Conversational agents in physical and psychological symptom management: A systematic review of randomized controlled trials

Background

Effective management of physical and psychological symptoms is a critical component of comprehensive care for both chronic disease patients and apparently healthy individuals experiencing episodic symptoms. Conversational agents, which are dialog systems capable of understanding and generating human language, have emerged as a potential tool to enhance symptom management through interactive support.

Objective

To examine the characteristics and effectiveness of conversational agent-delivered interventions reported in randomized controlled trials (RCTs) in the management of both physical and psychological symptoms.

Design

A systematic review.

Methods

A comprehensive search was performed in Pubmed, ACM Digital Library, CINAHL, EMBASE, PyscInfo, Web of Science, Scopus and gray literature sources from their inception to Oct 2024. Search terms included “conversational agent”, “symptom”, “randomized controlled trial” and their synonyms and hyponyms. Duplicates were identified by EndNote, and titles, abstracts and full texts were independently screened according to predefined criteria. Data extraction focused on basic study characteristics and conversational agent details, with The Cochrane Risk of Bias 2.0 tool employed for bias assessment.

Results

The search yielded 2756 articles and 29 were finally included for review. The included studies predominantly came from developed countries (n = 23) and were conducted between 2020 and 2024 (n = 24). The studies frequently evaluated the feasibility and acceptability of conversational agent interventions (n = 14), with a predominantly focus on psychological symptoms (depression, anxiety, etc.) (n = 17). A few studies focused on physical symptoms (pain, etc.) (n = 4), while others addressed both symptoms (n = 8). Twenty-five distinct conversational agents (Woebot, Tess, etc.) were evaluated, utilizing platforms ranging from proprietary applications to common messaging channels like WeChat and Facebook Messenger. Cognitive Behavioral Therapy (CBT) was a commonly integrated approach (n = 22), with rule-based dialogs (n = 22) as the most commonly dialog system methods and Natural Language Processing (NLP) (n = 15) as the predominant AI techniques. The median recruitment and completion rates were 72 % and 79 %, respectively. The majority of studies reported positive user experiences and significant symptom management improvements (n = 22). However, risk of bias was high in seventeen studies and presented some concerns in nine others.

Conclusions

Conversational agents have shown promise in enhancing both physical and psychological symptom management through positive user experiences and effectiveness. However, the high risk of bias identified in many studies warrants caution in interpreting these findings. Future research should prioritize the methodological quality of RCTs to strengthen the evidence base supporting the use of conversational agents as a complementary tool in symptom management.