Integrating a behavior change model into the conversation design of a low-cost chatbot to address vaccine misinformation: A Feasibility Study (Preprint)

Abstract

The COVID-19 pandemic highlighted the pros and cons of current health communication approaches. Social media became a primary medium by which people received information. Misinformation, disinformation, and conspiracy theories undermined the propagation of sound health information during the pandemic, so much so that the WHO called it an infodemic. Inaccurate and false information severely impacts public health, delaying individual health choices to take preventative measures, and influences vaccine uptake. Using social media for health communication is increasing in popularity due in part to the ability to reach more significant numbers of people. The use of chatbots in healthcare has significantly increased in recent years. Chatbots enable users to have humanlike conversations on various topics and can vary widely in their complexity and functionality.

This paper aims to describe the development and feasibility testing of a low-tech chatbot called VWise. VWise is designed to engage participants in correcting misinformation around COVID-19 vaccinations by utilizing Motivational Interviewing (MI) as a behavior change model.

We developed personas for both the bot and the potential participants. Using our participant personas, the research team engaged in mock conversations. Transcriptions of the conversations were coded to identify phases of MI and where conversations diverged from these phases. Several iterations of sample dialogues were created. We selected to use ManyChat, a cloud-based platform, for its easy-to-use interface. Relevant participant responses were first stored into variables and mapped to a pre-configured Google Sheet that became our data set for analysis. No identifying data, generated by ManyChat was included in our data set. Responses to qualitative questions were deductively coded by two independent researchers. VWise was pilot tested on a group of 33 participants.

Out of 33 participants, 17(51%) chose to continue the conversation with VWise till the end as a mark of engagement. When they were asked about the influence the conversation had on them, 10 (6 -fully vaccinated, 4- partially vaccinated) expressed positive opinion, 1 partially-vaccinated expressed a neutral opinion, and 6 participants did not answer (5-fully vaccinated, 1-non-vaccinated). A validated tool, Chatbot Usability Questionnaire (CUQ) was included at the very end of the conversation. Of the 17 people who concluded the chat, 13(76.5%) participants filled out the CUQ. The mean score was 70.9, (SD±19.4), median score was 78.1, with the lowest and highest scores being 34.4 and 95.3 respectively.

This study presented the development and feasibility testing of VWise, a low-tech chatbot aimed at addressing COVID-19 vaccine misinformation through engaging participants in the behavior change process led by MI techniques. A high level of engagement with the bot was demonstrated. Our study highlights that low-tech bots are a viable option for use in health communication and in the promotion of behavior change. Our conversational model, based on MI, was successful is observing change talk and resistance in participants, furthering the argument that NLP is not absolutely necessary to produce observations of readiness for behavior change.