Conversational Agents for Sustaining Engagement to eHealth

In recent years, there has been a significant rise in digital health technologies including smartphone apps. These technologies potentially solve a critical issue in health care delivery — wide dissemination of evidence-based interventions in a scalable manner. As such, it is understandable that there is a certain optimism and enthusiasm about eHealth technologies and their potential ability to reshape health care.

However, most eHealth technologies suffer from low engagement and adherence issues. For example, data from the PTSD Coach app — developed by the Department of Veterans Affairs (VA) — indicate that only 5.5% patients remained engaged after a year. Successful interventions for better health outcomes often require high engagement and adherence from participants. The failure of current technologies to sustain longitudinal engagement thus can lead to non-optimal outcomes and undermine their efficacy. As such, there is an immediate and urgent need to develop new theoretical frameworks and methodologies to address the engagement and adherence issues.

We aim to address these critical issues by developing conversational agents (CAs). These domain-specific agents are designed to provide interactive psychoeducation, assessment tools, and personalized illness management strategies while maintaining social presence and process accountability to sustain user engagement.

The development of such CAs will involve solving a number of technical and design challenges. Specifically, CAs require turn-taking and interactivity, which results in active information flow in both ways rather than just passive delivery of information content. This involves i) natural language processing for understanding user input and broader context; ii) dialogue management; iii) adaptive interaction design. We also leverage active listening and supportive accountability theory to ensure social presence and sustained user engagement.

The outcomes of this project will not only advance the theory and methods for designing CAs but also address an urgent need of the broader eHealth community.



Support for this research is provided by the National Science Foundation under Grant No. 1850287.


Hee Jeong Han
Sanjana Mendu
Beth K. Jaworski
Jason E Owen
Saeed Abdullah