Chronic Diseases Management through AI-assisted Chat-Based Online Health Communities

Abstract

Chronic diseases have emerged as one of the most pressing challenges in global public health. These diseases are shaped by long-term patterns in lifestyle and behavior, such as dietary choices, activity levels, and psychological resilience. Traditional management often depends on periodic hospital visits, long-term medication, and self-managed lifestyle changes between clinical appointments. This approach places a heavy burden on healthcare systems and provides users with limited support outside clinical settings. Many users receive little daily guidance, and the lack of continuous and personalized management makes it difficult to maintain healthy behaviors over time. These limitations highlight the need for scalable and professional management models that can deliver continuous behavioral guidance in everyday life.

In recent years, advances in digital health technologies and Artificial Intelligence (AI) have created new possibilities for chronic diseases management (CDM). Wearable devices, mobile health applications, and online health communities (OHCs) now enable large-scale and personalized support for daily health behaviors. Within OHCs, community operations, peer interaction, and data-informed guidance help users adjust their lifestyles, strengthen health awareness, and maintain positive behavioral patterns. At the same time, AI has become increasingly important in CDM, particularly in areas such as dietary assessment, health monitoring, and behavioral analysis. These tools provide timely and professional support that enhances the responsiveness and precision of digital health interventions.

Building on these technological advances, this thesis introduced a new AI-assisted chat-based OHCs for CDM. We evaluate this approach through a large-scale real-world application: the Ping An Million Chronic Disease Program, which has served nearly 1.5 million users over the past four years. Using the large-scale user data from this program, this thesis evaluates the effectiveness of this management model, analyzes the application of AI technology within the community, and evaluate whether this model can be effectively extended to multiple CDM settings.

This thesis consists of three main parts.

The first part introduces the novel chat-based OHC model through real-time group interactions and structured behavioral tracking. We draw on real-world data from over 60,000 users and quantify the effectiveness from the combined mechanism of social support, social comparison, and self-monitoring. We examine four types of social support (informational, emotional, instrumental, appraisal), three types of social comparison (upward, lateral, downward), and three types of self-monitoring (health indicator check-ins, meal logging, health education) across a 21-day intervention and one-year follow-up, and use hierarchical mixed-effects models to evaluate their effects on obesity, hypertension, and diabetes. We find that all four types of social support are positively associated with weight loss, with informational support showing the largest association. Only downward comparison is significantly associated with better outcomes, suggesting that realistic and achievable benchmarks are more motivating than aiming for much higher or similar standards. The effectiveness of self-monitoring is disease-specific: health indicator check-ins and meal logging are most effective for weight loss, while health education is essential for hypertension and diabetes. Mid-sized groups (around 100 users), higher speaking rates, and lower dropout rates are associated with better outcomes, highlighting the importance and potential of platform design.

The second part examines the role of AI compared with human nutritionists in chronic disease management through a 21-day weight-loss intervention. We conduct a randomized controlled trial with 802 users and evaluate the two approaches from four perspectives: direct outcomes, behavioral improvements, user feedbacks, and nutritionist assessments. The results show that both AI and human nutritionists lead to meaningful weight-loss improvements, with AI demonstrating even stronger effects. AI performs comparably to human nutritionists in promoting behavioral change and offers a more balanced distribution of social support, suggesting that AI can deliver more stable and scalable guidance. User feedbacks are similar across the two groups, while nutritionists rate AI more favorably in terms of accuracy and content quality, which indicates AI's potential to provide high-quality feedback at scale. Evidence from more than 20,000 additional real-world users further confirms that AI can deliver reliable, scalable, and cost-efficient support in online health communities without compromising effectiveness.

The third part extends the AI-assisted chat-based OHCs model to Metabolic Dysfunction Associated Fatty Liver Disease (MAFLD). We conduct a randomized controlled trial with 667 users and evaluate the effectiveness of user engagement and a multi-agent intervention design. The study implements a multi-agent intervention design that integrates dietary guidance, exercise support, report interpretation, and health education together. The results show that this design improves short-term outcomes and highlights how motivation, health literacy, and structured support jointly shape chronic disease management. Engagement plays a critical role in MAFLD management, as highly engaged users achieve better outcomes even compared with users who have higher baseline disease risk. These findings validate the potential of this management model to be applied across a wider set of CDM settings.

Date of Award	16 Apr 2026
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Siyang GAO (Supervisor)