Applying Multiple Data Collection Tools to Quantify Human Papillomavirus Vaccine Communication on Twitter

As media evolves, more information about human papillomavirus (HPV) vaccination is shifting to social media platforms such as Twitter. Health information consumed on social media may be especially influential for segments of society such as younger populations, as well as ethnic and racial minorities, who may be less likely to access health information through formal news sources, health care providers, and other more traditional resources. The objectives of this study, which used interdisciplinary research, were to quantify HPV vaccine communication on Twitter and to develop a methodology to improve the collection and analysis of Twitter data.

The researchers collected Twitter data using 10 keywords related to HPV vaccination from August 1 2014 to July 31 2015. Prospective data collection used the Twitter Search API, and retrospective data collection used Twitter Firehose. Using a codebook to characterise tweet sentiment and content, the researchers coded a subsample of tweets by hand to develop classification models to code the entire sample using machine learning procedures. They also documented the words in the 140-character tweet text most associated with each keyword. They used chi-square tests, analysis of variance, and nonparametric equality of medians to test for significant differences in tweet characteristic by sentiment.

A total of 193,379 English-language tweets were collected, classified, and analysed. Positive sentiment was the largest type of sentiment in the sample, with 75,393 positive tweets (38.99% of the sample), followed by negative sentiment with 48,940 tweets (25.31% of the sample). In addition, positive tweets had, on average, many more followers than negative tweets, averaging 8,022 followers compared to only 4,772 followers, respectively. ("This suggests that more influential users, as measured by the number of followers, are tweeting about the vaccine more positively than negatively.") Positive and neutral tweets constituted the largest percentage of tweets mentioning prevention or protection (20,425/75,393, 27.09% and 6477/25,110, 25.79%, respectively), compared with only 11.5% of negative tweets (5647/48,940; P<.001). Nearly one-half (22,726/48,940, 46.44%) of negative tweets mentioned side effects, compared with only 17.14% (12,921/75,393) of positive tweets and 15.08% of neutral tweets (3787/25,110; P<.001). ("This can be important information for health promotion and communication campaigns, specifically in terms of tailoring a message and joining a particular conversation. As tweets that contain information on side effects are more likely to be part of a negative conversation, tweeting 'side effects are minimal,' that is, downplaying negative sentiment, may not be the most effective way to communicate this information.")

Associated words varied with each keyword, with HPV being associated with personal words such as I, me, and have, and #HPV being associated with January (cervical cancer awareness month), prevent, and learn. Words associated with the specific vaccine-related keywords (e.g., HPV vaccine, Gardasil, and Cervarix) varied greatly, with more positive and preventive words being associated with HPV vaccine (e.g., read, to prevent, for girls), and more negative words being associated with Gardasil and Cervarix (e.g., cdcwhistleblower, exposed).

A majority of tweets included a URL (138,059/193.379, 71.39%), nearly half included a hashtag (86,966/193.379, 44.97%), and just over half included a mention (112,049/193.379, 57.94%). Tweets coded as having positive sentiment toward HPV vaccine and no mention of HPV vaccine had a significantly higher use of URLs (57,800/75,393, 76.66%, and 18,898/25,110, 75.26% containing links, respectively) as compared with negative sentiment (34,491/48,940, 70.48%) (P<.001). Positive sentiment showed the greatest use of hashtags (36,638/75,393, 48.60%); the researchers explain: "The hashtag (#) is an important feature on Twitter that categorizes tweets based on keyword and makes it easier to search other tweets with that same hashtag and keyword. By including #HPV in a tweet, users are able to click on #HPV and read other messages that have also included #HPV, thus acting as a 'social search' function. This search feature may play an important role in raising awareness, as demonstrated by the associated word clusters.") Negative sentiment showed the greatest use of mentions (31,085/48,940, 63.52%). ("Mentions are a way to communicate directly to other users (more directed messaging and communication), and may serve as a way for agenda setters and opinion leaders to emerge in a network...and to control message diffusion....Use of mentions may be a mechanism for negative and alternative messages about the HPV vaccine, which are in the minority, to appear to gain clout and recognition.")

Building on this particular study, future research could "examine how various stakeholders, including parents, youth, health care providers, and health care systems, communicate about the HPV vaccine and identify opportunities to strengthen vaccine uptake and completion. Furthermore, identifying how communication trends are associated with behavioral outcomes, that is, actual vaccine uptake, will be an important next phase of this area of inquiry."

In conclusion: "Understanding the content and implications of conversations that form around HPV vaccination on social media can aid health organizations and health-focused Twitter users in creating a meaningful exchange of ideas and having a significant impact on vaccine uptake.....In terms of public health surveillance, our study demonstrated that, despite an often negative-leaning frame and discussion of HPV vaccine on social media, the greatest percentage of HPV vaccine tweets are positive. Understanding effective dissemination channels will help connect campaigns with 'elite' users and media who have many followers, and consequently may lead to a wider reach of message....This area of research is inherently interdisciplinary, and this study supports this movement by applying public health, health communication, and data science approaches to extend methodologies across fields."