Externally Validated Machine Learning Models for 30-/90-/180-Day Unplanned Readmission After Head and Neck Cancer...

Area of Science:

• Oncology
• Health Informatics
• Machine Learning

Background:

• Unplanned readmissions after head and neck cancer (HNC) hospitalizations are frequent and expensive.
• Predictive models are needed to identify high-risk patients for targeted interventions.

Purpose of the Study:

• To develop and externally validate machine learning (ML) models for predicting unplanned HNC readmissions.
• To assess model performance across short (30-day) and longer-term (90/180-day) horizons.

Main Methods:

• Utilized the Nationwide Readmissions Database (2016-2020) for adult, nonelective HNC admissions.
• Engineered 247 discharge predictors and trained multiple ML models, including XGBoost.
• Evaluated model discrimination (AUC), calibration, and clinical utility using decision curve analysis and SHAP for interpretability.

Main Results:

• XGBoost demonstrated strong discrimination (AUC 0.725/0.746/0.756 for 30/90/180 days).
• Models showed acceptable calibration and clinical utility across various risk thresholds.
• Key predictors included artificial airway/nutrition status, discharge timing/disposition, and disease severity.

Conclusions:

• An XGBoost ML model effectively predicts short- and long-term unplanned readmission risk in HNC patients.
• The model, trained on administrative and clinical data, can support targeted discharge planning.
• This approach has the potential to improve patient outcomes and reduce healthcare expenditures.