Conversational Agents Supporting Self-Management in People With a Chronic Disease: Systematic Review

Background Conversational agents (CAs) are increasingly used as a promising tool for scalable, accessible, and personalized self-management support of people with a chronic disease. Studies of CAs for self-management of chronic disease operate within a multidisciplinary domain: self-management originates from (behavioral) psychology and CAs stem from intervention technology, while diseases are typically studied within the biomedical context. To ensure their effectiveness, structured evaluations and descriptions of the interventions, integrating biomedical, behavioral, and technological perspectives, are essential. Objective We aimed to examine the design and evaluation of CAs for self-management support of chronic diseases, focusing on their characteristics, integration of behavioral change techniques, and evaluation methods. The findings will guide future research and inform intervention design. Methods We conducted a systematic search in the PubMed and Embase databases to identify studies that investigated CAs for chronic disease self-management, published from January 1, 2018, to April 15, 2024. Full-text journal articles, published in English, studying the efficacy or effectiveness of a CA in the context of self-management for chronic diseases in adults were included. Data extraction was guided by conceptual frameworks to ensure comprehensive reporting of intervention and methodologies: the behavioral intervention technology model and the CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile Health Applications and Online Telehealth) checklist. Risk of bias was assessed using the Risk of Bias 2 tool and the Risk of Bias in Non-randomized Studies-of Interventions (ROBINS-I) tool (version 2). Results In total, 25 studies were included, primarily focusing on text-based, rule-based CAs delivered via a mobile apps. The chronic diseases predominantly targeted were diabetes and cancer. Commonly identified clusters of behavior change techniques were “shaping knowledge,” “feedback and monitoring,” “natural consequences,” and “associations.” However, reporting of behavior change techniques and their delivery was lacking, and intervention descriptions were limited. Studies were mostly in the early phase, with a great variety in intervention descriptions, study methods, and outcome measures. Conclusions Advancing the field of CA-based interventions requires transparent intervention descriptions, rigorous methodologies, consistent use of validated scales, standardized taxonomy, and reporting aligned with standardized frameworks. Enhanced integration of artificial intelligence–driven personalization and a focus on implementation in health care settings are critical for future research.