Gardasil on Twitter: A Content Mining Study Examining Message, Context, and Source Characteristics of Human Papilloma Virus (HPV) Vaccine-Related Tweets

Genital human papilloma virus (HPV) is the most common sexually transmitted virus in the US. Effective HPV vaccines remain controversial and have poor uptake among many adolescent groups. Social media have potential for promoting the HPV vaccine, but little is known about the content or sentiment of social media discussions regarding the vaccine. The purpose of this study was to examine message, context, sentiment, and source characteristics of original HPV- and HPV vaccine-related Twitter posts (tweets) over a 3-month time period. This content mining study used publicly available data purchased from Twitter. One-thousand tweets were randomly selected from a sample of 45,260 English language tweets matching our search criteria and collected from December 2014-January 2015. Tweets were manually coded according to source, context, message, and sentiment characteristics using a coding document developed from previous research. Individuals posted just over half (55%) of tweets, whereas organizations posted 40%. More than half (57%) of all coded tweets were positive, and less than one-fifth were negative. Organizations’ tweets were more likely to be positive than individuals’ tweets. tweet accurate, positive HPV vaccination information via Twitter.


Introduction
Genital human papillomavirus (HPV) is the most common sexually transmitted virus in the United States. It is estimated that more than 90% of all sexually active men and 80% of sexually active women will be infected with at least one type of HPV [1]. HPV is associated with cervical, vaginal, and vulvar cancers in women; penile cancer in men; and anal and oropharyngeal cancers in both men and women [2] [3] [4] [5] [6]. HPV can also cause genital warts and warts in the throat [7] [8] [9].
In 2014, the Food and Drug Administration (FDA) approved a vaccine that has the potential to prevent approximately 90% of anal, cervical, vaginal, and vulvar cancers [10]. The recombinant HPV 9-valent vaccine (Gardasil 9, Merck) covers 9 HPV types-5 more than Gardasil and 7 more than Cervarix (2 other FDA-approved HPV vaccines) [10]. Gardasil vaccines are frequently delivered in 3 separate doses over a short period of time. HPV vaccines have the potential to either reduce or exacerbate disparities in HPV-related diseases, depending on vaccine uptake which varies based on racial/ethnic group, income, and language [11].
While Gardasil 9, Gardasil, and Cervarix all have demonstrated positive effects in reducing multiple types of cancer, the vaccines remain controversial.
Reasons include the fact that the vaccines prevent a virus that is transmitted only via sexual contact, differentiating it from most other recommended vaccines [12]. The cervical cancer prevalence and death rate in the US are also both relatively low. Furthermore, some parents believe giving their children this vaccine implies consent to engage in sexual activity, confers a false sense of protection from sexually transmitted infections (STIs), and will lead to earlier first sex [12].
There is also a vocal anti-vaccine movement in which individuals believe that children already have too many vaccinations on the immunization schedule; vaccines in general are "bad"; HPV vaccines are not safe; and/or long-term side effects are unknown [12].
Given these controversies, HPV vaccine uptake remains low, especially in key priority populations. Such health inequities remain a critical adolescent health concern. For example, in 2014, 56% of 13-year-old and 78% of 17-year-old girls reported completing the 3-dose vaccine series [13]. However, fewer males-47% of 13-year-olds and 62% of 17-year-olds-reported series completion [13]. Moreover, important racial disparities exist. Compared to their non-Hispanic White Health peers, non-Hispanic Black females remain less likely to report 3-dose series completion [13]. Although HPV vaccination coverage among females has increased nationally, HPV coverage lags behind tetanus-diphtheria-acellular pertussis (Tdap) vaccine and meningococcal conjugate coverage [13].
Social media represent a potential vaccine promotion modality, as Blacks, Hispanics, and adolescents and young adults who are within the indicated age-range for receiving the HPV vaccine are heavy users of social media [14]. Social media are interactive Web sites and applications that enable users to create, share, comment on, and, modify content [15]. One type of social media, the microblog, consists of the sharing of short pieces of information to the public. One of the best known microblogs is Twitter, started in 2006. Ten years later, Twitter averages 310 million (66 million US) monthly active users [16] [17], who post content in the form of "tweets", or messages of ≤140 characters. Twitter is the social media platform of choice for 18% of teens, beating out Facebook (17%) [18].
Hispanic and Black individuals use Twitter more frequently than their White peers [19]. These young people represent important priority populations for HPV vaccine promotion efforts given both the generally inequitable promotion and uptake within these populations and the critical importance of effectively disseminating evidence-based information to those who are eligible to receive the vaccination.
In order to analyze social media posts, several studies have utilized tweets to examine myriad health issues such as influenza surveillance [20] [21] [22] and tobacco use [23], as well as to assess dissemination of information on antibiotics [24] and to examine attitudes toward HPV and the HPV vaccine [25] [26] [27] [28] [29]. Many of these studies utilized a popular method for analyzing the content of tweets known as sentiment analysis which entails using machine learning methods to analyze tweets in terms of the opinions they express, capturing public opinion about a concept or issue [30]. Investigators have theorized that knowledge gained about HPV and HPV vaccines through sentiment analysis could be harnessed in developing HPV vaccine promotion strategies and messaging through social media. In an effort examine this potential application, tweets about influenza vaccinations were mined to assess the sentiments, or opinions, toward vaccinations as well as how those sentiments could be mapped onto large networks for potential public health outreach [31]. That study found that while sentiments toward influenza vaccination vary, they are often shared within a virtual network of like-minded individuals that would make messages targeting vaccination behaviors difficult to effectively disseminate. Several studies found similar results when examining the content of tweets that discuss the HPV vaccine such that sentiments toward the vaccine are frequently shared within a social community that harbors those same beliefs [26] [32] [33]. These communities could be reached through several distinct sources such as doctors, organizations, celebrities, and fellow community members. This further highlights a research gap that exists between HPV vaccine information on Twitter and adequately disseminating to the audiences that should consume this information.
The aim of this study was to examine message, context, and source characteristics related to original Twitter posts regarding the HPV vaccine over a 3-month time period. Specifically, our aims were to describe the source characteristics of twitter users (people and organizations) who post original messages on Twitter about HPV and the HPV vaccine, examine the types of HPV-and HPV vaccine-related information shared on Twitter, and assess the general sentiment toward the HPV vaccine among social media users who tweet about the vaccine. This study will add to the current literature by reporting sentiments expressed by organizations as well as by individuals.

