Conversational agents in healthcare: a systematic review

Area of Science:

• Digital Health
• Human-Computer Interaction
• Health Informatics

Background:

• Conversational agents (CAs) with natural language capabilities are increasingly explored for health applications.
• Understanding their characteristics, applications, and evaluation is crucial for advancing digital health tools.

Purpose of the Study:

• To systematically review the characteristics, applications, and evaluation measures of CAs with unconstrained natural language input for health-related purposes.
• To identify research gaps and inform future study designs in this emerging field.

Main Methods:

• A comprehensive search of multiple databases (PubMed, Embase, CINAHL, PsycInfo, ACM Digital) was conducted.
• Studies involving CAs with unconstrained natural language input and user interaction evaluation measures were included.
• Screening was performed by independent reviewers, with inter-coder agreement assessed using Cohen's kappa.

Main Results:

• 17 articles on 14 CAs met the inclusion criteria, with most using finite-state or frame-based dialogue management.
• Half of the CAs supported consumer self-care tasks; one randomized controlled trial (RCT) showed a significant reduction in depression symptoms (d=0.44, p=0.04).
• Patient safety was infrequently evaluated across the reviewed studies.

Conclusions:

• The use of CAs for health is an emerging research area with limited, predominantly quasi-experimental studies.
• Current research rarely evaluates the efficacy or safety of these agents.
• Future research should prioritize robust experimental designs and standardized reporting for CAs in healthcare.