Design, development, and use of conversational agents in rehabilitation for adults with brain-related neurological conditions: a scoping review.

Abstract

The objective of this review was to identify how conversational agents are designed and used in rehabilitation for adults with brain-related neurological conditions. Adults with brain-related neurological conditions experience varied cognitive and functional challenges that can persist long term. However, rehabilitation services are time- and resource-limited, and novel rehabilitation approaches are warranted. Conversational agents provide a human-computer interface with which the user can converse. A conversational agent can be designed to meet specific user needs, such as rehabilitation and support. Studies focused on the design and use of conversational agents for rehabilitation for people aged 18 years or older with brain-related neurological conditions were considered for inclusion. Eligible publication types included peer-reviewed publications (quantitative, qualitative, and/or mixed methods study designs; research protocols; peer-reviewed expert opinion papers; clinical studies, including pilot trials; systematic or scoping reviews), full conference papers, and master's or PhD theses. Eligible types of research included prototype development, feasibility testing, and clinical trials. Online databases, including MEDLINE, Scopus, ProQuest (all databases), Web of Science, and gray literature sources were searched with no date limit. Only English publications were considered due to a lack of resourcing available for translations. Title and abstract screening and full-text review were conducted by two independent reviewers. Data extraction was shared by three independent reviewers. The data extraction instrument was iteratively refined to meet the requirements of all included papers, and covered details for technological aspects and the clinical context. Results are presented narratively and in tabular format, with emphasis on participants, concept and context, and data extraction instrument components. Eleven papers were included in the review, which represented seven distinct conversational agent prototypes. Methodologies included technology description (n = 9) and initial user testing (n = 6). The intended clinical cohorts for the reported conversational agents were people with dementia (n = 5), Parkinson disease (n = 2), stroke (n = 1), traumatic brain injury (n = 1), mixed dementia and mild cognitive impairment (n = 1), and mixed dementia and Parkinson disease (n = 1). Two studies included participants who were healthy or otherwise from the general community. The design of the conversational agents considered technology aspects and clinical purposes. Two conversational agent prototypes incorporated a speaking humanoid avatar as reported in five of the papers. Topics of conversation focused on subjects enjoyable to the user (life history, hobbies, where they lived). The clinical purposes reported in the 11 papers were to increase the amount of conversation the user has each day (n = 4), reminiscence (n = 2), and one study each for anxiety management and education, Parkinson disease education, to obtain and analyze a recording of the user's voice, to monitor well-being, and to build rapport before providing daily task prompts. One study reported clinician oversight of the conversational agent use. The studies had low sample sizes (range: 1-33). No studies undertook effectiveness testing. Outcome measures focused on usability, language detection and production, and technological performance. No health-related outcomes were measured. No adverse events were reported, and only two studies reported safety considerations. Current literature reporting the design and use of conversational agents for rehabilitation for adults with brain-related neurological conditions is heterogeneous and represents early stages of conversational agent development and testing. We recommend, as per our customized data extraction instrument, that studies of conversational agents for this population clearly define technical aspects, methodology for developing the conversation content, recruitment methods, safety issues, and requirements for clinician oversight.