Twitter and Health Moses Flash, Williams College ’15 SUMR, LDI Program Summer 2014 Raina Merchant, MD MSHP Outline

• Introduction • Research Question • Methodology • Results/Discussion • Questions Introduction: What is Twitter?

• Created and made public in 2006

• 140 character microblogs

• 271 million active users – 500 million tweets per day – 35+ languages – 77% outside US

• “To give everyone the power to create and share ideas and information instantly, without barriers”.

• Source: https://about.twitter.com/company Research Question

• Prior work has identified that Twitter is being used for detecting outbreaks earlier than the CDC, tracking policy implementation, sentiment regarding the ACA, identifying variability in language use by region, and monitoring human behavior.

• We sought to identify peer-reviewed publications using Twitter as a database to answer health and health care related questions.

• Evaluating use of Twitter as a research tool can help health scientists better understand the opportunities and limitations of this social data Methodology

• Databases: PubMed and Web of Science

• Keywords “Twitter and health”

• Exclusion: publications/studies that weren’t twitter/health related, or were either an editorial, perspective review, case study.

• Inclusion: studies that used twitter as a tool to analyze or obtain data on a health-related topic. Methodology

• Data extracted from each article: – Author, year, title, journal, research field, methodology, use of Twitter, tweets analyzed, objective, outcomes, time frame of study and conclusions

• Created themes categorizing the use of twitter in each study

• All studies were reviewed independently by 2 coders Results

496 Articles

Web of Science: Embase: 62 178 PubMed: 256

98 met inclusion/ exclusion criteria Use of Twitter for Health Research

Bucket Number of articles (percentage) Health Information Dissemination 23 (24%) Psychosocial/Behavioral 22 (22%) Health Discourse 19 (19%) Surveillance 17 (17%) Sentiment 6 (6%) Engagement 3 (3%) Marketing 3 (3%) Study Recruitment Tool 2 (2%) Regional Health Surveying 2 (2%) Prediction 1 (1%) Adv Chronic Kidney Dis BMJ quality & safety International Journal of Journal of the Medical Transl Behav Med JournalsMedical that Informatics Publish Informatics this Association Research Alcohol Body Image J Adv Nurs Methods Inf Med Urology Am J Infect Control Br J Sports Med J Am Coll Radiol Online J Issues Nurs BMJ quality & safety Am J Trop Med Hyg Computers in Human Ophthalmic Physiol Journal of the American Dental Behavior J Biomed Semantics Opt Association American Journal of Crisis J Cancer Educ PLoS One Cardiology Tobacco Control Angle Orthod Cyberpsychol Behav J Dent Res Prev Chronic Dis Soc Netw BMJ Open Annali dell'Istituto Disaster Med Public J Health Commun Regen Med Superiore di Sanita Health Prep Int J Eat Disord J Med Internet Res Emerg Med Annals of Thoracic Resuscitation Journal of Medical Internet Medicine Research Artif Life Epilepsy Behav J Med Toxicol. Science International Journal of Human- Computer Studies Artificial Intelligence in Female Pelvic Med Journal of Medicine Reconstr Surg. J Oncol Pract Communication Health Policy BMC Cancer Fertil Steril J Phys Act Health Stud Health Technol Inform BMC Public Health Health Commun JAMA Derm Telemed J E Health BMC Res Notes The Journal of Theoretical biology and Social Psychological Antimicrobial medical modelling and Personality Chemotherapy Science Health Information Dissemination: What type of information is available on Twitter?

• 24% of articles

• Omurtag K et. al. 2012 “The ART of social networking: how [Society for Assisted Reproductive Technology SART member clinics are connecting with patients online”

• Evaluated 384 registered clinics and 1,382 posts – 30% of clinics had linked social networking page. – 89% affiliated with nonacademic centers – 31% of posts provided information, 28% included advertising 19% offered support, 17% were irrelevant to the target audience and 5%involved patients requesting information.

Psychosocial/Behavioral: Analyzing social and behavioral traits, response to stimulus. • 22% of articles

• Ritter RS, Preston JL, Hernandez I 2014 ”Happy Tweets: Christians Are Happier, More Socially Connected, and Less Analytical Than Atheists on Twitter”

• Sampled followers from 5 prominent Christian and 5 prominent atheist figures.

• Operationalized happiness as relative frequency of words in LIWC positive emotion dictionary to the frequency of words in the negative emotion dictionary

• Christians expressed more happiness in their tweets and talked more about social processes Table 1. Descriptive Statistics and Zero-Order Correlations.

Ritter R S et al. Social Psychological and Personality Science 2013;5:243-249

Copyright © by Social and Personality Psychology Consortium Health Discourse: What is being discussed about a health topic, by who, and why? • 19% of articles

• Love B, Himelboim I, Holton A, Stewart K 2013 “Twitter as a source of vaccination information: content drivers and what they are saying”.

