Pilot RCT of PIPE-AI for Prediabetes and Diabetes Self-Management

A Personalized Interactive Patient Empowerment Artificial Intelligence Platform (PIPE-AI) Enhanced by DiabetesGPT in Prediabetes and Diabetes Patients in Primary Healthcare Settings

This pilot randomized controlled trial evaluates the feasibility, effectiveness, and acceptance of the Personalized Interactive Patient Empowerment Artificial Intelligence Platform (PIPE-AI) enhanced by DiabetesGPT among adults with prediabetes or diabetes in Hong Kong primary healthcare settings. Participants will be randomized in a 1:1 ratio to an intervention group receiving PIPE-AI or to a waitlist control group receiving usual care during the first 3 months. Both groups will complete baseline and 3-month assessments, including patient-reported outcome measures, clinical outcomes, healthcare service utilization, and intermittent continuous glucose monitoring using Abbott FreeStyle Libre 2 Plus sensors.

This study is a pilot randomized controlled trial designed to assess the feasibility, effectiveness, and acceptance of the PIPE-AI platform enhanced by DiabetesGPT before broader implementation in primary healthcare settings. The PIPE-AI platform integrates individualized risk assessments and a locally fine-tuned large language model, DiabetesGPT, to generate personalized health advice and patient empowerment support for people with prediabetes or diabetes.

Approximately 50 participants will be recruited and randomized in a 1:1 ratio to the intervention group or waitlist control group. Eligibility will be assessed by nurses and doctors in participating primary healthcare settings. After written informed consent, trained research assistants will conduct baseline assessments, including demographic and socioeconomic data, lifestyle behaviours, medical history, current medications, patient-reported outcome measures, and venous blood sampling for fasting glucose and HbA1c.

Randomization will be performed before recruitment by a statistician using R software. After baseline assessment, participants will receive an opaque letter containing their assigned group and follow-up instructions. Research assistants involved in subject recruitment and baseline assessment will be blinded to grouping to reduce measurement bias where operationally feasible.

Participants in the intervention group will install the PIPE-AI app and receive instructions on how to use it. An individualized patient empowerment programme will be provided during the 3-month follow-up period. App login frequency will be monitored on the server, and reminders may be sent by SMS, WhatsApp, or WeChat if a participant does not log in within one week after recruitment or stops logging in for more than one month.

Participants in the control group will follow a waitlist approach. During the first 3 months, they will receive usual care, including group-based patient empowerment programmes where available through the District Health Centre, general guidance from the Hong Kong Reference Framework for Diabetes Care for Adults in Primary Care Settings, or routine diabetes education and management services in their respective clinics. After the 3-month follow-up assessment, control group participants will be given access to the app.

All RCT participants in both groups will use Abbott FreeStyle Libre 2 Plus continuous glucose monitoring sensors intermittently during the 3-month trial. After successful enrolment and baseline assessments, participants receive two CGM sensors for use during Weeks 1-4. If CGM data are successfully collected in the study system, the research team will contact the participant near the end of the trial to collect a third CGM sensor for use during Weeks 11-12. CGM data will be downloaded or exported for research analysis using pseudonymized study IDs.

Participants in both groups will complete a 3-month follow-up assessment, including patient-reported outcome measures, clinical outcomes, healthcare service utilization, and CGM-related data where applicable. Main analysis will adopt an intention-to-treat principle, with per-protocol analysis as sensitivity analysis. As this is a pilot RCT, CGM analyses will be exploratory and will assess feasibility, adherence, completeness of CGM data collection, glycaemic profiles, and signal of change for future definitive trials.

Inclusion Criteria:

• Adults aged 18 years or older.
• Diagnosed with prediabetes or diabetes by their family doctors.
• Able to understand Chinese.
• Possess a mobile phone that can install the app developed for this project.

Exclusion Criteria:

• Under legal guardianship.
• Comorbid schizophrenia.
• Intellectual disability.
• Incompetent in giving consent.
• Other significant cognitive disorders.