Predicting Discharge Dates From the NICU Using Progress Note Data.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Neonatal Care

Background:

• NICU discharge delays often stem from nonmedical factors like equipment needs and parental education.
• Predictive modeling can identify medically ready patients, enabling proactive planning for discharge.

Purpose of the Study:

• To develop and validate a machine learning model to predict the days to discharge (DTD) for NICU patients.
• To identify key clinical features that indicate a patient's readiness for discharge.

Main Methods:

• A retrospective study analyzed 26 features from daily progress notes of 4693 NICU patients.
• Supervised machine learning, specifically random forest classifiers, was employed to predict DTD.
• Model performance was evaluated across different patient subpopulations.

Main Results:

• The model demonstrated strong predictive accuracy (AUC 0.723-0.865) for DTD across most patient groups.
• Neurosurgery patients showed lower model performance compared to premature, cardiac, and GI surgery groups.
• Key predictors for discharge readiness included oral feeding percentage and weight.

Conclusions:

• Clinical data from daily progress notes can accurately predict NICU patient discharge readiness.
• This predictive capability facilitates timely discharge planning and resource allocation.