An interpretable machine learning approach for predicting clinically important gastrointestinal bleeding in...

Area of Science:

• Critical Care Medicine
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Clinically important gastrointestinal bleeding (CIGIB) is a significant complication in critically ill patients, associated with increased ICU stay and mortality.
• Existing methods for identifying high-risk patients for CIGIB are insufficient.
• No prior studies have utilized machine learning to predict CIGIB in the ICU or identify its key predictors.

Purpose of the Study:

• To develop and validate machine learning models for predicting CIGIB in intensive care unit (ICU) patients.
• To identify key clinical predictors of CIGIB using machine learning interpretability techniques.

Main Methods:

• A retrospective single-center study included 7357 ICU patients admitted between 2017 and 2024.
• Machine learning models (XGBoost, Random Forest, L1-logistic regression) were trained on data from the first 24 hours of ICU admission.
• Model performance was evaluated using AUROC, precision, recall, and F1 scores, with SHAP values used for predictor evaluation.

Main Results:

• The XGBoost model achieved the highest predictive performance with an AUROC of 0.84.
• Key predictors of CIGIB included APACHE III scores, hematocrit, APTT, creatinine, and respiratory rate.
• Invasive mechanical ventilation and stress ulcer prophylaxis were not among the top 20 predictors.

Conclusions:

• This study is the first to apply machine learning for CIGIB prediction in ICU patients, demonstrating high accuracy and interpretability.
• The developed model shows potential for guiding early interventions and prophylaxis strategies.
• Further multi-center studies and interventional trials are recommended to validate these findings.