Effect of Artificial Intelligence on Medication Safety and Prescribing Errors in Emergency Care: A Systematic Review

Abstract

Background: Medication-related harm remains a major patient-safety problem in emergency care, where time pressure, incomplete clinical information, and complex prescribing decisions increase the risk of error. Artificial intelligence is increasingly used to support safer prescribing, but its effect in emergency-care medication management remains uncertain.
Methods: A systematic review of PubMed-indexed primary studies was conducted. Clinical trials and cohort studies evaluating artificial intelligence interventions for medication safety or prescribing errors in emergency or acute care were included. Data were extracted systematically, risk of bias was assessed, and findings were synthesized narratively without meta-analysis.
Results: Nine studies were included, comprising two randomized trials and seven cohort studies. Artificial intelligence–supported systems generally improved prescribing safety: severe drug-related problems decreased from 15.3% to 2.8% in one randomized study, and sedation-type order errors decreased from 0.39% to 0.037% in another (odds ratio 0.094). Predictive performance was favorable across studies, with area under the receiver operating characteristic curve values ranging from 0.81 to 0.91, sensitivity reaching 0.99, and one alert-filtering model suppressing 54.1% of alerts while maintaining 0.99 sensitivity. Additional benefits included 43% alert-driven prescription changes and 87.1% detection of dose errors.
Conclusions: Artificial intelligence appeared to improve medication safety and reduce prescribing errors in emergency-related care, particularly when integrated as clinician-supportive decision support rather than as a replacement for clinical judgment. However, the evidence remained heterogeneous and was dominated by retrospective designs, indicating the need for larger prospective emergency-specific studies before wider implementation.