Accepted for/Published in: JMIR Medical Informatics
Date Submitted: May 21, 2025
Date Accepted: Apr 21, 2026
Development and Interpretability Analysis of a Stacking Ensemble Model for Early Prediction of Nutritional Risk in ICU Patients: Retrospective Cohort Study
ABSTRACT
Background:
Malnutrition in critically ill patients significantly impacts clinical outcomes.
Objective:
This study aims to develop an AI-driven model for real-time malnutrition risk monitoring in ICU settings, addressing the limitations of delayed manual screening.
Methods:
We analyze clinical records of 49,679 ICU patients from the MIMIC-IV database. A voting ensemble model (LGC-VE) was designed to process irregular time-series data, supported by dynamic feature standardization and interpretability analysis (SHAP).
Results:
The LGC-VE model achieves an accuracy of 92.87% and AUC of 0.96, outperforming single models (LSTM: 91.77%, CNN: 90.76%) and conventional tools. SHAP analysis identifies serum chloride concentration (mean SHAP value=0.42), serum bicarbonate (0.38), and BMI (0.35) as top predictors.
Conclusions:
This study proposes an interpretable AI model for ICU malnutrition monitoring, prioritizing timely interventions through efficient computational performance. Future work will validate its scalability across diverse clinical environments.
Citation
The author of this paper has made a PDF available, but requires the user to login, or create an account.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.