Related Experiment Video
Updated: Jul 7, 2025
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.6K
Predicting ICU readmission risks in intracerebral hemorrhage patients: Insights from machine learning models using
Jinfeng Miao1, Chengchao Zuo1, Huan Cao1
1Department of Neurology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, No.1095 Jiefang Avenue, Wuhan 430030, China.
Journal of the Neurological Sciences
|December 26, 2023
Summary
Machine learning models can predict intensive care unit (ICU) readmissions in patients with intracerebral hemorrhage (ICH). Key predictors include hydrocephalus, sex, neutrophils, Glasgow Coma Scale (GCS), oxygen saturation (SpO2), and creatinine levels.
Area of Science:
- Neurology
- Medical Informatics
- Critical Care Medicine
Background:
- Intracerebral hemorrhage (ICH) is a severe stroke subtype with high mortality and frequent complications.
- While acute ICH management is well-studied, factors leading to intensive care unit (ICU) readmission are less understood.
- This study addresses the gap in knowledge regarding ICU readmission predictors for ICH patients.
Purpose of the Study:
- To develop and validate machine learning (ML) models for predicting ICU readmissions in patients with intracerebral hemorrhage (ICH).
- To identify key clinical factors that significantly influence the likelihood of ICU readmission in ICH survivors.
Main Methods:
- Utilized retrospective data from 2242 ICH patients across the MIMIC-III and MIMIC-IV databases.
- Employed recursive feature elimination with cross-validation (RFECV) to identify significant predictors.
- Developed and validated four ML models: AdaBoost, RandomForest, LightGBM, and XGBoost, using SHapley Additive exPlanations (SHAP) for feature interpretation.
Main Results:
- ICU readmission rates were 9.6% (MIMIC-III) and 10.6% (MIMIC-IV).
- The LightGBM model achieved the highest performance with an AUC of 0.736.
- Significant predictors identified by SHAP analysis included hydrocephalus, sex, neutrophils, Glasgow Coma Scale (GCS), oxygen saturation (SpO2), and creatinine.
Conclusions:
- The LightGBM model shows promise for predicting ICU readmissions in ICH patients.
- Identified clinical factors like hydrocephalus and GCS are crucial for readmission risk assessment.
- Findings can aid in optimizing patient care and ICU resource allocation, with a call for prospective validation.