Data Source
To address our study aims, we purchased social media data from Twitter. Those

Coding
Despite the growing use of sentiment analysis through machine learning methods, there is evidence that machine learning methods do not always align well with social science objectives [34]. For this study, investigators opted to use a qualitative coding methodology whereby we developed a coding document, adapted from sentiment analysis and other content analysis research (e.g. HPV vaccine information on YouTube [35] [36] [37], to manually extract information from this sample of tweets. To pilot test the coding document, we collected data via the Twitter streaming API between September 11 and November 20, 2014. Using Health fied Twitter badge), and/or organization. Tweets posted by individuals were coded using information from the user's profile page indicating whether the individual self-identified as a parent, child, and/or spiritual/religious person; if they had a particular political affiliation; and their reported profession (e.g. journalist, physician). If the tweet was posted by an organization, organization type was coded using information from the user's profile (e.g. business, media outlet, nonprofit, government). Since we pulled data from the poster's profile, these data are not considered anonymous. For example, some "personal" information was recorded, such as poster name (e.g. Kinsey Institute) and affiliation.
All tweets were coded for sentiment using the following categories: "neutral", "negative", "ambiguous", "positive", and "other". "Neutral" tweets were defined as those using language indicating neither approval nor disapproval of the vaccine. "Negative" tweets used language disapproving of the HPV vaccine, whereas "positive" tweets used language approving of the vaccine. "Ambiguous" tweets contained both approving and disapproving language. Tweets coded as "other" did not fit into any of the other sentiment categories. Message tone was also coded as to whether tweets contained concerns about civil liberties, that the vaccine is a hoax, and that the vaccine is dangerous.
Guided by previous research [35], we coded tweets for specific information about the HPV vaccine. The following yes/no categories were coded for message frames, or mention of the vaccine being a: cancer prevention, STI prevention, or genital wart prevention tool. We also coded for other vaccine-related information (e.g. whether the tweet mentioned vaccine eligibility criteria, safety concerns, side effects) and the source that the tweet credited (e.g. Centers for Disease Control and Prevention [CDC], medical doctor). We coded for whether tweets gave a firsthand account of an experience with HPV or the vaccine, person writing about the personal experience was a child or adult, and personal experience came from a parent/relative of someone who had received or would receive the HPV vaccine. Finally, we coded messages for other HPV-related information (e.g. how HPV is transmitted, how to protect against infection, symptoms).

