Optimizing Patient-Manager Interactions in Chronic Care Using Generative Artificial Intelligence QA Dialogue Systems

Chronic disease management requires continuous care to prevent exacerbations, yet providing 24/7 accessible medical advice is a significant challenge for care providers. The current model of periodic patient visits with standard dialogue responses fails to adequately address the individualized needs of patients and caregivers. This lack of personalized guidance undermines treatment adherence and satisfaction. To bridge this gap, we developed a novel approach leveraging generative AI (GAI) to optimize patient-manager interactions. Focusing on prevalent chronic conditions such as diabetes, chronic kidney disease, and dementia, we created intelligent dialogue agents that deliver tailored, actionable nursing recommendations aligning with each case's unique circumstances. The developed methodology began by curating a corpus of 30 frequently asked questions on home care, emotional support, and social factors from online patient data across the three diseases. We employed prompt engineering to precisely define the virtual roles of nurses and physicians while implementing a “Rephrase and Respond” (RAR) protocol to enhance query comprehension and generate reliable solutions. The RAR approach solves problems by question restating, followed by contextualized recommendations from the language model. Clinical expert feedback iteratively refined the system's outputs. The evaluation result showed an average gain of 10% in response and an increase of 5% in readability after optimization. The optimized RAR method of large language models effectively improved the quality of nursing advice for chronic disease, aligned with patients' emotional and social needs. The RAR method in medical advisory areas, including chronic disease management and rehabilitation care, enhances the quality and practicality of nursing advice in these domains.