Interpretable prediction of prolonged length of stay for patients undergoing lumbar disc herniation surgery based on machine learning and SHAP

BACKGROUND: Prolonged length of stay (PLOS) is a key indicator of decreased healthcare quality. This study aims to identify the most suitable machine learning model for predicting PLOS in patients undergoing lumbar disc herniation surgery. METHODS: We retrospectively analyzed 1050 patients who underwent lumbar disc herniation surgery at two tertiary medical centers. Prolonged length of stay was defined as hospitalization duration exceeding the 75th percentile. Multivariable logistic regression was used for preliminary feature screening. Seven machine learning models were developed and compared for PLOS prediction. Model performance was evaluated from three aspects: discrimination, calibration and clinical utility. SHAP analysis was applied to interpret the optimal model’s characteristics and individual case predictions. RESULTS: Four features significantly associated with PLOS were identified for model development: diabetes, open surgery, activated partial thromboplastin time (APTT), and disc height index (DHI). The GBM model demonstrated superior performance, achieving an AUC of 0.751 (95% CI: 0.684–0.818) on the test set and an AUC of 0.794 (95% CI: 0.683–0.905) on the external validation set. SHAP analysis identified DHI as the most influential feature for determining PLOS risk in the model. Furthermore, we have developed a freely accessible web-based calculator, available at: https://hbss2222.shinyapps.io/PLOS/ . CONCLUSION: We successfully developed a machine learning model for predicting prolonged length of stay in patients undergoing lumbar disc herniation surgery, which holds potential for assisting clinical decision-making.