Co-Design and Usability Study of T2Cura: Conversational AI for Post-Diagnosis Type 2 Diabetes Support (Preprint)

BACKGROUND Background: As the country with the largest population of patients with type 2 diabetes mellitus (T2DM) globally, China faces multiple challenges including inefficient doctor-patient communication, uneven distribution of medical resources (particularly in remote areas), and low patient health literacy. Additionally, sociocultural factors such as limited consultation time and absence of follow-up mechanisms in traditional medical models, patients' cognitive biases regarding the disease (such as equating "sugar" with "sweet foods"), and high-carbohydrate dietary patterns further exacerbate patients' post-diagnosis self-management difficulties. Moreover, elderly patients and those with low educational backgrounds often struggle to understand complex medical instructions, while language and digital literacy barriers—especially in rural regions—further hinder their ability to access and act on care guidance. OBJECTIVE Objective: This study aimed to develop an intelligent system for post-diagnosis patient care to enhance disease awareness and self-management capabilities among patients with T2DM in the initial stages following diagnosis, and to evaluate the medical accuracy of system-generated content, the usability of human-computer interaction, and the application value of personalized intervention recommendations in post-diagnosis care management. We also sought to explore how sociolinguistic adaptability and culturally relevant content personalization could improve patient adherence and communication effectiveness. METHODS Methods: This study systematically elucidated the co-creation process and architectural design of the T2Cura system. We engaged multiple stakeholders, including clinical endocrinology specialists, T2DM patients, and their family members, through co-creation workshops to identify key needs in the post-diagnosis management phase and derive corresponding optimization strategies. Based on these strategies, we constructed a system prototype and conducted beta testing to evaluate its functional performance and effectiveness in simulated scenarios. In addition, we conducted a real user test, and the results showed that T2Cura has a significant effect in improving users' disease knowledge and health literacy. The system also received positive feedback in terms of usability and user engagement. RESULTS Results: This paper presents T2Cura—a continuous care system based on a hierarchical management framework that integrates a closed-domain knowledge base with localized explanation capabilities for medical terminology, combined with voice recognition and dietary recommendation modules. The system supports T2DM patients in the post-diagnosis phase to understand diagnostic and treatment information and obtain personalized care recommendations through AI-mediated dialogue. Beta testing results demonstrated that the system exhibited capabilities in recognizing medical terminology in conversations, generating reasonable and easily understood responses, and the voice recognition module showed good adaptability to various dialect inputs, with no misleading content generated. In particular, the system performed well in handling colloquial and ambiguous expressions, demonstrating robustness in real-world communication environments and effectively avoiding semantic errors. CONCLUSIONS Conclusions: This study validated the application potential of the T2Cura system in supporting post-diagnosis health management for patients with type 2 diabetes mellitus. The system integrates large language models with carefully constructed multi-source databases, effectively enhancing patients' understanding of medical terminology while demonstrating excellent performance in medical content accuracy, human-computer interaction usability, and adaptability to local languages. Its modular design allows for future integration with other chronic disease management systems and supports continuous iteration based on user feedback, indicating strong scalability and long-term value.