Analysis
We conducted all analyses using IBM ® SPSS ® Statistics (Release 22.0.0.0). Means and standard deviations were calculated for continuous variables. Frequencies and percentages were calculated for all categorical variables. We also conducted a descriptive temporal analysis of tweets, organized by date of original message posting and by tweet sentiment/tone. Lastly, a chi-square analysis was conducted to determine statistical differences between source types (i.e. individual versus organization) on tweet sentiment.

Source Characteristics
Of the 1000 manually coded tweets, just over half (n = 548, 54.8%) were posted Health by individuals, whereas organizations posted 404 tweets (40.4%). The remaining tweets came from profiles that were unclear in terms of representing an individual or organization. Among the 548 tweets posted by individuals, the largest identified types were parents (9.5%), journalists (8.2%), and physicians (7.3%; Table 1). Only one tweet came from a dental professional, and no tweets from children/teens were identified. Among the 404 tweets from organizations, the largest identified types were health information providers (43.6%), businesses (37.4%), non-profit/advocacy groups (16.6%), and healthcare organizations (13.4%).

Tweet Sentiment
More than half of tweets (n = 571, 57.1%) were coded as having a positive tone.
Less than one-fifth of tweets were coded as having a negative tone (n = 188, 18.8%), whereas 195 (19.5%) were coded as neutral. Forty-six tweets were coded as either ambiguous (1.4%) or other tone (3.2%).   • Also Breaking: seat belts don't promote car accidents! "No increase in risky sexual activity w/HPV vaccine" • @ThrupennyBit And now we know cervical cancer can be prevented by the HPV vaccine. Who wouldn't want THAT for their kids?

Message Characteristics
The largest message frame related to the HPV vaccine as a cancer prevention tool, as 18% of tweets mentioned that the vaccine can prevent cancer (Table 3).
Very few tweets in this sample included HPV vaccine-related information (e.g. girls are vaccine-eligible; Table 3). Attributing credit to sources of information in tweets was very uncommon in this sample, with the most commonly cited sources in tweets being medical doctors (3.0%; Table 3). Few tweets were delivered from a first person perspective. Of tweets posted by individuals, 23 (4.2%) included a personal account from individuals who discussed a firsthand experience with HPV or the vaccine. Only 0.6% originated from the perspective of a parent/relative of a potential or actual HPV vaccine recipient. Finally, few tweets included information about HPV, such as its link with cancers and how HPV can be transmitted (Table 3).

