Artificial Intelligence in Medication Safety Systems for the Prevention of Perioperative and Inpatient Medication Errors

Abstract

Background:
Medication errors are a major source of preventable harm in perioperative and inpatient care, and artificial intelligence (AI) is increasingly used to strengthen medication safety systems.
Methods:
PubMed was searched from inception to September 2025 for English-language human studies evaluating AI-enabled medication safety interventions in hospital perioperative or inpatient settings. Two reviewers screened records, extracted data in duplicate, and appraised risk of bias using Joanna Briggs Institute checklists; results were synthesized narratively without meta-analysis.
Results:
Eleven studies, predominantly retrospective cohorts or pre–post implementations, were included across adult wards, intensive care, pediatric/neonatal intensive care, and intravenous administration monitoring, with sample sizes ranging from 311 ICU prescribing episodes to 3,481,634 alert events. Discrimination for identifying high-risk prescriptions or patients was moderate to high (AUROC 0.74–0.97). Improvements were reported in key process measures, including reduced pediatric ICU dosing deviations (RR 0.21; 95% CI 0.05–0.96), higher reconciliation discrepancy yield with admission prioritization (45% vs 21%; RR 2.13; 95% CI 1.40–3.24), and lower alert burden (54.1% reduction at sensitivity 0.99; precision 0.192); one deployment triggered alerts for 0.4% of orders and 43% prompted order changes.
Conclusions:
AI-enabled medication safety systems were associated with improved identification and management of high-risk medication scenarios and more efficient safety surveillance. Evidence remained heterogeneous, and preventable adverse drug events and longer-term patient outcomes were infrequently measured.