Development of a vaccine informatics system and its application to identifying the impact of vaccine debate on immunization rates during a global pandemic

Vaccine debate has been on social media for more than a decade, and a surge of anti-vaccine activities on social media has been detected during prior disease outbreaks. Nonetheless, how this debate changes and impacts the uptake rates for crucial vaccines during the COVID-19 pandemic remains unknown. The long-term goal is to counteract the negative impact of misinformation on digital platforms that threatens public health. The overall objectives of this application are to develop a publicly accessible vaccine informatics system to track vaccine debate, and to test the impact of vaccine debate on COVID-19 (if developed by 2021), flu, and HPV immunization rates during the onset of a global pandemic. The central hypothesis is that vaccine debate will increase and become more negative during the pandemic, leading to lower vaccine uptake rates. The rationale for this project is that discovering how vaccine debate changes and influences vaccine uptake rates during a pandemic will be critically important for managing and preventing disease spread. The central hypothesis will be tested by pursuing two specific aims: 1) Develop a vaccine informatics system to identify the frequency and valence of vaccine debate during and following the pandemic compared to the pre-pandemic baseline; and 2) Apply this system to identify the causal impact of vaccine debate on immunization rates during the pandemic. Under the first aim, ~1 million social media posts will be collected, and a deep-learning algorithm for classifying multimodal social media posts will be developed. This algorithm will address potential bias and noise in human annotations of vaccine debate that is increasingly politicized. The classification results will be tabulated in a Web portal so that daily and weekly statistics about pro- and anti-vaccine posts will be readily available. Under the second aim, a multimethod approach will be proposed that resolves the current barriers in research on vaccine refusal. This approach will use a survey of 2,000 individuals who represent the US population. The survey responses will be combined with the respondents' prior engagement with vaccine debate retrospectively collected from social media. These engagement data will be then classified by the machine- learning algorithm developed in Aim 1. This research is innovative because it proposes a robust co-teaching framework for addressing noisy human annotations of vaccine debate. It also proposes a statistical modeling technique that involves heterogenous metrics obtained from a multi-method approach for hypothesis testing. These innovations are timely and urgent as the current time presents a rare opportunity to identify the impact of vaccine debate on public health during the onset of a global pandemic. The feasibility of this proposed research is clear from the solid preliminary datasets collected from 2018-2020 that establish the pre-pandemic baseline. The proposed research is significant because it will produce a public barometer of vaccine debate and provide a methodological breakthrough in uncovering the reasoning behind refusing crucial vaccines during the global pandemic.