Performance Evaluation of a Supervised Machine Learning Pain Classification Model Developed by Neonatal Nurses.

Area of Science:

• Neonatal care
• Machine learning in healthcare
• Pain assessment technologies

Background:

• Early-life pain can negatively impact neurodevelopment.
• Current pain assessment methods are inconsistent and nurse-dependent.
• Facial expression tools lack brain-based pain evidence.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for pain classification.
• To improve the objectivity and consistency of pain assessment in neonates.
• To create a brain-based pain recognition system.

Main Methods:

• Utilized the Neonatal Facial Coding System (NFCS) with video data from 49 neonates.
• Employed a human-centered design for Embedded Machine Learning Solutions.
• Trained and tested ML models using a 70% training and 30% testing data split.

Main Results:

• The best ML model achieved 98.5% recall and 0.98 Area Under the Curve (AUC).
• ML model performance significantly surpassed NICU nurses' AUC of 0.68.
• Nurse interrater reliability varied across different pain assessment tasks.

Conclusions:

• The ML model demonstrates superior accuracy in identifying neonatal pain compared to human assessment.
• Findings support the development of an automated, continuous, brain-based pain monitoring system (PRAMS).
• This technology has the potential to enhance neonatal pain management and outcomes.