Predicting ICU Readmission in Patients With Cerebral Infarction: A Machine Learning Approach Using Neurophysiological...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Neurology

Background:

• Intensive care unit (ICU) readmission for cerebral infarction (CI) patients poses a significant clinical challenge.
• Predictive modeling can aid in identifying high-risk patients for targeted interventions.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based predictive model for ICU readmission in CI patients.
• To utilize neurophysiological and clinical data from the MIMIC-IV database for model development.
• To compare the performance of various ML models for predicting ICU readmission.

Main Methods:

• A retrospective cohort of 3,348 CI patients from the MIMIC-IV database was analyzed.
• Feature selection was performed using LASSO regression.
• Multiple ML models (Decision Tree, KNN, LightGBM, Naïve Bayes, Random Forest, SVM, XGBoost) and logistic regression were developed and evaluated.

Main Results:

• The logistic regression model demonstrated the highest predictive performance with an AUC of 0.682 (95% CI: 0.630-0.733).
• Significant predictors for ICU readmission included peptic ulcer disease, glucocorticoid use, potassium levels, and red blood cell count.

Conclusions:

• ML models show efficacy in predicting ICU readmission for CI patients.
• Logistic regression offers a clinically interpretable approach for patient risk stratification.
• Identifying key predictors can inform clinical decision-making and resource allocation.