Personalized Risk Prediction for 30-Day Readmissions With Venous Thromboembolism Using Machine Learning.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Venous thromboembolism (VTE) poses a significant risk for hospital readmissions.
• Accurate prediction of VTE readmission risk is crucial for patient management.
• Existing risk stratification methods may require enhancement through advanced analytics.

Purpose of the Study:

• To develop and validate machine learning models for predicting 30-day VTE readmission risk.
• To identify patients at high risk for VTE after hospital discharge.
• To provide a tool for personalized risk assessment.

Main Methods:

• Retrospective analysis of structured electronic health records (EHRs).
• Development and evaluation of three predictive models: logistic regression, balanced random forest, and multilayer perceptron.
• Utilized data from a single academic hospital.

Main Results:

• The study included 158,804 admissions, with 2,080 (1.31%) VTE-positive cases.
• The balanced random forest model exhibited superior predictive performance compared to logistic regression and multilayer perceptron.
• A high-performing, validated risk prediction tool was developed.

Conclusions:

• Machine learning models, particularly the balanced random forest, can effectively predict 30-day VTE readmission risk.
• The developed tool aids in identifying high-risk patients for targeted interventions.
• This predictive model can inform clinical decisions to improve patient outcomes.