Mining Human Papillomavirus HPV Vaccination Health Beliefs from
Mining Human Papillomavirus (HPV) Vaccination Health Beliefs from Twitter Using Attentive Recurrent Neural Network Data Analytics and Social Media 2. 0 S 53 Jingcheng Du UTHealth School of Biomedical Informatics Twitter: @Jingcheng. Du #AMIA 2018 Supervisor: Cui Tao, Ph. D.
Disclosure I and my spouse/partner have no relevant relationships with commercial interests to disclose. AMIA 2018 | amia. org 2
Learning Objectives After participating in this session the learner should be better able to: • Understand a systematic way to mine health beliefs on HPV vaccination from Twitter • Be familiar with the use of deep learning in understanding the contents of Twitter AMIA 2018 | amia. org 3
Vaccines Save Lives, However ? • 74% of the pediatricians report encountering a parent who refused or delayed vaccines • For the parents that adhere to the schedule, 25% of them feel it may not be the best or safest way to immunize youngsters • The increasing rates of nonmedical exemptions on vaccinations • 1. 2. 3. Major measure for vaccine refusal and delay American Academy of Pediatrics. "Documenting parental refusal to have their children vaccinated. " (2014). https: //www. livescience. com/16346 -parents-distrust-delay-vaccines. html Dredze M, Broniatowski DA, Smith MC, et al. Understanding Vaccine Refusal. Am J Prev Med 2016; 50: 550– 2. AMIA 2018 | amia. org 4
HPV • HPV infection • Nearly 80 million Americans are currently infected with some type of HPV • Most HPV infections (9 out of 10) go away by themselves within two years • Sometimes, HPV infections will last longer, and can cause certain cancers, including cervical cancer, anal cancer, etc. • Every year in the United States, HPV causes 33, 700 cancers in men and women • HPV vaccines 1. 2. 3. • Cancer prevention: prevent most of the cancers caused by HPV infection (about 31, 200) • HPV vaccination completion rates remain low (significantly lower than other recommended adolescent vaccinations) • The good understanding of health beliefs from the public is a crucial step for developing effective vaccine promotion strategies https: //www. cdc. gov/hpv/parents/questions-answers. html. https: //www. cancer. gov/about-cancer/causes-prevention/risk/infectious-agents/hpv-fact-sheet https: //www. mdanderson. org/content/dam/mdanderson/documents/prevention-and-screening/HPV-Statement. pdf AMIA 2018 | amia. org 5
Motivation • Limitations in traditional survey based methods • Labor consuming • Time consuming • Expensive to reach large-scale populations • Difficulties in tracking the changes • Social media 1. 2. 3. 4. • Creation and exchange of user generated contents in real time • Enormous popularity, with lots of health related discussion Kaplan AM, Haenlein M. Users of the world, unite! The challenges and opportunities of Social Media. Bus Horiz 2010; 53: 59– 68. http: //www. pewinternet. org/fact-sheet/social-media/ Dredze M, Broniatowski DA, Smith MC, et al. Understanding Vaccine Refusal. Am J Prev Med 2016; 50: 550– 2. Dube E, Bettinger JA, Halperin B, et al. Determinants of parents’ decision to vaccinate their children against rotavirus: results of a longitudinal study. Health Educ Res 2012; 27: 1069– 80. AMIA 2018 | amia. org 6
Methods Overview - Pipeline Figure 1: the overview of deep learning based framework AMIA 2017 | amia. org 7
Methods • Twitter data, collected by keywords (Streaming API) • Unlabeled, 956, 262 tweets from 2014 -01 -01 to 2017 -12 -31 • Labeled, 6000 tweets, three annotators • Health Belief Model (HBM) • The most widely used conceptual frameworks to explain why people adopt behaviors that lead to healthy lives • HBM constructs can influence the likelihood of HPV vaccination intention and uptake • Text classification • Related to HBM (contains one or more health beliefs) • Susceptibility, Severity, Benefits, Barriers 1. Donadiki EM, Jiménez-García R, Hernández-Barrera V, Sourtzi P, Carrasco-Garrido P, de Andrés AL, Jimenez-Trujillo I, Velonakis EG. Health Belief Model applied to non-compliance with HPV vaccine among female university students. Public Health. 2014 Mar 1; 128(3): 268 -73. . AMIA 2018 | amia. org 8
