Large Language Models for Mitigating Disease-Related Anxiety Among Patients With Diabetes and Low Health Literacy: A Systematic Review

Abstract

Background: Low health literacy among diabetic adults exacerbates health inequities in the United States, with disproportionately higher prevalence in racial minorities. Such disparities raise healthcare costs due to excessive medical service utilization. Methods: This systematic review followed the PRISMA 2020 guidelines. Relevant studies published from January 2020 to April 2026 were retrieved from five databases. Two independent reviewers completed study screening, data extraction and quality assessment. Results: Twenty-six eligible studies were included, covering four types of LLM-based interventions. These interventions effectively improved patients’ disease understanding, self-efficacy and treatment compliance, and relieved anxiety. LLMs were proven scalable and cost-efficient to narrow health equity gaps, despite the digital divide acting as a major constraint. Conclusion: LLMs are promising scalable tools to reduce diabetes-related anxiety among low-health-literacy populations. It is recommended to integrate AI technologies into public health strategies to mitigate systemic healthcare disparities.