Tweet for Behavior Change: Using Social Media for the Dissemination of Public Health Messages

Background

• Social media offers targeted, interactive health-promotion opportunities, including reaching hard-to-reach groups and creating dynamic campaigns, but research remains limited on campaign design, success measures, and whether likes or shares imply behavior change.Source: Background · p. 2

Aim

• The study aimed to test the feasibility of designing, implementing, and evaluating a bespoke social media intervention to disseminate skin cancer prevention messages.Source: Aim · p. 3

Research Objectives

• Objectives covered campaign feasibility and effects on attitudes and knowledge, message frames, diffusion strategies, process measures and data access, and suitable control groups.Diffusion strategies included promoted messages, influencers, and a Thunderclap; demographic data of interest included gender.Source: Research Objectives · p. 3

Why a Skin Cancer Campaign?

• A 750-person postcode-stratified survey found 41.1% did not know melanoma was most serious, 49.2% considered tans healthy, and 6.4% checked skin frequently.The regional Public Health Agency identified skin cancer as a campaign priority based on gaps in sun-safe attitudes and knowledge; the study assessed baseline and postcampaign levels.Source: Why a Skin Cancer Campaign? · p. 3

Why Twitter?

• Twitter was selected because posts are public and its APIs enable real-time access; streaming data were filtered for campaign hashtags and cleaned of blank, spam, and irrelevant tweets.Source: Why Twitter? · p. 3

Design

• A quasi-experimental interrupted time-series comparison used a crossover: charity-account Phase 1 (May 1–July 14, 2015), two-week washout, then bespoke-account Phase 2 (August 1–September 30).Phase 1 was two weeks longer to account for reduced interactions during a national holiday period; the protocol followed the CONSORT-EHEALTH checklist.Source: Design · p. 3

Control Group

• Northern Ireland was the campaign-exposed area and Wales the control; geo-located tweets and predefined keyword volumes were compared before and after the campaign.Keyword volumes were used to track campaign-related messages in each location.Source: Control Group · p. 3

Intervention Development and Implementation

• Phase 1 used informative, story, shock, humor, and opportunistic or responsive messages about skin surveillance and sun protection; Phase 2 used similar content.Phase 2 also included a Thunderclap on September 1, 2015, and both phases used £10 promoted posts targeted to Northern Ireland residents aged 18 or older.Source: Intervention Development and… · p. 4, Measures · p. 4

Measures

• The team tracked impressions, engagements (including likes, comments, shares, and retweets), and engagement rate, while aiming to infer user gender from Twitter names despite limited profile data.Impressions were post views, and engagement rate was engagements divided by impressions.Source: Measures · p. 4

Pre- and Postintervention Web-Based Survey

• The Web-based survey followed CHERRIES reporting guidance and used 37 multiple-choice questions covering demographics, skin-cancer prevention, and psychosocial mediators.It was modeled on the Cancer Research UK SunSmart survey and took approximately 15–20 minutes to complete.Source: Pre- and Postintervention Web-Based… · pp. 4–5
• Social-media advertisements invited Northern Ireland adults over 18 in April and October 2015; paid promotion, an iPad Mini prize draw, and age-and-region stratification supported recruitment.Promotion was capped at £15 on Facebook and £10 on Twitter; a £15 Facebook limit could potentially reach 770–2000 Northern Ireland residents aged 18–65+.Source: Pre- and Postintervention Web-Based… · p. 5
• The primary outcomes were pre/post changes in sun-protection attitudes and skin-cancer knowledge; surveys were included if respondents completed the final question, without IP-based duplicate checks.Differences between the preintervention and postintervention periods served as the intervention-impact assessment.Source: Pre- and Postintervention Web-Based… · p. 5

Data Analysis

• Focus-group and workshop transcripts were anonymized and analyzed through repeated reading, initial coding, and an a priori thematic Framework method.The themes concerned the perspectives of professionals and users.Source: Data Analysis · p. 5
• Survey data on social-media use, demographics, and skin-cancer knowledge and attitudes were summarized with frequencies and cross-tabulations; significance tests were omitted for feasibility-study reasons.Cross-tabulations compared responses by gender, age, and other sociodemographic characteristics.Source: Data Analysis · p. 5
• Analytics were gathered for individual posts, message frames, and the overall campaign, with frequencies used to compare impressions, engagement, and shares across frames.Source: Data Analysis · p. 5

Measuring Twitter Analytics

• Adoreboard-provided Twitter Firehose data and metadata supported counts of tweet impressions and engagements, hashtag use, retweets, and pre/postcampaign search-term frequencies.The service provider enabled greater access to Twitter data than public APIs.Source: Measuring Twitter Analytics · p. 5

Demographic Characteristics

• The pre- and postcampaign surveys had 337 and 429 participants; respondents were disproportionately female and more educated than Northern Ireland’s population.Both samples had similar age distributions, but fewer postcampaign households reported income above £20,001 (40.4% vs 52.8%).Source: Demographic Characteristics · p. 5

Attitudes to UV Exposure and Skin Cancer Prevention

• Postcampaign, fewer respondents liked tanning (60.5% to 55.6%), viewed tan as healthy (55.9% to 52.7%) or attractive (48.6% to 43.7%), while sun-protection belief rose (62.6% to 65.0%).These were trends in attitudes, not evidence of behavioral change.Source: Attitudes to UV Exposure and Skin… · p. 6

Knowledge of Skin Cancer Prevention

• Postintervention, more respondents correctly identified skin cancer as the most common cancer (28.4% to 39.3%) and melanoma as most serious (49.1% to 55.5%).Awareness also rose for midday sun strength (91.3% to 93.5%) and fair-skin protection needs (73.8% to 77.6%).Source: Knowledge of Skin Cancer Prevention · p. 7

