Exploring the use of conversational agents in healthcare: an overview, conceptual analysis and case study

• Main Author: Dhinagaran, Dhakshenya Ardhithy
• Other Authors: Lorainne Tudor Car
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Science::Medicine::Computer applications
• Online Access: https://hdl.handle.net/10356/163179
• Institution: Nanyang Technological University

Conversational agents (CAs) or chatbots are computer programs designed to participate in human-like conversations with their users. The growing interest in these agents is evident in their employment in numerous fields including business, retail, and even healthcare. The need for new initiatives to promote healthy living is apparent in Singapore given its highly stressful and fast-paced way of life, insufficient time for physical activity, and frequent dining out. Singapore is technologically astute and always on the lookout for novel initiatives, such as the use of CAs, to improve patient care and support healthcare staff. This project was grounded in the underlying principles of evidence-based medicine and employed a multitude of study designs and methodologies. Firstly, a (1) scoping review was conducted to map out and evaluate the existing evidence on CAs in healthcare. Secondly, a (2) conceptual framework for rule-based healthcare CA design and development was proposed based on existing mHealth and CA development frameworks. Thirdly, a (3) healthy lifestyle change CA was developed and piloted based on guidance from the proposed conceptual framework and (4) co-design/needs assessment with members of the public through a series of semi-structured interviews. (5) Follow-up interviews were then conducted with participants from the pilot study who reported on their experience interacting with the healthy living CA and their recommendations for its improvement in the future. Finally, a (6) systematic assessment of CA apps was performed in Google Play (android) and Apple (iOS) stores to map out and assess the characteristics and behavior change techniques used in those healthcare CAs currently available to the public for healthy lifestyle behavior change. The results indicated that this field is still in its nascent stages with a predominance of small case studies and a dearth of large-scale randomized controlled trials. Most of the CAs reported in the literature were standalone smartphone apps that were text-based and employed machine learning. The conceptual framework used existing frameworks as a foundation and culminated in a five-step framework guiding the user through defining the aim, creating an identity, developing the content, building the conversation flow, and piloting the CA. The CA designed and developed for a healthy lifestyle change, Precilla, aimed to include those recommendations made by volunteer participants in the needs assessment interviews. Their suggestions included providing educational information on diabetes and prediabetes, risk factors, and risk prevention. They also suggested a friendly personality and simple, layman terms to be used in the conversations with Precilla. Some participants were very satisfied with Precilla, found her informative, and thought the information shared was very valuable and a good reminder to make healthier choices with regards to healthy eating, physical activity, sleep, and stress. Conversely, others were disappointed in the lack of artificial intelligence and would have liked to have the opportunity to “chit-chat” with Precilla without any specific endpoint. One of the main recommendations for improvement suggested introducing some personalization to the CA in terms of more tailored content and message delivery timings. Incentives such as rewards and point collection were also of interest and participants had high regard for formal and informal group support in enabling them to achieve their healthy lifestyle change goals. Finally, the systematic assessment of healthy living apps from the app stores indicated that the pool of CA apps in this area is fairly small. The apps focus on just one area of healthy living either diabetes prevention, wellness and physical activity, sleep improvements, or healthier eating. The apps were all free with only a couple offering in-app purchases and no advertisements were featured. CA personality was sometimes neglected in most apps which came across as very rigid, factual, and not human-like. This Ph.D. project has made several valuable contributions such as a timely update of the current progress and status of this field of research on healthcare CAs through the review and synthesis of evidence mapped in the scoping review and apps assessment. It has also proposed some guidelines on CA development which currently have not been strictly defined. Finally, a test of the feasibility and acceptability of a healthcare conversational agent for healthy living in Singapore revealed the applicability and/or necessity of such an agent and could propose future potential applications in Singapore’s healthcare system. This project has laid a good foundation for future research to tap into.