Interpretable Machine Learning Model for Predicting Prolonged Postoperative Length of Stay in Lung Cancer Patients Undergoing Day Surgery: A Retrospective Cohort Study.

BACKGROUND

Prolonged postoperative length of stay (p-LOS) in lung cancer patients is associated with poorer prognosis and increased healthcare burden, highlighting the need for early identification of high risk individuals. This study aimed to develop and validate a machine learning model to predict p-LOS in patients undergoing day surgery for lung cancer.

METHODS

A retrospective analysis was conducted on 1009 patients who underwent day surgery for lung cancer. Prolonged p-LOS was defined as hospitalization exceeding 48 hours. Feature selection was performed using Lasso regression, and predictive models were developed using 6 machine learning algorithms with internal validation. Model performance was evaluated using AUC and other metrics, and the SHAP method was applied to interpret feature contributions and individual predictions.

RESULTS

Among the included patients, 128 (12.69%) experienced prolonged p-LOS. The LightGBM model demonstrated optimal performance, achieving an AUC of 0.947 (0.919-0.974). Key predictive factors included fatigue, subcutaneous emphysema, pain, and cough.

CONCLUSIONS

The developed machine learning model accurately predicts prolonged p-LOS risk in lung cancer patients following day surgery, with high discriminative ability and interpretability. It shows potential for supporting early clinical intervention and optimizing postoperative care.