Conversational Agents for Body Weight Management: A Systematic Review (Preprint)

BACKGROUND Obesity is a public health issue worldwide. Conversational agents (CAs), also frequently called chatbots, are computer programs which simulate dialogue between people. Due to better accessibility, cost-effectiveness, personalization, and compassionate patient-centered treatments, CAs are expected to have the potential to provide sustainable lifestyle counseling for weight management. OBJECTIVE This systematic review aimed to critically summarize and evaluate clinical studies on CAs for weight management in order to propose a future direction for effective weight management practice and research using CAs. METHODS PubMed, EMBASE, and the Cochrane library (CENTRAL) were searched up to April 18th, 2022. Studies were included if CAs were used for weight management, and they had a capability of unconstrained natural language input. No restrictions were imposed on study design, language, or publication type. Quality of the included studies was assessed using a Cochrane risk of bias assessment tool or Critical Appraisal Skills Programme (CASP) checklist. The extracted data from the included studies were tabulated and narratively synthesized as substantial heterogeneity was expected. RESULTS Eight studies met the eligibility criteria: three randomized controlled trials (RCTs) and five uncontrolled before-and-after studies. The CAs in the included studies aimed at behavioral changes through education, advice on food choices, or counseling via psychological approaches. Of the included studies, only three studies reported a significant weight-loss outcome (1.3-2.4kg decrease at 12-15 weeks of CA use). The overall quality of the included studies was judged as low. CONCLUSIONS The findings of this systematic review suggest 1) CAs with unconstrained natural language input can be utilized as a feasible interpersonal weight management intervention by promoting engagement in psychiatric intervention-based conversation simulating treatments by healthcare professionals; but 2) currently there is a paucity of evidence. Well-designed rigorous RCTs with larger sample sizes, longer treatment duration, and follow-up focusing on CAs’ acceptability, efficacy and safety are warranted.