With diabetes projected to affect over 1.3 billion individuals globally by 2050, delivering personalised support has become a critical component of disease management. This requires recognising the diverse and complex needs of patients, particularly those with chronic conditions such as diabetes. A recent study analysed over 500,000 secure patient messages using artificial intelligence (AI) and natural language processing (NLP) techniques to understand patient concerns and develop AI-based tools for more responsive care. These tools were assessed by clinicians for their usefulness and associated risks, providing valuable insights into the potential of AI in supporting tailored patient care.
Understanding Patient Concerns Through Secure Messaging
Secure patient portals are a widely used channel for individuals to communicate their health concerns. The analysis of more than 528,000 endocrinology-directed messages from over 11,000 patients with diabetes revealed key themes in patient concerns. Dietary management and meal-related questions were among the most frequently discussed topics, alongside interpretation of laboratory results and administrative challenges related to insurance and appointment scheduling.
Must Read: The Role of AI in Restoring Human-Centred Healthcare
Twelve core themes were identified, including medication dosage, use of diabetes technology (such as glucose monitors), hypoglycaemia, bone health and thyroid management. Interestingly, thyroid-related messages formed a distinct, siloed cluster, suggesting the need for targeted communication strategies. These insights underscore the depth and variety of issues patients seek help with, highlighting an opportunity for AI to deliver targeted, consistent responses that align with patient needs and preferences.
AI as a Clinical Assistant: Usefulness and Perceived Risks
Using generative AI models such as ChatGPT-4, researchers drafted potential AI tools designed to assist healthcare professionals in responding to the identified patient concerns. Five experienced endocrinologists then evaluated these tools. Overall, the AI assistance received high usefulness ratings, particularly in areas related to patient education and administrative efficiency.
Highly rated tools included the creation of educational content for dietary management, automated responses to administrative queries and templated replies for lab result explanations. These functions could significantly reduce the workload on clinicians while enhancing patient understanding and engagement. However, AI tools that involved direct interpretation of real-time patient data or message triaging were perceived as riskier. Concerns centred around the potential for misinterpretation and over-reliance on automated systems without human oversight.
Tailoring Care to Diverse Needs and Demographics
The study also examined how patient demographics influenced the content of messages. Differences were observed across race, ethnicity, sex and marital status. For instance, White and female patients were more likely to raise issues about bone health and mental distress, while male and non-White patients more frequently discussed dietary and device-related topics. These findings suggest that AI tools must be adaptable to address varying needs among different patient groups.
During the COVID-19 pandemic, the volume of messages increased significantly, particularly those related to meal planning and dietary concerns. This shift reflects the heightened vulnerability of patients with diabetes during the pandemic and the growing reliance on digital health platforms. AI-driven tools could support real-time updates about medication availability, policy changes and direct scheduling support, which are especially beneficial during healthcare disruptions.
By leveraging NLP and generative AI, the study successfully identified key areas where patients with diabetes seek support and explored practical AI applications to address these needs. Clinician evaluations affirmed the value of AI in enhancing administrative workflows and patient education, although cautious integration is advised for tools dealing with real-time data interpretation.
The results affirm that AI has the potential to transform patient communication and tailored support in diabetes care. It also underscores the importance of considering demographic variations and ensuring equitable access to AI-driven tools. Future developments should prioritise transparency, safety and inclusivity to maximise the benefits of AI in patient-centred healthcare. As digital platforms become more integral to chronic disease management, the thoughtful integration of AI will be crucial in achieving long-term health outcomes.
Source: npj Digital Medicine
Image Credit: iStock