Influence of Message Frames on Social Media

• Shocking tweets had the most impressions (n=2369), humorous tweets the most engagement (n=148), and informative tweets the most retweets (17 followers).The most retweeted story, shock, and humor messages were shared 7, 9, and 10 times, respectively.Source: Influence of Message Frames on Social… · p. 7
• By median, shocking messages led in impressions (565), engagements (15.5), and retweets (2.5), while humorous messages led in engagement rate (2.5%).Source: Influence of Message Frames on Social… · p. 7

Influencers

• Influencer tweets generated more impressions, especially on shock posts (n=11,349) and story posts (n=9612); promoted posts did not notably increase impressions, engagement, or retweets.Source: Influencers · p. 7

Opportunistic Messages

• For opportunistic messages, the maxima were 2993 impressions, 103 engagements, and 8 retweets on a single tweet.Source: Opportunistic Messages · p. 8

Thunderclap

• Thunderclap reached its 100-supporter activation goal with 122 supporters and a social reach of 454,207.Source: Thunderclap · p. 8
• Thunderclap-related tweets with an influencer drew 11,740 impressions versus 2527 without one; top support tweets achieved 2527 impressions, 2 engagements, and 7 retweets.Source: Thunderclap · p. 8

Twitter Analytics

• The campaign generated 417,678 tweet impressions, 11,213 post engagements, and 1211 retweets; one tweet led with 11,349 impressions, 811 engagements, and 17 retweets.Source: Twitter Analytics · p. 8

Is There an Appropriate Control Group for a Social Media Campaign?

• Geo-tagged keyword searches returned 15,964 Northern Ireland tweets in April 2015 and 14,168 in October, versus 50,164 Wales tweets in April and 51,634 in October.Source: Is There an Appropriate Control Group… · p. 9
• After the campaign, keyword counts increased in Wales but decreased in Northern Ireland, and the designated campaign hashtags did not appear in Wales search results.Source: Is There an Appropriate Control Group… · pp. 9–10

Principal Findings

• The study found Twitter feasible for public health messaging, enabling dynamic, tailored messages to reach audiences in real time.Source: Principal Findings · p. 10, Investigating the Impact of a Bespoke… · p. 10

Investigating the Impact of a Bespoke Social Media Campaign on Skin Cancer Attitudes and Knowledge

• Web-based surveys showed a trend toward improved skin cancer knowledge and attitudes, but cannot accurately represent overall population knowledge.Reported gains included awareness that sun protection can reduce skin cancer risk and greater awareness of skin cancer severity.Source: Investigating the Impact of a Bespoke… · p. 10

Investigating the Impact of Employing Different Message Frames on Social Media

• Shocking messages generated the most impressions, humorous messages the greatest engagement, and informative messages were retweeted most.Source: Investigating the Impact of Employing… · p. 10

Promoted Messages

• Promoted tweets produced fewer impressions and retweets than organic or influencer posts, though targeting may benefit campaigns with narrower audiences.The study targeted adults aged 18 and older across Northern Ireland.Source: Promoted Messages · p. 11

Influencers

• Influencer seeds increased impressions and engagements, but selection should consider message and audience fit, not just follower count.Seed users differed in influence, and predictive models may help assess message propagation and campaign impact.Source: Influencers · p. 11

Thunderclap

• The Thunderclap campaign reached 122 supporters and more than 450,000 people, suggesting it can spread awareness when adequately promoted and explained.Source: Thunderclap · pp. 11–12

Determining the Appropriate Process Evaluation Measures and Access to Data for a Social Media Campaign

• Impressions, likes, and retweets do not establish message support or predict behavior, so social media campaigns need adapted evaluation methods.The study delivered a Web-based survey through social media; netnography, crowdsourcing, and photo or video elicitation were suggested for deeper assessment.Source: Determining the Appropriate Process… · p. 12

Investigating Whether There Is an Appropriate “Control Group” for a Social Media Campaign

• A suitable control group was unavailable, while weather, clustered tweets, and Twitter’s unrepresentative user base limited inference.The 2015 summer was unusually cold and wet, and #eek impressions may have been driven by an influencer; the study also noted unresolved privacy and identification concerns. Traditional randomized trials and process evaluation guidance offer little direction for social media interventions.Source: Investigating Whether There Is an… · pp. 12–13

Limitations

• Contamination across study phases was a key methodological concern; future research should use a phase-based pre-post design with an adequate wash-out period.Source: Limitations · p. 13
• The Twitter population may be unrepresentative, and limited campaign traffic and interaction mean findings should be interpreted with caution.Source: Limitations · p. 13
• Influencers’ follower numbers and campaign engagement were limited, and the study generated hypotheses requiring testing in a larger, definitive trial.Source: Limitations · p. 13

• A 24 KB PDF file, publichealth_v3i1e14_app1.pdf, is listed as the multimedia appendix.Source: Multimedia Appendix 1 · p. 13

• The appendix is labeled “Intervention development.”Source: Multimedia Appendix 2 · p. 13
• The appendix is provided as a 194KB PDF file named publichealth_v3i1e14_app2.pdf.Source: Multimedia Appendix 2 · p. 13

• The appendix is provided as a 56KB PDF file.Source: Multimedia Appendix 3 · p. 13

Multimedia Appendix 4

• The appendix presents a survey advertisement on Twitter.It is provided as a 160KB PDF file.Source: Multimedia Appendix 4 · p. 14

• UV means ultraviolet, and CHERRIES means Checklist for Reporting Results of Internet E-Surveys.Source: Abbreviations · p. 17