ESPEN 2021 Abstract Submission

Topic: Critical care

Abstract Submission Identifier: ESPEN21-ABS-1592

FEEDING INTOLERANCE AS A PREDICTOR OF CLINICAL OUTCOMES IN CRITICALLY ILL PATIENTS: A MACHINE LEARNING APPROACH

O. Raphaeli^*, ¹, C. Hajaj ¹, I. Bendavid ², A. Goldstein ¹, E. Chen ¹, P. Singer ²

¹Industrial engineering and management, Ariel University, Ariel, ²Intensive Care Unit , Rabin Medical Center, Beilinson Hospital, Petah Tikva , Israel

Rationale: Feeding intolerance (FI) in enterally fed patients is a common problem associated with adverse outcomes in critically ill patients. Yet, the exact role of FI regarding mortality and morbidity remained controversial. The aim of the study is to investigate whether FI occurrences, along 72 hours of admission, has incremental prognostic significance in predicting clinical outcomes in critically ill patients.

Methods: We included adult patients (2012-2018) admitted at Beilinson hospital ICU for more than 48 hours. FI definition is based on the occurrence of “large” gastric volumes, GI symptoms and “inadequate” delivery of enteral nutrition (Reintam Blaser et al., 2021). Admission conditions and FI occurrences, along 72 hours, were analyzed by machine learning (ML) classification algorithms predicting several mortality and morbidity outcomes metrics. Prediction performance was assessed by the area under the curve (AUROC) of ten-fold cross-validation and validation sets. The study was approved by the ethics committee of Beilinson.

Results: The dataset comprised of 1,782 patients who received enteral nutrition (EN). The median (IQR) age was 62 (48-72) years, BMI 26.5 (23-31). Main admission conditions: surgical (47%), trauma (27%) and medical (25%). Five ML algorithms were trained and tested (Python software). The best performing algorithm was Random Forest classifier. Models included admission conditions only achieved AUC of 73% - 77% (according to outcome metric), while the addition of FI occurrences along 72 hours achieved AUC of 82% - 87%, respectively. Valuable predictors were mainly large GRV (>250 mL) and inadequate delivery of enteral nutrition.

Conclusion: FI occurrences along 72 hours of ICU admission has an incremental prognostic significance in an ML approach predicting adverse clinical outcomes in enterally fed critically ill patients

References: Reintam Blaser A. , Deane AM, Preiser JC, Arabi YM, Jakob SM. Enteral Feeding Intolerance: Updates in Definitions and Pathophysiology. Nutr Clin Pract. 2021 Feb;36(1):40-49

Disclosure of Interest: None Declared

Keywords: Enteral Feeding Intolerance, enteral nutritional support, intensive care unit, Machine learning