Building Prediction Models for 30-Day Readmissions Among ICU Patients Using Both Structured and Unstructured Data in...

Area of Science:

• Healthcare Analytics
• Clinical Informatics
• Machine Learning in Medicine

Background:

• Intensive care unit (ICU) readmissions negatively impact patient health and hospital financial performance.
• High readmission rates increase mortality risk and reduce hospital profitability due to higher costs and decreased reimbursement.
• Accurate prediction of ICU readmissions is crucial for mitigating adverse outcomes and improving hospital efficiency.

Purpose of the Study:

• To develop and assess machine learning models for predicting 30-day readmission rates among ICU patients.
• To identify key predictors for ICU readmissions using both structured and unstructured clinical data.
• To evaluate the performance of various machine learning algorithms in forecasting patient readmissions.

Main Methods:

• Utilized the MIMIC-III database containing structured (demographics, labs, comorbidities) and unstructured (discharge summaries) patient data.
• Developed and compared multiple machine learning models, including Logistic Regression, to predict 30-day ICU readmissions.
• Evaluated model performance using the Area Under the Receiver Operating Characteristic curve (AUROC).

Main Results:

• The Logistic Regression model demonstrated the best performance, achieving an AUROC of 75.7%.
• Feature combinations including demographics, laboratory tests, comorbidities, and discharge summaries were explored.
• The study highlights the predictive power of machine learning for ICU readmission risk.

Conclusions:

• Machine learning models show significant potential for accurately predicting ICU readmissions.
• Early identification of high-risk patients can enable targeted interventions to reduce readmissions.
• Leveraging clinical data, including unstructured text, enhances predictive capabilities for patient outcomes.