Discussion
The discussion surrounding the HPV vaccine in original Twitter posts appears to be positive, as more than half of tweets examined in our research contained language approving of the vaccine. These findings differ from some of the existing research on social media and the vaccine, including across 2 content analyses of YouTube videos [37] [38], a qualitative study [39] and content analysis [40] of online news stories, and a content analysis of top search engine results [36]. In more recent research, Bahk et al. [41] found that mainstream media were dominated by positive HPV vaccine sentiment (over a one-year period, the weekly average was 75.5% positive), but on Twitter the predominant sentiment toward the HPV vaccine was negative (over a one-year period, the weekly average was 74.9% negative). The Bahk et al. study included retweets in analyses, however, and it is possible that negative material is retweeted more frequently. In addition, our study indicates that the quantity, sentiment, and tone of tweets varies over time and is linked to "newsworthy" events, such as FDA approval for a new vaccination, research results being disseminated, and political debates. Although HPV vaccine sentiment seems to vary across time, studies, and platforms, research indicates that negative sentiments seem to be more influential, at least regarding HPV vaccine sentiment. In a controlled experiment, for example, Nan and Madden [42] found that college students exposed to a negative blog post about the vaccine "perceived the vaccine as less safe, held more negative attitudes toward the vaccine, and had reduced intentions to receive the vaccine" compared to those who had no blog exposure. However, those exposed to the positive blog had no changes in perceptions, intentions, or attitudes regarding the vaccine [42]. Twitter-specific data on HPV vaccine sentiment suggest that we could stratify tweets based on social network connectedness to target large swaths of consumers that share specific sentiments regarding the vaccine [28] [43]. Targeted tweets or posts could then be disseminated to those community members to improve their sentiment regarding HPV vaccination. While this is one potential approach to effectively disseminating messages regarding HPV vaccination, there is still a critical need to optimize methods to counteract tweets providing negative and incorrect information, especially since HPV-related tweets often occur in quick spikes.
On Twitter, the HPV vaccine is framed largely as a cervical cancer prevention tool, with less focus on its potential to prevent genital warts or other cancers. In addition, there were more tweets describing how the HPV vaccine could help females, potentially making it seem less relevant for males. According to Bigman et al. [44], participants who learned about HPV vaccine effectiveness believed more in its effectiveness for preventing cervical cancer and felt more positively toward the vaccine for cervical cancer prevention. Adolescent medicine specialists and public health educators should expand communication regarding the HPV vaccine's ability to prevent other types of cancers and warts, which can occur in males and females.
Very few tweets in this sample included HPV-and HPV vaccine-related information. Calloway et al. [45] and Habel et al. [40] found that US newspaper coverage and online news stories of the HPV vaccine lacked detailed information on HPV, which could contribute to people misunderstanding the complexity of cervical cancer and HPV. While the 140 character limit for tweets constricts the amount of information that can be shared, professionals can share detailed information via Twitter by including links with information about HPV and the vaccine. Individuals (most commonly parents, journalists, and/or physicians) shared most (54.8%) tweets, but few tweets were written in first-person. Organizations (mostly health information providers, non-profit/advocacy groups, and healthcare organizations) shared about two-fifths of tweets. Additionally, the majority of negative sentiments were tweeted by individuals instead of by organizations.
Thus, organizations may be a potential source of information that would reach more followers outside of social networks that share homogenous beliefs about the vaccine. Other types of healthcare personnel-nurses, allied health professionals, health educators, and dental professionals-might consider communicating correct and positive information about the HPV vaccine via Twitter.
These practitioners could also be targeted as recipients of positive messages in the hopes that they would then disseminate the information to patients.

Study Limitations
This research should be considered within the context of its limitations. While it is clear that some healthcare providers-mainly physicians-disseminate HPV vaccine information via Twitter, our study reveals an opportunity for a greater variety of healthcare personnel to provide accurate and positive HPV vaccination information on this platform. In addition, healthcare providers should be aware of messages that parents, adolescents, and young adults are receiving about the vaccine, especially in times of substantial coverage of an HPV vaccine news topic. Future directions could also target these various healthcare practitioners to then disseminate positive messages regarding HPV vaccinations to appropriate recipients. Finally, it is critical that public health professionals learn more about how to harness the power of social media like Twitter to better inform the decision making of parents, adolescents, and young adults related to the HPV vaccine.

Declarations
Funding: This work was supported in part by the Center for Human Dynamics in the Mobile Age (for use of its SMART Dashboard), and seed money provided by San Diego State University.
Ethical approval: None required. Guarantor: EWB. Contributorship: EWB, JRH, and JLF researched literature and/or conceived the study. EWB, AN, JLF, and KJW was involved in protocol development and data analysis. EWB and JRH wrote the first draft of the manuscript. All authors reviewed and edited the manuscript and approved the final version of the manuscript.