Artificial Intelligence in Antimicrobial Stewardship: An Umbrella Review of Systematic Reviews
1
Systematic Reviews Network (SRN), Calabar, Nigeria
Popul. Med. 2026;8(Supplement Supplement 1):A218
ABSTRACT
INTRODUCTION:
Artificial intelligence (AI) has been increasingly proposed as a tool to strengthen antimicrobial stewardship (AMS) and address the growing threat of antimicrobial resistance (AMR). While numerous systematic reviews have evaluated AI applications in AMS, their findings are heterogeneous, and the overall strength of the evidence base remain unclear. This umbrella review aimed to summarize, compare, and critically appraise the evidence from systematic reviews on AI applications in AMS.
METHODS:
An umbrella review was conducted of systematic reviews published between 2022 and 2025. Searches were performed in PubMed and Google Scholar. Eligible reviews synthesized evidence on AI or machine learning applications in antimicrobial stewardship among human populations. Data extracted at the review level, included scope, AI methods, AMS domains, outcomes, and limitations.
RESULTS:
Six systematic reviews, encompassing over 150 primary studies and more than 1.8 million patients, were included. Supervised machine learning models, particularly random forests, gradient-boosting algorithms, support vector machines, and logistic regression, dominated the evidence base. Large language models (LLMs) were assessed in only one review. AI applications focused primarily on prediction of antimicrobial resistance, early infection detection, risk stratification, and empiric antibiotic optimization, with limited evaluation of prescribing behavior, antibiotic duration, or long-term stewardship outcomes. Quantitative meta-analyses reported moderate-to-good predictive performance (pooled AUC 0.72; accuracy 75%; sensitivity 77%; specificity 74%), with higher negative predictive values indicating potential utility for rule-out tasks. Evidence linking AI use to downstream clinical or stewardship outcomes was limited, heterogeneous, and largely restricted to reductions in inappropriate antibiotic prescribing in small observational studies. LLMs demonstrated moderate performance on simple questions but poor reliability for complex clinical decision-making.
CONCLUSIONS:
AI shows promise in supporting predictive and decision-support functions within AMS, but are not yet suitable for autonomous antimicrobial decision-making. Current evidence is limited by geographic concentration, and scarce data on real-world clinical outcomes.
Artificial intelligence (AI) has been increasingly proposed as a tool to strengthen antimicrobial stewardship (AMS) and address the growing threat of antimicrobial resistance (AMR). While numerous systematic reviews have evaluated AI applications in AMS, their findings are heterogeneous, and the overall strength of the evidence base remain unclear. This umbrella review aimed to summarize, compare, and critically appraise the evidence from systematic reviews on AI applications in AMS.
METHODS:
An umbrella review was conducted of systematic reviews published between 2022 and 2025. Searches were performed in PubMed and Google Scholar. Eligible reviews synthesized evidence on AI or machine learning applications in antimicrobial stewardship among human populations. Data extracted at the review level, included scope, AI methods, AMS domains, outcomes, and limitations.
RESULTS:
Six systematic reviews, encompassing over 150 primary studies and more than 1.8 million patients, were included. Supervised machine learning models, particularly random forests, gradient-boosting algorithms, support vector machines, and logistic regression, dominated the evidence base. Large language models (LLMs) were assessed in only one review. AI applications focused primarily on prediction of antimicrobial resistance, early infection detection, risk stratification, and empiric antibiotic optimization, with limited evaluation of prescribing behavior, antibiotic duration, or long-term stewardship outcomes. Quantitative meta-analyses reported moderate-to-good predictive performance (pooled AUC 0.72; accuracy 75%; sensitivity 77%; specificity 74%), with higher negative predictive values indicating potential utility for rule-out tasks. Evidence linking AI use to downstream clinical or stewardship outcomes was limited, heterogeneous, and largely restricted to reductions in inappropriate antibiotic prescribing in small observational studies. LLMs demonstrated moderate performance on simple questions but poor reliability for complex clinical decision-making.
CONCLUSIONS:
AI shows promise in supporting predictive and decision-support functions within AMS, but are not yet suitable for autonomous antimicrobial decision-making. Current evidence is limited by geographic concentration, and scarce data on real-world clinical outcomes.
| eISSN: | 2654-1459 |
We process personal data collected when visiting the website. The function of obtaining information about users and their behavior is carried out by voluntarily entered information in forms and saving cookies in end devices. Data, including cookies, are used to provide services, improve the user experience and to analyze the traffic in accordance with the Privacy policy. Data are also collected and processed by Google Analytics tool (more).
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
