforts [42] and lobbying for international tobacco regulatory policy changes [16]. Moreover, a news content analysis in low- and middle-income countries found that half of the 50 identified news articles covering US MRTP authorization during the 6 months after the announcement included misleading “reduced risk” language [51] even though the US FDA’s decision only specifies “reduced exposure” to harmful and potentially harmful chemicals. This may lead to incorrect risk perceptions around IQOS and other HTPs in and outside of the United States and calls for the needs of close surveillance of public opinion and effective health communication campaigns to inform appropriate regulatory policies and interventions globally.

This study also comes with limitations that are not uncommon in similar studies examining social media content. First, this study only considered posts from publicly accessible accounts and may not reflect the attitudes of people with private Twitter accounts. Additionally, because of the analytical capacity, we limited our corpus to English language only. This potentially misses much conversation happening in non–English speaking countries such as Japan, Germany, and Italy, where IQOS seems to have gained greater popularity [52], as well as conversations conducted in non-English languages commonly spoken in the United States (eg, Spanish). Previous research suggests that Twitter users’ information-sharing networks largely cluster and segregate as a function of language and topic [53], and Spanish speakers in the United States may be more likely to receive and share online misinformation in general [54]. However, it is important to note that a recent analysis of Spanish tobacco-related tweets found similar themes to previous English-only studies [55].

Some of the personal testimonials may have come from influencers connected to and sponsored by the tobacco industry, without clear disclosure within the tweet [28,56]. We did not investigate the nature of the accounts posting tweets—if the text reflected a person’s direct or indirect experience and/or opinion toward IQOS, we counted it as a personal testimonial. A similar approach was applied to news/information tweets, where tweets from tobacco industry public relations accounts may have inadvertently been included. A future study should explore the origins of such tweets and how the differences across account types may differentially impact viewers’ perceptions.

The location of tweets (eg, US vs non-US) would have been helpful to better understand the impact of the US-based policy events on global tobacco control policy; however, because of the small number of tweets (approximately 2%) that include location information, such analyses could not be conducted. Future content analyses may provide further insights if they can verify information where the posts originated.

Moreover, our analyses focused on the textual content of tweets, not including external links or images. This may have led to some loss of information. For example, some tweets were vague about which authorization they referred to (eg, “Senate plenary to deliberate regulation of import, manufacturing, sales, distribution, and use of vapes and HTPs”), so it was unclear whether they should be coded as US or foreign policy. It is also possible that some additional themes and subthemes would have emerged if we included images in the coding process [57]. This study also collected data only from Twitter from a specific time period, and therefore the results may not generalize to other popular social media platforms such as Facebook, Instagram, or TikTok that have different format features, platform regulations, and user populations.

Our study suggests that social media content analysis can be a useful tool for public opinion surveillance related to emerging tobacco products, including the sentiment of social media users regarding emerging tobacco products. Such information should be actively used to determine what kind of interventions and communication campaigns should be prioritized. This study categorized and analyzed 3 different types of tweets discussing HTPs: personal testimonials, news/information, and direct sales/marketing. Our findings suggest that after the US FDA’s IQOS MRTP authorization, many Twitter users held neutral or positive perspectives toward HTPs. They perceived IQOS as being less harmful or safer than cigarettes, even though the MRTP authorization was limited to “reduced exposure” claims, not “reduced risk.” Given that many people turn to social media as a source of information, examining news/information tweets also provides insight into the nature of information about HTPs that people are exposed to. This includes some areas for concern, including misrepresentation of MRTP authorization as “FDA approval.” This study also raises awareness about cross-border marketing and sales of HTPs and calls for stricter regulatory policy and effective enforcement on social media platforms globally. This kind of content analysis provides more detailed information driven by the themes of users’ posts and a different perspective from traditional surveys, although continuous research is required to address its representativeness.