Development and Validation of a Web-Based Pediatric Readmission Risk Assessment Tool

Area of Science:

• Pediatric Healthcare Informatics
• Clinical Risk Prediction
• Health Services Research

Background:

• Unplanned readmissions (URs) in pediatric care present a significant challenge for healthcare systems.
• Accurate prediction of URs is crucial for effective resource allocation and patient management.
• Existing methods for predicting pediatric URs often lack accuracy or are not readily applicable before discharge.

Purpose of the Study:

• To develop and validate accurate algorithms for predicting pediatric unplanned readmissions within 3, 7, and 30 days of hospital discharge.
• To create a tool that utilizes pre-discharge clinical data for risk stratification.
• To support timely and informed discharge planning decisions in pediatric care.

Main Methods:

• Utilized a large retrospective cohort of over 1.3 million pediatric admissions from the Children's Hospital Association Pediatric Health Information System (2016-2017).
• Employed gradient boosting trees (XGBoost) to model complex interactions among patient predictors.
• Developed prediction models for 3, 7, and 30-day unplanned readmission risk.

Main Results:

• The study identified 1.6% of patients with 3-day URs, 2.4% with 7-day URs, and 4.4% with 30-day URs.
• Prediction models demonstrated strong discrimination, with AUCs of 0.771 (3-day), 0.778 (7-day), and 0.811 (30-day).
• A publicly available pediatric readmission risk score prediction tool was developed.

Conclusions:

• Pediatric unplanned readmission risk can be accurately predicted using information available before patient discharge.
• The developed algorithms and tool can be integrated into electronic medical record systems for practical application.
• This predictive capability can significantly enhance hospital discharge planning and potentially reduce readmission rates.