Artificial intelligence for infection surveillance, risk stratification, and antimicrobial decision support in acute-care hospitals: a scoping review.

Background: Artificial intelligence (AI) has increasingly been proposed to strengthen infection surveillance, early risk stratification, antimicrobial decision support, and selected workflow functions in acute-care hospitals. However, the literature remains clinically heterogeneous, methodologically uneven, and conceptually fragmented, with technical performance often interprested too readily as evidence of clinical effectiveness. This scoping review aimed to map and synthesise the empirical literature on AI applications for infection surveillance and related hospital applications, while explicitly distinguishing technical performance from clinical utility, implementation relevance, and patient benefit. Methods: We conducted a scoping review in accordance with PRISMA 2020 and PRISMA-ScR guidance. CINAHL, Cochrane Library, Embase, PubMed, Scopus, and Web of Science were searched for English-language empirical studies. Eligible studies examined AI applications relevant to infection surveillance, risk prediction, detection, antimicrobial decision support, or workflow-relevant hospital functions in acute-care settings. Findings were synthesised narratively by application domain and translational stage. Results: Database searches yielded 884 records; 628 unique records underwent title and abstract screening, 180 full texts were assessed, and 39 studies were included. The literature was dominated by retrospective model-development and validation studies; no randomised trials or robust comparative evaluations under routine clinical conditions were identified. Evidence clustered around surgical-site infection, urinary-tract-infection-related outcomes, ventilator-associated pneumonia, bacteraemia, sepsis, resistant organisms, and infection-related mortality. Across these domains, AI models generally showed moderate-to-high discriminatory performance, particularly for surgical-site infection surveillance and prediction. A smaller body of evidence suggested potential operational value in antimicrobial prescribing support, early risk stratification, real-time bacteraemia prediction, and reduction of manual surveillance workload. However, implementation evidence was sparse and heterogeneous, with limited assessment of usability, adoption, trust, workflow redesign, sustained real-world performance, or patient-level benefit. Conclusion: AI shows substantial promise as an adjunct to infection surveillance and selected hospital infection-management tasks, but the current evidence base is considerably stronger for technical accuracy than for clinical effectiveness, implementation success, or patient benefit. Stronger prospective, externally validated, and implementation-oriented studies are needed before firmer claims can be justified.