Predicting ICU readmission using grouped physiological and medication trends.

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Predictive Modeling

Background:

• Intensive care unit (ICU) readmissions are associated with high mortality and prolonged hospital stays.
• Accurate prediction of ICU readmission risk can guide clinical decisions and prevent avoidable readmissions.
• Current models often rely on snapshot physiological data, neglecting valuable temporal trend information.

Purpose of the Study:

• To develop a more accurate prediction model for 30-day ICU readmission risk.
• To investigate the predictive value of temporal trends in physiological and medication data.
• To enhance existing ICU readmission prediction by incorporating detailed trend analysis.

Main Methods:

• Utilized the Multiparameter Intelligent Monitoring in Intensive Care II (MIMIC-II) dataset.
• Converted time-series data into trend graphs and applied frequent subgraph mining to extract temporal trends.
• Grouped trends using non-negative matrix factorization and trained a logistic regression model.

Main Results:

• The developed model demonstrated superior performance compared to existing methods.
• Achieved an approximate 4% improvement in the area under the receiver operating characteristic curve (AUC).
• Identified grouped physiological and medication trends as strong predictors of ICU readmission risk.

Conclusions:

• Temporal trends in physiological and medication data contain significant predictive information for ICU readmission.
• Incorporating these grouped trends as complementary features to snapshot measurements enhances prediction accuracy.
• This approach offers a more comprehensive method for predicting 30-day ICU readmission risk.