Interpretable Machine Learning Model for Predicting Prolonged Postoperative Length of Stay in Lung Cancer Patients...

Area of Science:

• Oncology
• Medical Informatics
• Surgical Outcomes Research

Background:

• Prolonged postoperative length of stay (p-LOS) in lung cancer patients is linked to worse prognosis and higher healthcare costs.
• Early identification of patients at risk for p-LOS is crucial for optimizing care.
• Day surgery for lung cancer necessitates precise prediction of postoperative stay.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting p-LOS in lung cancer patients undergoing day surgery.
• To identify key clinical factors contributing to prolonged postoperative hospitalization.
• To enhance early clinical intervention strategies for lung cancer surgery patients.

Main Methods:

• Retrospective analysis of 1009 lung cancer patients undergoing day surgery.
• Definition of p-LOS as hospitalization exceeding 48 hours.
• Feature selection via Lasso regression and ML model development (6 algorithms) with internal validation; SHAP for interpretability.

Main Results:

• 12.69% of patients experienced p-LOS.
• The LightGBM ML model achieved an AUC of 0.947 (95% CI: 0.919-0.974).
• Significant predictors of p-LOS included fatigue, subcutaneous emphysema, pain, and cough.

Conclusions:

• The developed ML model accurately predicts p-LOS in lung cancer day surgery patients.
• The model exhibits high discriminative ability and interpretability.
• This predictive tool can support timely clinical interventions and optimize postoperative management.