JMIR Publications

ABSTRACT

Background:

Prescription errors are a relevant challenge in critically ill patients under polypharmacy. Considering this, a growing presence of artificial intelligence based softwares to avoid these situations is noticeable.

Objective:

This study evaluated the role of NoHarm.ai, an AI-based clinical decision support tool, in detecting and preventing dosage discrepancies in an intensive care unit.

Methods:

Two-stage study: exploratory phase (September–December 2023), with a theoretical review on AI and clinical safety; and a prospective observational phase (March–July 2024), which analyzed “overdose” or “underdose” notifications provided by the AI integrated into the electronic health records of a 22-bed general ICU at a tertiary hospital in southern Brazil.

Results:

A total of 73 notifications were registered in prescriptions for 32 patients (59.4% male; 40.6% female). Most interventions (85%) were evaluated within 24 hours, with overdose alerts predominating, especially vancomycin (65.6%). Among accepted interventions, there were 13 dose increases, 5 reductions, and 7 suspensions. Patients over 65 years old had higher mortality (60%), and overall outcomes were 18 discharges and 14 deaths. There was no statistically significant difference in renal or hepatic function between cases with or without dosage adjustment, possibly due to the sample size.

Conclusions:

The integration of AI into clinical pharmacy practice demonstrated relevant acceptance and effectiveness in detecting dosage inconsistencies, highlighting the potential of hybrid human–machine models to enhance medication safety. Larger multicenter studies are needed to validate clinical applicability and standardize efficacy and safety metrics.