Developing Conversational User Interfaces To Deliver Evidence-Based Health Interventions
Restricted (Penn State Only)
- Author:
- Mendu, Sanjana
- Graduate Program:
- Informatics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 05, 2024
- Committee Members:
- Xinning Gui, Major Field Member
Saeed Abdullah, Chair & Dissertation Advisor
Andrew High, Outside Unit & Field Member
Kenneth Huang, Major Field Member
Dongwon Lee, Professor in Charge/Director of Graduate Studies - Keywords:
- healthcare
human-computer interaction
mixed-methods
conversational user interface
generative AI
smart speaker
chronic pain
physical activity
health intervention - Abstract:
- Although healthcare is a universal need, many people lack access to quality healthcare delivery services. Technology-supported interventions have become a popular option for addressing this gap. Conversational User Interfaces (CUIs), in particular, have emerged as a novel tool for supporting longitudinal engagement with health interventions due to human-like qualities of interaction. Despite the large body of existing work in this area, limited research has focused on designing CUIs within the context of evidence-based health interventions. This dissertation aims to address this gap by shedding light on key considerations for developing CUIs in alignment with existing methodological frameworks for health intervention delivery. Specifically, I draw upon findings from the design, development, and evaluation process for two unique CUIs for different health intervention contexts. The first CUI is a voice assistant to support engagement with mindfulness practices for individuals living with chronic pain. I evaluated the system using semi-structured interviews with experienced mindfulness facilitators to ensure prototype fidelity to the original intervention (MBSR) followed by a 4-week pilot study with individuals living with chronic pain. The second CUI is a theory-informed prompt engineering architecture to develop personalized, principle-based intervention content to promote physical activity using generative AI. I evaluated the resulting messages and images using both computational metrics and subjective feedback from human raters to understand the diversity and acceptability of content. While the messages have not yet been deployed in an end-to-end message delivery pipeline, these findings serve as a meaningful foundation for such a future deployment. This dissertation makes three main contributions. Firstly, my findings sheds light on novel mechanisms to support the dynamic personalization of digital health intervention delivery while maintaining fidelity to theoretical frameworks. Secondly, this work highlights opportunities and challenges in leveraging CUIs in the context of health intervention delivery. Lastly, this work provides unique insights into the acceptability and feasibility of using CUIs to support marginalized populations and move beyond traditional pathways for health intervention delivery.