Beyond Human Error: Building Intelligent Resilience for Medication Safety in the ICU.

Area of Science:

• Patient Safety
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Medication errors (MEs) in intensive care units (ICUs) pose a significant risk to patient safety.
• Traditional voluntary reporting systems detect less than 2% of MEs, leaving a substantial surveillance gap.
• Existing methods for ME detection are insufficient for ensuring patient safety in ICUs.

Purpose of the Study:

• To critically review the literature on artificial intelligence (AI) and implementation science for improving medication safety in ICUs.
• To propose a five-layer Intelligent Safety Stack to bridge the surveillance gap in medication error detection.
• To explore how AI and sociotechnical frameworks can enhance patient safety in critical care settings.

Main Methods:

• A critical narrative review of peer-reviewed literature from 2000-2025.
• Synthesis of evidence on medication safety, predictive analytics, generative AI, engineering controls, and sociotechnical frameworks.
• Development of a proposed five-layer Intelligent Safety Stack framework.

Main Results:

• Reported ME incidence varies widely (1.32%–31.7%) due to methodological differences.
• The proposed Intelligent Safety Stack includes Standardised Ontology, Intelligent Surveillance, Signal Optimisation, Generative Stewardship, and Engineering Controls.
• Engineering Controls, such as smart pump interoperability, demonstrated up to a 54.8% reduction in administration errors.

Conclusions:

• Relying solely on error counting is inadequate for sustainable medication safety.
• A sociotechnical integration of the Intelligent Safety Stack is crucial for improving ICU patient safety.
• Success in medication safety should be measured by rescue rates, not just error prevalence.