Development and validation of a machine learning-based detection system to improve precision screening for medication...

Area of Science:

• Medical Informatics
• Neonatal Medicine
• Machine Learning in Healthcare

Background:

• Medication errors (MEs) are a significant concern in Neonatal Intensive Care Units (NICUs).
• Predicting and preventing MEs requires understanding complex contributing factors.
• Existing models for ME prediction in neonates are limited.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting medication errors in NICU patients.
• To identify key patient-related and care provider-related parameters associated with MEs.
• To provide a tool for targeted screening and prevention of MEs in neonates.

Main Methods:

• Prospective, observational cohort study in a 22-bed NICU.
• Analysis of 11,908 medication orders for 412 NICU patients.
• Development and validation of ML algorithms to predict ME presence.

Main Results:

• Physician-related and nurse-related MEs were observed in 42.2% and 57.0% of patients, respectively.
• Key predictors included total drugs, drug classes (anti-infectives, nervous system, etc.), Apgar score, postnatal age, nurse/physician work hours, and shifts.
• The predictive model demonstrated high performance with an AUC of 0.920.

Conclusions:

• This is the first validated ML model to predict MEs in NICU patients using work environment and pharmacotherapy data.
• The model shows promise for targeted, precision screening to prevent medication errors in neonates.
• The developed model is accessible online for potential clinical implementation.