JMIR Publications

ABSTRACT

Background:

Artificial Intelligence (AI) offers a novel approach to enhance antimicrobial stewardship (AMS). Existing systematic reviews focus on AI’s predictive performance, overlooking the implementation gap between AI’s potential and its use.

Objective:

This systematic review aimed to identify and synthesise peer-reviewed literature on the factors influencing AI use and acceptability in AMS.

Methods:

Eight databases were searched from inception to September 2025 to identify perceived barriers and enablers to AI use in AMS. Data on factors influencing AI use and its acceptability were deductively coded into domains from the Theoretical Domains Framework (TDF) and The Technology Acceptance Model (TAM3) respectively. Inductive thematic analysis within domains provided deeper insights into the specific influences on AI use. The Theory and Techniques Tool (TaTT) and Affordability, Practicability, Effectiveness, Acceptability, Side-effects, Safety (APEASE) criteria were used to identify potential Behaviour Change Techniques (BCTs) to address these influences.

Results:

Twelve primary studies were included. Barriers/enablers reported were mostly from HCPs (92%), across healthcare settings and roles. Key factors influencing AI use in AMS fell within five domains: Beliefs about Consequences (100% of included studies; e.g. impact on patient safety), Environmental Context & Resources (92%; e.g. usability), Memory, Attention & Decision Processes (75%; e.g. guidance under uncertainty), Knowledge (75%; e.g. understanding of system functionality), and Social/Professional Role and Identity (67%; e.g. professional autonomy and expertise). Domain mapping to the TAM3 identified Reasonable Demonstrability, Job Relevance, and Output Quality as key determinants to Perceived Usefulness, and Perception of External Control and Computer Self-Efficacy as key determinants of Perceived Ease of Use. Potential BCTs to address these influences include: Feedback on behaviour and outcomes of behaviour, Information about health consequences, and Adding objects to the environment.

Conclusions:

Key influences on AI use in AMS were identified across five TDF domains, alongside five key determinants utilising the TAM3 impacting its perceived usefulness and perceived ease of use, highlighting targets for intervention. Further primary research is needed on factors influencing AI use in AMS, particularly in agriculture and veterinary care, and on the effectiveness of various intervention strategies.