Deep-Learning Model for Mortality Prediction of ICU Patients with Paralytic Ileus.

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Biomedical Informatics

Background:

• Paralytic Ileus (PI) in Intensive Care Units (ICUs) is associated with high mortality rates.
• Existing predictive models for PI mortality are often complex and lack reliability.
• Predicting mortality in ICU patients with PI is challenging due to extensive data and numerous variables.

Purpose of the Study:

• To develop and validate a deep-learning framework for predicting mortality in ICU patients diagnosed with PI.
• To identify key clinical and demographic factors influencing mortality risk in PI patients.
• To enhance the accuracy and reliability of mortality prediction for PI patients in critical care settings.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care IV (MIMIC-IV) dataset, including 1017 ICU patients with PI.
• Employed SHAP (SHapley Additive exPlanations) analysis to identify significant predictive features.
• Developed a deep neural network (DLMP) using six clinical lab items (Anion gap, Platelet, PTT, BUN, Total Bilirubin, Bicarbonate) and one demographic variable.

Main Results:

• The DLMP framework achieved a high Area Under the Curve (AUC) score of 0.887.
• DLMP demonstrated superior performance compared to existing predictive models for ICU patients with PI.
• The model effectively reduced feature complexity by focusing on six critical lab items and one demographic variable.

Conclusions:

• The developed Deep Learning Model for Mortality Prediction (DLMP) offers a significant advancement in predicting mortality for ICU patients with PI.
• DLMP provides a more accurate and reliable prognostic tool, aiding families and clinicians in understanding patient conditions.
• The model shows potential for improving clinical trial design and facilitating further research in critical care prognostication.