Predictive modeling for 14-day unplanned hospital readmission risk by using machine learning algorithms.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Early unplanned hospital readmissions increase patient harm, healthcare costs, and negatively impact hospital reputation.
• Identifying patients at high risk for readmission is crucial for implementing targeted interventions.
• This study focused on developing predictive models for 14-day unplanned readmissions.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 14-day unplanned hospital readmissions.
• To identify key predictors associated with unplanned readmissions.
• To provide a tool for early identification of high-risk patients.

Main Methods:

• Retrospective cohort study of 24,722 adult patients from a university hospital.
• Utilized logistic regression, random forest, extreme gradient boosting, and categorical boosting (Catboost) algorithms.
• Model performance evaluated using precision, recall, F1-score, AUROC, and AUPRC.

Main Results:

• The Catboost model demonstrated superior performance, achieving an AUROC of 0.9903 and AUPRC of 0.7515.
• Incorporating 21 influential features improved the Catboost model's performance (AUROC: 0.9909, AUPRC: 0.7711).
• The 14-day unplanned readmission rate in the cohort was 1.22%.

Conclusions:

• Developed machine learning models reliably predict 14-day unplanned readmissions.
• Models identified influential features, particularly diagnosis-related, for risk prediction.
• These models can facilitate early discharge planning and transitional care to prevent readmissions.