Effectiveness of Artificial Intelligence-Based Health Education Accurately Linking System (AI- HEALS) for Type 2 Diabetes Mellitus Self-Management: A Cluster Randomized Clinical Trial

<title>Abstract</title> <p> Background Type 2 diabetes mellitus (T2DM) requires sustained self-management, for which digital health interventions offer scalable support. However, evidence on their long-term effectiveness in real-world primary care remains limited. We evaluated the effectiveness of the Artificial Intelligence-based Health Education Accurately Linking System (AI-HEALS), an AI-driven platform delivering personalized T2DM self-management support. Methods In this cluster-randomized controlled trial (July 2023–July 2024), 45 community health centres in Beijing, China, were randomly allocated 1:1 to AI-HEALS plus treatment-as-usual (TAU) or TAU alone, enrolling 406 adults with T2DM aged 18–75 years. Proactive intervention delivery occurred over the first 3 months. The primary outcome was change in glycated haemoglobin (HbA _1c ); secondary outcomes included self-management behaviours, diabetes-related skills, and psychological measures, assessed at 3, 6, 12, and 18 months. Outcomes were analysed using linear mixed-effects models, reporting adjusted between-group differences at each time point and overall across follow-up. Results Of 406 participants randomized, 391 were included in the analysis (AI-HEALS, n = 202; TAU, n = 189). HbA _1c declined in both arms, with no significant between-group difference at 18 months (adjusted mean difference, − 0.22%; 95% CI, − 0.54 to 0.10) or overall across follow-up (− 0.21%; 95% CI, − 0.48 to 0.06). AI-HEALS was associated with a small overall improvement in diabetes-related skills (1.69 points; 95% CI, 0.05 to 3.33) and higher dietary adherence at 6 and 12 months, though neither dietary adherence nor self-care activity showed a significant sustained overall effect. Medication adherence, self-efficacy, sleep quality, perceived stress, and quality of life did not differ significantly between groups. Conclusions AI-assisted personalized health education improved selected self-management behaviours, most consistently diabetes-related skills, but did not yield a sustained glycaemic benefit beyond usual care over 18 months. AI-HEALS offers a scalable digital health approach for community diabetes care, though durable metabolic benefit may require more sustained or adaptive interactive support. Trial registration ChiCTR2300068952 (2 March 2023). </p>