We conducted a search to define a longitudinal set of tweets related to menthol cigarettes between February 1, 2012 and January 31, 2013 (shown in Figure 1 [39]). This time period provides a year of data and covers a time period prior to regulatory action on menthol cigarettes globally.

This observational study used data collected by the Health Media Collaboratory (HMC) [40]. The HMC accessed Twitter status updates through the Gnip PowerTrack Firehose (Gnip, Inc) [41], which prospectively collects all Tweets from public Twitter feeds.

The HMC created two archives that collected all Twitter status updates containing the keywords “cig” or “cigarette” (known as the cigarette archive) or the keyword “smoking” (the smoking archive). We chose those keywords after an extensive process of precision testing of terms likely to obtain tobacco-related content (Health Media Collaboratory, Procedures Manual, unpublished, 2012). The Gnip PowerTrack collected the content of the tweet, metadata such as username (who made the tweet), date and time the update was made, and hashtags (user-generated descriptive tags).

At the time of our search for menthol cigarettes, the smoking archive had 32,588,116 tweets and the cigarette archive had 14,528,963. We identified an initial set of keywords including general terms, major menthol cigarette brands, slang terms, and common misspellings that we expected to generate relevant menthol cigarette-relevant content (see Table 1). We filtered the smoking and cigarette archives for each term, resulting in 133,485 tweets.

Once the searches were completed, 2 coders independently coded 100 randomly selected tweets from each keyword or all tweets if less than 100 for menthol relevance. We calculated a kappa statistic for each term to assess interrater reliability. For any term with at least .8 kappa score, one author (SWR) verified the coding and generated a measure of precision (percentage of identified tweets relevant to menthol cigarettes). For any keyword less than .8 kappa score, we coded another 100 tweets until we reached this level of interrater reliability. Of the 42 keywords searched, we achieved a kappa of at least .8 on the first attempt in 41 (98%) keywords.

We used DiscoverText (Texifter, LLC), a cloud-based text analytic software with active machine learning algorithms, allowing human coders to “train” the computer on a coding schema using a selection of tweets. If the keyword generated less than 40% of relevant tweets from the 100-tweet sample, we dropped all tweets with that keyword. If a keyword generated 80% or more relevant tweets in the 100-tweet sample, we kept all tweets. If the keyword generated 40% to 80% relevant tweets, we used the machine learning to assign a probability of a given tweet being menthol cigarette relevant, validated the machine estimates using a second 100-tweet sample coded by human coders, and retained only tweets with greater than 0.8 probability of being menthol cigarette relevant based on the machine classification. Of the 29,957 tweets assessed by the classifier, we retained 68.37% (n=20,483).

After merging and deduplicating these sets of tweets across archives and keywords, we ended with 94,627 unique tweets to include in content coding.

We developed a codebook (see Table 2 for brief description of codes; Multimedia Appendix 1 provides the broader codebook) for directed content analysis [42], building on empirically derived constructs but allowing for emergent categories in the following areas: themes, populations, other tobacco or substances, tweet characteristics, user smoking status, and sentiment toward menthol cigarettes.

We used the codebook to code a random sample of 7000 tweets weighted by prevalence of keywords in the menthol cigarette-relevant dataset. Thus, even tweets with low-prevalence keywords had a probability of being selected equal to their prevalence in the larger dataset. Retweets were coded the same as the original tweet. We coded the entire tweet content of modified retweets, but coded sentiment on the modification alone. Each tweet was coded with as many categories as were applicable.

Next, 4 coders independently coded 500 tweets, reconciling after each 100 and clarifying the codebook, as needed, including adding or broadening codes. Coders also used “other” (write-in) responses to code for new categories that were not in the initial codebook. We added prevalent other categories to the codebook, and reviewed previously coded tweets for the presence of any applicable other categories. Once coders reached kappa of .8 reliability across raters and categories, individual coders coded the remaining tweets independently. Coders flagged unclear tweets, which were adjudicated by a second coder (SWR, CLJ, or MB). We excluded non-English language tweets and those that contained relevant keywords but were not about menthol cigarettes (eg, smoking on Newport Beach, or enjoying Newport, RI).

Of the 7000 coded tweets, we excluded 343 (4.90%) as being either non-English or irrelevant to cigarette smoking or menthol cigarettes, with 95.1% retrieval precision and 90.4% retrieval recall within the HMC archive [46], leaving an analytic sample of 6657 tweets.

For these tweets, we provide univariate statistics of the frequency of tweets and provide qualitative narrative about frequent or notable content in each category. When examining differences in sentiment by other categories, we conducted chi-square tests to assess statistically significant differences in positive or negative sentiment toward menthol cigarettes.

