Implementation determinants and outcomes of artificial intelligence in surgical healthcare: a systematic review...

Artificial intelligence (AI) is increasingly being integrated into surgical healthcare to enhance perioperative risk assessment, improve diagnostic accuracy, refine clinical workflows, support intraoperative decision-making and reduce postoperative complications. AI applications including machine learning-based predictive analytics, computer vision, robotic assistance and clinical decision support systems have shown promising clinical performance across surgical specialties. However, knowledge on real-world implementation, barriers, facilitators and system-level outcomes in surgical healthcare remains fragmented with limited synthesis of implementation determinants. This systematic review aims to identify, categorise and synthesise global evidence on implementation strategies, barriers, facilitators, adoption outcomes and reported impacts of AI applications in surgical healthcare settings.

This protocol is reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 guidelines and has been registered with PROSPERO (CRD420261305972). A systematic search will be conducted across PubMed/Medline, CINAHL (EBSCO), Web of Science, Scopus, Global Index Medicus, medRxiv, IEEE Xplore and ACM Digital Library for studies published between January 2000 and December 2025. Study designs include interventional (randomised controlled trials), observational, implementation, qualitative and mixed-methods studies and programme evaluations. The primary outcome is barriers and facilitators to AI implementation, and synthesised using framework synthesis (Carroll et al). Secondary outcomes include implementation outcomes (eg, adoption, feasibility, acceptability, fidelity, penetration, and sustainability), implementation strategies employed, reported effects on surgical quality metrics, workflow efficiency, and cost and resource utilisation. Secondary implementation outcomes will be characterised using the Reach, Effectiveness, Adoption, Implementation, Maintenance framework. Where three or more studies report comparable quantitative outcome data with effect estimates, random-effects meta-analysis will be conducted; no meta-analysis will be performed for the primary outcome. Risk of bias will be assessed using the Mixed Methods Appraisal Tool. Certainty of evidence will be evaluated using Grading of Recommendations Assessment, Development and Evaluation (GRADE) for quantitative findings and GRADE-CERQual for qualitative evidence. Subgroup analyses by geographic setting, surgical specialty and AI modality are planned, conditional on data availability (minimum three studies per subgroup).

Ethical approval is not needed, and the results will be disseminated via peer-reviewed publication and presentation at conferences relevant to this field.