Artificial intelligence chatbot-assisted preoperative education and perioperative anxiety in surgical patients: A systematic review and structured narrative synthesis of randomized controlled trials.

BACKGROUND

Preoperative anxiety may adversely affect surgical patients' experiences of anesthesia, pain, recovery, and care. Artificial intelligence (AI) chatbots are increasingly used for preoperative education, but their effects on perioperative anxiety remain uncertain. This review evaluated the effects of AI chatbot-assisted preoperative education on perioperative anxiety in surgical patients.

METHODS

This review was registered in PROSPERO on 16 June 2026 (CRD420261425412) and followed PRISMA 2020 and SWiM. PubMed, Web of Science, Scopus, Embase, Cochrane CENTRAL, Europe PMC, ClinicalTrials.gov, and WHO ICTRP were searched for eligible randomized trials. Anxiety outcomes were classified by construct, validation, score type, and timing; where STAI state- and trait-anxiety scores were reported separately, only the state-anxiety score was analyzed. Study-specific Hedges-corrected SMDs and 95% CIs were displayed without pooling because heterogeneity and directional inconsistency made a summary estimate uninterpretable. Risk of bias and evidence certainty were assessed using RoB 2 and GRADE, respectively.

RESULTS

Five randomized controlled trials involving 549 participants were included. The study-level effects varied substantially in both direction and magnitude, ranging from an SMD of - 0.59 (95% CI - 0.82 to - 0.36), favoring AI chatbot-assisted education, to an SMD of 1.25 (95% CI 0.73 to 1.77), favoring conventional physician-led education. The remaining estimates included effects favoring AI-assisted education as well as estimates compatible with benefit or harm. Differences across surgical settings, chatbot architectures, comparator conditions, anxiety measures, and human oversight precluded a clinically meaningful pooled estimate. The certainty of evidence was very low because of risk of bias, inconsistency, and imprecision.

CONCLUSIONS

Evidence from five heterogeneous randomized controlled trials remains insufficient to determine whether AI chatbot-assisted preoperative education reduces perioperative anxiety. The absence of a consistent effect should not be interpreted as definitive evidence of ineffectiveness, particularly given the very low certainty of evidence. Larger, well-designed trials are required.