A Feasibility Study on Usability and Trust among Population Groups of a Medical Avatar Supported by Large Language Models with Retrieval Augmented Generation

Healthcare professionals have limited time to support patients and their relatives, but their information needs are high. Therefore, the Radboud University together with the Canisius Wilhelmina Hospital hospital developed a speaking virtual hu-man avatar which, contrary to many avatars, uses a Large Language Model (LLM) enhanced with Retrieval Augmented Generation (RAG). The RAG tech-nique enables medical information supplied by the hospital to be utilized during interactions, rather than generic LLM information. Two videos were produced, one presenting a patient-avatar interaction regarding a total hip surgery, and an-other one presenting an interaction between a relative of a patient and the avatar concerning postoperative delirium. A survey was conducted among adults over 40 from the Netherlands, the UK and the USA to study the effects of gender, country and education level on usability and trust, which are important factors for avatar acceptance. Participants watched videos, imagining themselves as the pa-tient (video 1) or relative (video 2), and rated the constructs on a 7-point Likert scale (0-6). 165 persons (MeanAge=51.6, SDAge=8.9, Male=80, Female=85) completed the survey. In the patient role, participants scored the usability as M=4.61 (SD=0.97) and trust as M=3.92 (SD=1.10), all above the mean scale value. In the role as relative to the patient, participants scored usability as M=4.64 (SD=1.08) and trust as M=4.31 (SD=1.06). No effects were found of gender, country and education level.