The feasibility and preliminary effectiveness of a machine learning (ML)-based digital assistant to promote physical activity in older adults

This study will involve adults 65 years of age or older syncing their tracker to an existing ML-based physical activity digital assistant ‘MoveMentor’. The intervention administered to participants is of a 14-week duration. All participants will participate in the intervention. This intervention uses a smartphone app that integrates with a physical activity tracker (such as a Fitbit). The primary feature of this app is a digital assistant for physical activity. The minimum time that participants are asked to interact with this app will last no more than 5-10 minutes, which will be completed at a pace determined by the participants. This study will assess the feasibility (usability, acceptability, and engagement), preliminary effectiveness, and user experience. The MoveMentor digital assistant incorporates several key features: 1) Generative Pre-trained Transformers (GPT) to simulate natural conversation, 2) reinforcement (machine) learning to optimise the timing and content of messages, and 3) real-time data from a tracker, weather, and Global Positioning System (GPS). The digital assistant will provide tailored advice on physical activity based on user inputs.. Aimed at motivating participants to increase their physical activity and adhere to national physical activity guidelines, the intervention offers detailed sessions delivered as a series of interactive "Conversations," powered by DialogFlow to facilitate natural interaction. During the conversations, the digital assistant engages participants with specific questions (e.g., ‘What stops you from being active?’) about their physical activity, health status, self-efficacy, and social support. Based on their responses, participants will receive personalised advice on the benefits of being active (i.e., why be active) and strategies for increasing activity (i.e., how to be active); for example "Did you know that exercising, walking, and gardening have a very positive effect on wellbeing?". This tailored advice aims to help individuals understand the importance of being physically active and provides evidence-based steps to their engagement in it. The digital assistant will engage participants in personalised conversations providing content tailored to support and improve adherence to the physical activity intervention. These conversations are informed by real-time physical activity data collected from activity trackers. To encourage and support physical activity, participants will receive just-in-time personalised ‘Nudges’ through smartphone notifications. These nudges are generated by a reinforcement learning algorithm that considers factors like the frequency, timing, and context, to deliver tailored suggestions. Considering a variety of data sources, the algorithm ensures that nudges are sent at the most effective times for each individual. These data sources include: 1) data from preferences and settings during sign-up, 2) personal data (e.g., health status, age) via Conversations, 3) physical activity monitor (e.g., Fitbit), 4) participant preferences (e.g., likes and dislikes, 5) GPS (e.g., at work), and 6) weather conditions (e.g., favourable conditions for outdoor activity). Additionally, participants can initiate a ‘Questions and Answers’ session with the digital assistant at any time. They can interact through voice or text to ask a variety of physical activity related questions, including: 1) knowledge questions (e.g., ‘I have sprained by ankle, how do I stay active?’), 2) activity status questions (e.g., ‘How many steps did I take yesterday?’), 3) goal-based questions (e.g., ‘Am I meeting my activity goals?’), 4) location-based questions (e.g., ‘Where are there flat walking pathways?’), 5) suggestions (e.g., ‘I feel like being active today, what do you suggest?’). In addition to interacting with the digital assistant, MoveMentor aids users in goal setting and action planning. Participants receive a daily activity goal that gradually increases over time based on their performance. If they are not consistently meeting the goal, it will gradually decrease. The action plan helps participants outline key activity details, such as: 1) which physical activities they will engage in, 2) when during the day they will do them, and 3) how many times a week they plan to be active.