Neonatal Adverse Events' Trigger Tool Setup With Random Forest

Area of Science:

• Neonatal Medicine
• Medical Informatics
• Clinical Data Analysis

Background:

• Neonatal adverse events (AEs) pose significant risks to hospitalized infants.
• Early detection of AEs is crucial for timely intervention and improved outcomes.
• Existing methods for AE detection may lack efficiency and reliability.

Purpose of the Study:

• To develop and validate a novel trigger tool for the detection of neonatal adverse events (AEs).
• To identify critical variables predictive of AEs in neonates using machine learning.
• To assess the effectiveness of the developed trigger tool in a real-world clinical setting.

Main Methods:

• Utilized a random forest (RF) algorithm to build a predictive model from the medical records of 782 neonates.
• Selected critical variables through recursive elimination to form the trigger tool.
• Validated the trigger tool by reviewing medical records of 655 neonates with birth weights ≥1500 g.

Main Results:

• The RF model identified six key variables: diarrhea, antibiotic use, fever, death, skin damage, and suspected necrotizing enterocolitis.
• The trigger tool demonstrated a sensitivity of 70.7% and specificity of 92.0% in predicting AEs.
• The overall positive predictive value of the trigger tool was 0.686, with skin damage, iatrogenic diarrhea, and fever being the most common AEs.

Conclusions:

• The developed neonate-focused trigger tool, based on the RF algorithm, is efficient and reliable for identifying AEs.
• This tool can aid in the early detection of adverse events in hospitalized neonates weighing ≥1500 g.
• The findings support the integration of this trigger tool into clinical practice for enhanced neonatal care.