L.G.P. Peters1,2, G. Spada1, Dr. S. Schneider3, Dr. A. Middelweerd1, Dr. S. Wopereis1,2
1Research group Biomedical Signals and Systems, University of Twente, Enschede, Nederland. 2Research group Microbiology & System biology, TNO, Leiden, Nederland. 3Research group Human Media Interaction, University of Twente, Enschede, Nederland.
Lotte.peters@utwente.nl
Background: Nutrition is an increasingly important part of type 2 diabetes (T2D) treatment. However, accurate dietary assessment methods, like questionnaires, are often burdensome and fail to sustain user engagement, resulting in unreliable data. The Nutri+ questionnaire is a validated, web-based tool for assessing habitual intake, but faces usability and engagement challenges. Digital tools, including large language model-based systems, offer new opportunities to improve both assessment quality and user experience. An AI chatbot has therefore been developed which, through conversation and using engagement techniques, asks about dietary intake. In this pilot study, the Nutri+ chatbot was evaluated against the Nutri+ questionnaire.
Methods: A mixed-methods, co-creation study involving healthy adults comprised three phases. In the user testing phase (n=33), a personalised and a system-assigned group, as well as two engagement techniques (gamification and nudging), were evaluated using the User Engagement Scale (UES; 1–5) and the Chatbot Usability Questionnaire (CUQ; 0–100).
Results: The Nutri+ chatbot achieved higher user engagement than the Nutri+ questionnaire (UES 3.22 versus 2.82; p<0.001). No clear differences were found between the personalised (UES 3.14) and system-assigned group (UES 3.30), nor between the gamified (UES 3.29) and nudged versions (UES 3.19). The average usability score for the Nutri+ chatbot was 70.6/100. The system-assigned group scored higher on usability than the personalised group (CUQ 76.6 versus 64.3; p=0.02), whilst the gamified and nudged versions achieved comparable scores (CUQ 75.1 versus 68.3).
Discussion/Conclusion: The Nutri+ chatbot represents a promising step towards a more interactive way of collecting dietary intake. The results suggest that a conversational approach can increase user engagement, and thereby potentially improve the quality of dietary intake data and long-term compliance upon repeated assessments. Healthcare professionals will be able to provide more personalised dietary advice when more accurate dietary intake data becomes available. Next, the chatbot can be tested with people living with T2D.