Perceived Susceptibility Explanation • The assessment of the risk of getting HPV infections Sample tweet • hpv is so common almost everyone will be infected with the virus. but it can cause cancer. so why wait? vaccinate! Annotation agreement Prevalence in labeled tweets • 0. 727 AMIA 2018 | amia. org 9
Perceived Severity Explanation • The assessment of whether HPV infections are sufficient health concerns Sample tweet • learn about the human papillomavirus (hpv), which causes almost all cases of cervical cancer: … Prevalence in labeled tweets Annotation agreement • 0. 807 AMIA 2018 | amia. org 10
Perceived Benefits Explanation • Benefits of HPV vaccine in protecting against HPV infections and HPV infections induced cancers, etc. Sample tweet • health lifestyle | here's how the hpv vaccine can help cut the risk of cancer in gay men | news > Prevalence in labeled tweets Annotation agreement • 0. 831 AMIA 2018 | amia. org 11
Perceived Barriers Explanation • Side effects of HPV vaccines; cost of getting HPV vaccines; negative news & reports on HPV vaccines, etc. Sample tweet • hpv vaccine is associated with serious health risks Annotation agreement Prevalence in labeled tweets • 0. 834 AMIA 2018 | amia. org 12
Methods • Deep learning model • Bi-directional attentive Recurrent Neural Network (RNN) model • Baseline models • Traditional machine learning models • Extra trees, logistics regression, support vector machine • Feature: average embedding • Neural network baseline models • Bi-directional RNN model, Convolutional Neural Network model (Kim) • Evaluation • Tweets were split into training, validation and testing (7: 1: 2) • Precision, recall, f score AMIA 2018 | amia. org 13
RNN with Attention Binary label Augment sequence models to capture the salient information Capture information in both directions Glo. Ve Twitter Embedding Figure 2: the architecture of RNN with attention framework AMIA 2018 | amia. org 14
Comparison with ML Baseline Figure 2: comparison with traditional ML models, measured in F score AMIA 2018 | amia. org 15
Comparison with DL Baseline Figure 3: comparison with other DL models, measured in F score AMIA 2018 | amia. org 16
Prediction on Un-labeled Tweets Susceptibility Severity Benefits Barriers Figure 4 -a: Prevalence of HBM constructs in 2014 -2017 Figure 4 -b: longitudinal analysis of health belief of HPV vaccination: the green shallow represents the total number of HBM related tweets for each month; the colored lines represent prevalence of each HBM constructs (ratio of construct related tweets to total number of tweets, within each month) AMIA 2018 | amia. org 17
Why the shift? Top Google News in 2014 AMIA 2018 | amia. org 18
Why the shift? Top Google News in 2015 AMIA 2018 | amia. org 19
Why the shift? Top Google News in 2016 AMIA 2018 | amia. org 20
Why the shift? Top Google News in 2017 AMIA 2018 | amia. org 21
Discussion & conclusion • To our knowledge, it is the first effort to understand HPV vaccination health beliefs from social media • Performed a longitudinal analysis of health beliefs for a large-scale population by leveraging state-of-the-art deep learning model and social media data • Among the early efforts on mining social media data for cancer related communication, this study could facilitate targeted interventions for cancer prevention AMIA 2018 | amia. org 22
Acknowledgements • UTHealth • Dissertation committee • Dr. Fang Li • Dr. Cui Tao • Dr. Yang Xiang • Dr. Hua Xu • Yuxi Jia • Dr. Yong Chen (UPenn) • Dr. Degui Zhi • Dr. Sahiti Myneni • Dr. Yang Gong • Dr. Trevor Cohen (U of W) • Dr. Yaoyun Zhang • Texas A&M • Dr. Lu Tang • University of Florida • Dr. Jiang Bian • Grants • NIH R 01 LM 011829 • NIH R 01 AI 130460 • CPRIT RP 160015 co-author AMIA 2018 | amia. org 23
AMIA is the professional home for more than 5, 400 informatics professionals, representing frontline clinicians, researchers, public health experts and educators who bring meaning to data, manage information and generate new knowledge across the research and healthcare enterprise. AMIA 2018 | amia. org @AMIAInformatics @AMIAinformatics Official Group of AMIA @AMIAInformatics #Why. Informatics 24
Thank you! Email me at: jingcheng. du@uth. tmc. edu
Overall framework Figure 1 -a: Real-time access large-scale public perceptions of vaccines using social media AMIA 2018 | amia. org 26
- Slides: 26