Conversational agent interventions in diabetes care: a systematic review

Abstract

This systematic review aimed to evaluate the effectiveness, acceptability, and safety of conversational agent (CA) interventions in diabetes care. CAs are artificial intelligence driven tools that simulate human-like dialogue and have emerged as promising supports for self-management in chronic disease. We searched six electronic databases from inception to June 2024 and identified 16 eligible studies involving 9076 participants across 13 countries. Included studies varied in design, population, diabetes type, and intervention duration. Eleven studies assessed effectiveness, with most reporting improvements in glycemic control (e.g., HbA1c reductions of 0.3 % to 1.0 %), medication adherence, health behaviours (e.g., diet, physical activity), or mental health outcomes (e.g, anxiety). Thirteen studies examined acceptability and found that most users had positive emotional and motivational responses, though some expressed dissatisfaction with repetitive or impersonal interactions. Only four studies addressed safety, and while adverse events were rare, mechanisms such as clinical escalation protocols were inconsistently applied. Most studies were rated as weak in methodological quality, with small samples and limited use of control groups. In conclusion, CAs show promise as scalable, patient-centered tools for diabetes care. However, rigorous research is needed to better understand their clinical impact, safety, and suitability for diverse patient populations.