• Collected tweets with keywords related to vaccination – Most reposted, favorited, shared or response garnering tweets were considered “influential” – 2,580 tweets

• One-third of tweets were supportive of immunization. Health-focused sites, professional media, and medical organizations dominated shared links. Number (%) of Frequency by tone∗ Source sources (n = 341) Positive (%) Negative (%) Neutral (%) Local media 22 (6.4) 9 (40.9) 4 (18.2) 9 (40.9) National media 43 (12.6) 17 (39.5) 6 (19.0) 20 (40.5) Digital news 35 (10.3) 11 (31.4) 4 (11.4) 20 (57.1) aggregator Health specific 56 (16.4) 14 (25.0) 7 (12.5) 35 (62.5) Grassroots 28 (8.2) 6 (21.4) 7 (25.0) 15 (53.6) Alternative 18 (5.3) 1 (5.6) 6 (33.3) 11 (61.1) therapy Government 29 (8.5) 14 (48.3) 5 (12.2) 10 (34.5) agency Medical 41 (12.0) 10 (24.4) 6 (14.6) 25 (61.0) organizations Political 5 (1.5) 1 (20.0) 4 (80.0) 0 (0.0) organizations Advocacy 20 (5.9) 4 (20.0) 5 (25.0) 11 (55.0) organizations Other vaccine 20 (5.9) 4 (20.0) 1 (3.0) 15 (75.0) Other 24 (7.0) 5 (20.8) 1 (4.2) 18 (75.0) Total 341 (100.0) Surveillance: Monitoring outbreaks, mental disorders and other health issues. • 17% of articles

• Yom-Tov E 2014 “Detecting disease outbreaks in mass gatherings using internet data”

• Used a keyword filter for words associated with disease symptoms at nine festivals in the UK – Compared frequency of word use before and after festivals, as well as frequency of word use between other users after the festival.

• Disease symptom significant in two of the nine festivals, which agree with anecdotal data. Statistically significant symptoms from Twitter data for each event and three analysis methods.

Event Method 1 Method 2 Method 3 Cough Cough Tired, cough None Tired, pain, tremor Tired, flatulence

T in the Park Tired Tired, pain, cough Tired, cough Tired, pain, Depression Depression depression None Tired, pain, tremor Tired, fever Tired, pain, None None blindness Hajj Rash, wound Tired Tired Festival None Bleeding None

Download Festival None None None

RockNess None Phobia, swelling None Summarization of Findings and Limitations

• Twitter has mainly been utilized to analyze: – Dissemination of health information – Public discourse about health topics – Social Behavior – Disease surveillance

• Interesting uses of twitter – Analyze social behavior – Study recruitment

• Limitations may included the phrase used in the keyword filter and the number of databases explored. So what?

• Twitter is a great, innovative way to understand health – Unknown opportunities and limitations

• Thank you Dr. Merchant, Kevin and all the members of the Social Media Lab for the great summer research experience!!

• Questions? Sentiment: Public opinion on health issues.

• 6% of articles

• Emery SL et. al, 2014 “Are You Scared Yet? Evaluating Fear Appeal Messages in Tweets About the Tips Campaign”

• Determine how twitter users felt about the graphic imagery presented online by the CDC during an anti-smoking campaign

• Viewers of the imagery neither rejected nor dismissed the graphic and emotionally evocative material. Vast majority of users reflected message acceptance Engagement: Why are patients/ providers using twitter?

• 3% of articles

• De la Torre-Díez et. al. 2012 “A content analysis of chronic diseases social groups on Facebook and Twitter.”

• Looked at tweets related to colorectal cancer, breast cancer and diabetes

• Classified tweets into 5 main groups: – Fund Collecting groups, awareness groups, support groups, prevention groups and disease-fighting groups Marketing: How is Twitter being used to market products that effect health? • 3% of articles

• Huang J et. al, 2014 “A cross-sectional examination of marketing of electronic cigarettes on Twitter”.

• 73,672 tweets were captured between May and June 2012 – Coded for smoking cessation and safety mentions, and were classified as commercial or non-commercial – 90% classified as commercial

• Commercial tweets driven by a small group of highly active accounts – 94% included links to websites that often sell or promote e-cigs

• 34% of commercial tweets included mentions of prices or discounts Study Recruitment Tool: Using twitter as a means to recruit study participants. • 2% of articles

• Yuan P et. al, 2014 “Using Online Social Media for Recruitment of Human Immunodeficiency Virus-Positive Participants: A Cross- Sectional Survey”

• An online survey containing 112 items on demographics, HIV clinical outcomes, and use of technology was disseminated through Twitter and other networks. – 2034 survey clicks, 1404 met inclusion criteria and initiated survey and 1221 individuals completed the survey (87% retention). Regional Health Surveying: Tracking Health variance between locations. • 2% of articles

• Chen X, Yang X 2014 “Does food environment influence food choices? A geographical analysis through “tweets””

• Captured tweets with verified content that include information about a current or upcoming food choice and food outlets, including quality grocery stores and fast food restaurants in the region. – 81,543 tweets. • Regions with more healthful tweets and choices had more quality grocery stores. The number of fast food chains was not correlated with the presence of unhealthy tweets. Prediction: Using Twitter to predict health outcomes. • 1% of articles

• Al-Mohrej OA, et, al, 2014 “Will any future increase in cigarette price reduce smoking in Saudi Arabia”

• Developed questionnaire with information on demographic, socioeconomic factors, smoking history and personal opinion on price effects on cigarette consumption – Posted on twitter – 2057 responses and 802 current smokers – 92% of smokers were male

• 55% of smokers indicated a price increase from $US 2.67 to 8.27 would lead to smoking cessation.