Table 3 shows the frequency of themes, populations, other tobacco or substances, tweet characteristics, user smoking status, and sentiment.

Sentiment varied by category across themes, populations, other tobacco and substances, and tweet characteristics (omitting categories with <1% prevalence) Figure 2 shows. Based on a chi-square test, tweets coded as referencing themes of addiction (including concepts such as craving or desire for a menthol cigarette), smoking, taste or sensation, and industry or marketing; marijuana as an other substance; and tweet characteristics of commercial tweets and song lyrics were significantly more likely to be positive than negative toward menthol cigarettes. However, both smoking and taste or sensation themes had close to 40% negative sentiment. Positive smoking and taste sentiment included positive attributes of smoking (relaxing, relieving stress, taking smoke breaks) and enjoyment of menthol (pleasant minty taste, soothing or cooling sensation). Negative sentiment included negative mood (stress, anxiety, “bad nerves”), negative feelings about other people smoking (“#ItsNotCuteWhen you smoking ports <<<<”), or dislike of the taste or smell of menthol cigarettes (“Smoking a newport. #gross,” “I hate the scent of newports!! Cigarettes in general Smell terrible, but especially newports!”).

Tweets referencing themes of cessation and health, and populations of women, children and adolescents, and African Americans were significantly more likely to be negative toward menthol cigarettes. Almost no cessation-related tweets noted proven quit methods such as the use of FDA-approved cessation aids, talking with a physician, or using a quitline [47]. Many noted plans to quit after smoking a last cigarette or pack, or asked for support for their quit attempt (“Today is my last day smoking cigarettes....waving good bye Newports! Wish me luck!!”). A few noted trying to quit by using an e-cigarette (14/6657 tweets). A small number referenced making a quit attempt in relation to the US Centers for Disease Control and Prevention’s Tips from Former Smokers campaign (“Seriously those commercials of the people with the hole in their throat made me quit smoking I haven’t had a newport in days”) [48]. Some also tried to quit by switching to a nonmenthol cigarette (“I need to get off menthols. I need to quit but if I start smoking something disgusting it may be easier.”).

Most health tweets were negative, noting negative health consequences of smoking. However, positive tweets related to health expressed a misperception of menthol cigarettes as having medicinal use when sick (“Menthol cigarettes are my saviour with this cold”). We rarely found harm perceptions of menthol as less harmful than other cigarettes (“Oh natural menthol healthy cigarettes”), but the converse message that menthol was harmful was more prominent (“SMOKERS (menthol or nonmenthol) really WILL HAVE stroke. So better stop smoking.” “Menthol cigarettes? So they soothe your throat while giving you cancer? Kinda like being killed by a clown.”). Regarding populations, negative sentiment toward menthol cigarettes was driven by negative views of smoking among children and adolescents, women, and African American populations. Tweets referencing cigars, alcohol, and tobacco control had no significant differences between positive and negative sentiment toward menthol cigarettes.

Among smokers, 65.70% (2653/4038) expressed positive sentiment toward menthol (“Listening to music and smoking a port so relaxed”) (Figure 3). Only 27.34% of smokers (1104/4038) reported negative sentiment toward cigarettes in general (“I cant be doin these menthols man.! Gotta Quit Smoking period. Maybe tomorrow”) or specifically toward menthol-type cigarettes or a menthol cigarette brand (“Had a mayfair menthol, its like smoking mouthwash”). Smoker conversations about menthol cigarettes were dominated by discussion of smoking behavior, addiction or craving, taste preference, and marijuana.

Nonsmokers (320/351, 91.2%) held predominately negative sentiments about menthol cigarettes and smoking in general (“All the girls at my job be smoking them Newports and had the nerve to offer me one. No ma’am, never.”). Former smokers were generally negative about menthol (91/127, 71.7%) (“I remember when I used to be addicted to Newport cigs now the smell of cigarette smoke gives me a headache #happyiquit #cigsarenasty”). However, almost 30% (36/127, 28.4%) of former smokers were positive or ambivalent about menthol (“even though i quit smoking im tempted to try some marlboro menthol lights, i heard they stink less and are less harmful”).

Tweets about menthol cigarettes were most frequently about themes of smoking, taste or sensation, and addiction; African American populations; marijuana as an other substance; or song lyrics or pop culture references. Menthol content on Twitter was also driven by smokers who were more positive than others about menthol cigarettes. A prior study of tobacco-relevant content on Twitter found that most tweets were about personal experiences of use [34]. Liking or disliking the taste or sensation of menthol cigarettes found in a quarter of the tweets and in prior research [18,23] are understandable reasons for smoking menthol rather than nonmenthol cigarettes.

