Using Machine-Learning to Assess the Prognostic Value of Early Enteral Feeding Intolerance in Critically Ill...

Area of Science:

• Critical Care Medicine
• Clinical Nutrition
• Machine Learning in Healthcare

Background:

• Gastrointestinal intolerance during early enteral nutrition (EN) in critically ill patients has a controversial association with adverse outcomes.
• Early identification of patients at risk for enteral feeding intolerance (EFI) is crucial for improving clinical outcomes.

Purpose of the Study:

• To assess the prognostic value of EFI markers during early intensive care unit (ICU) stays.
• To predict early EN failure using a machine learning (ML) approach.

Main Methods:

• Retrospective analysis of 1584 adult patients receiving EN for >48 hours in an ICU.
• Utilized ML algorithms to analyze clinical data, EFI markers, and medications within 72 hours of admission.
• Assessed prediction performance using area under the receiver operating characteristics (AUCROC) via ten-fold cross-validation.

Main Results:

• ML models achieved AUCROCs of 0.73 for 90-day mortality and 0.71 for early EN failure.
• Gastric residual volume >250 mL on day two was a significant predictor in both models.
• Identified specific EFI markers with prognostic value for critical care patients.

Conclusions:

• ML effectively identified EFI markers associated with poor 90-day outcomes and early EN failure.
• Supports the early recognition and management of critically ill patients at risk for feeding intolerance.
• Further prospective and external validation studies are recommended to confirm findings.