Prediction model of recurrence in patients with lumbar disc herniation after unilateral biportal endoscopy spinal surgery- XGBoost machine learning model can be interpreted based on SHAP.

BACKGROUND

Patients with single-segment lumbar disc herniation(LDH) who have undergone unilateral biportal endoscopic (UBE) spinal surgery face the risk of postoperative recurrence. The potential factors for recurrence of lumbar disc herniation(rLDH) are not fully understood. AIMS: This study aimed to investigate the risk factors for rLDH after UBE spinal surgery and establish a risk factor prediction model using machine learning to predict the recurrence rate. STUDY

DESIGN

This study used LASSO regression for feature screening, XGBoost for machine learning model construction, ROC curve analysis, calibration curve analysis, clinical decision curve analysis, specificity, sensitivity, accuracy, and F1 score to evaluate the model’s performance.

RESULTS

The clinical data of 271 patients with LDH were retrospectively analyzed. All patients have undergone UBE spinal surgery. 47 patients relapsed, with a recurrence rate of 17.3%. The XGBoost model predicted a higher AUC for rLDH (training: 0.927, 95%CI: 0.882–0.958; Test: 0.866, 95%CI: 0.760–0.937). The importance of SHAP variables in the model in descending order is: “Width of Protrusion Base(WPB), “” Bone Removal Range(BRR), “"Modic Change,” " Type of LDH,” “Middle Vertebral space Height(MVH),” “Hyperlipidemia.”

CONCLUSION

The large WPB、high WPB、 extensive BRR, and preoperative Modic Changes are more likely to develop rLDH. Hyperlipidemia and prolapsed disc herniation are negatively correlated with rLDH. RELEVANCE TO CLINICAL PRACTICE: Improve the ability to predict the rLDH after UBE spinal surgery. This will enable clinicians to intervene early and improve postoperative outcomes.