Machine learning-based risk prediction model for medication administration errors in neonatal intensive care units: A...

Area of Science:

• Medical Informatics
• Neonatal Medicine
• Machine Learning in Healthcare

Background:

• Neonates are highly vulnerable to medication administration errors (MAEs) due to physiological immaturity and complex dosing.
• MAEs in neonates can lead to severe harm and significant healthcare costs.
• Developing predictive models for MAEs is critical for patient safety and healthcare efficiency.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting MAEs in neonatal intensive care units (NICUs).
• To identify key risk factors associated with MAEs in neonates.
• To provide a tool for timely interventions to reduce MAE occurrence.

Main Methods:

• A prospective, direct observational study in five Malaysian NICUs.
• Direct observation of nurses during medication preparation and administration for 1093 doses.
• Application of ten ML algorithms, with AdaBoost selected as the best performer based on F1-score.

Main Results:

• The AdaBoost model achieved an F1-score of 83.28%, accuracy of 77.63%, and an area under the ROC curve of 82.95%.
• Key predictors for MAEs included the intravenous route of administration, nurse working hours, and nursing experience.
• The study successfully identified influential features for MAE prediction in neonates.

Conclusions:

• An ML-based model, specifically AdaBoost, was successfully developed and validated for predicting MAEs in neonates.
• The model can assist healthcare providers in identifying high-risk situations and implementing timely interventions.
• This predictive approach holds potential for reducing harm and improving safety in neonatal medication administration.