Additionally, Twitter users, particularly smokers, frequently mentioned marijuana and menthol cigarettes, and such tweets tended to be positive. This type of linkage has been found related to use of these products, with adolescent menthol smokers more likely to use marijuana in the past 30 days than nonmenthol smokers [49,50]. Similar to the effect of Twitter content on perceptions of the use of hookah [37] and marijuana [51], these positive views of menthol could normalize menthol cigarette use among the Twitter followers of these users.

Discussion over Twitter focused on smokers’ concerns and tended to dwarf public health concerns, such as smoking cessation, health, and tobacco control. Some conversations about menthol cigarettes were cessation related but, as with other Twitter conversations about cessation, they lacked discussion of evidence-based strategies [52]. As in prior work, a small number of menthol smokers on Twitter may have perceived that switching to a nonmenthol cigarette is a good way to quit [26]. As in prior studies [18,53,54], we found that some smokers saw a medicinal effect of menthol cigarettes when they were sick. Studies have found that African American smokers may perceive health benefits of menthol or view menthol cigarettes as less harmful than nonmenthol cigarettes—beliefs that may interfere with quitting [14,18,20,23,24,26]. However, this view of menthol cigarettes as less harmful was not prevalent on Twitter and may reflect differences in broader population views of menthol. For instance, a national study of US smokers and nonsmokers found that 45.8% of adults believed that menthol and nonmenthol cigarettes were equally harmful; however, a sizable minority did not know whether menthol cigarettes were more or less harmful than nonmenthol cigarettes (40.8%) [21]. Few conversations on Twitter reflected concern for tobacco control policies. There was limited discussion about a menthol cigarette ban in the EU. After the TPSAC report in 2011 [12], but before an advanced notice of proposed rulemaking on menthol in cigarettes in 2013 [55], there was almost no Twitter discussion about a possible menthol cigarette ban in the United States or any other jurisdiction.

Finally, tweets reflected linkage of menthol cigarettes with African American populations historically targeted by menthol marketing and that have higher use prevalence [4,56]. Prior work has identified this “African Americanization” through targeted marketing and financial support for African American organizations [56]. African American menthol smokers have recognized that they have been targeted by menthol cigarette advertising [26]. This linkage between menthol cigarettes and African Americans is also seen on Twitter and is characterized by negative sentiment toward menthol cigarettes. Unfortunately, this negative sentiment seems to be driven by negative stereotypes about African American smokers rather than by rejection of the targeted marketing of menthol cigarettes to African Americans.

The strengths of this study are that it included a full year of global Twitter data from the Firehose, representing the entire corpus, rather than a sampling of Twitter content. Limitations are that, because we drew menthol content only from existing smoking and cigarette datasets, we may have missed menthol cigarette-relevant Twitter content without those terms. Additionally, due to this limitation, our research cannot be generalized to all tweets. We could not use Kool and Salem brand names as keywords because they were not precise enough to characterize menthol cigarette-relevant conversations. Future research in this area should consider using the menthol cigarette keywords beyond these existing archives and also examine potential substitution of menthol cigarettes with menthol flavors of other tobacco products such as e-cigarettes or cigars. The study lacked information about demographic or geographic information of Twitter users to understand differences in perceptions. However, themes found in prior research about menthol cigarettes [18-21,43] were also found in this study, suggesting these perceptions may be shared broadly. Future research can focus more in-depth on discussion of menthol cigarettes related to different populations. Another limitation was the cross-sectional nature of our analysis. Though the analysis was appropriate given our interest in the overall frequency of content about menthol cigarettes in this time period, a longitudinal analysis could provide information about changes in perceptions of menthol cigarette Twitter content.

Our findings present a relatively new way of assessing public opinions of menthol cigarettes through Twitter messages. Most messages are generated by smokers who have more positive sentiment toward these harmful combustible products. Misperceptions of menthol cigarettes having medicinal effects are prevalent, and positive linkage with marijuana is common. Addressing these common misconceptions and denormalizing menthol cigarette use, particularly for African American smokers, could ultimately save lives. In the United States, a menthol cigarette ban would be estimated to avert over 600,000 deaths by 2050, almost one-third among African Americans [57]. Framing public communication about menthol cigarettes to denormalize use in both online and offline forums is especially critical in light of potential regulation by the FDA, the EU, other jurisdictions, and localities. Future analyses could also use Twitter to examine reaction to regulations restricting menthol cigarettes among likely smokers and nonsmokers.