Early Prediction of Unplanned 30-Day Hospital Readmission: Model Development and Retrospective Data Analysis.

Area of Science:

• Health Informatics
• Clinical Decision Support
• Predictive Analytics

Background:

• Current readmission reduction strategies often begin too late in the patient care process.
• Preventive interventions during hospitalization are underexplored due to challenges in early risk prediction.
• Limited data availability early in hospital stays hinders timely risk assessment.

Purpose of the Study:

• To develop an early prediction model for unplanned 30-day hospital readmissions.
• To identify novel risk and protective factors associated with hospital readmissions.
• To improve patient outcomes through timely clinical interventions.

Main Methods:

• Utilized a large dataset of 96,550 patients from 205 hospitals (2016 Health Facts database).
• Developed prediction models using index admission data available within 24 hours and prior year encounter data.
• Employed extreme gradient boosting (XGBoost) and multivariate logistic regression for model development and factor identification.

Main Results:

• The best performing XGBoost model achieved an AUC of 0.753 (development) and 0.742 (validation).
• Identified 14 novel risk factors and 2 novel protective factors for readmission.
• The model demonstrates superior performance compared to widely used existing models.

Conclusions:

• The developed model enables early identification of patients at risk for readmission during hospitalization.
• Early risk identification allows clinicians to provide targeted attention and interventions for high-risk patients.
• The novel risk and protective factors enhance understanding of readmission determinants.