Gradient boosting machine model in predicting the short-term clinical effects of PELD in treating lumbar disc herniation / 中华骨科杂志

ABSTRACT

Objective:To evaluate the prognostic effects of gradient boosting machine (GBM) model on the short-term effects of percutaneous endoscopic lumbar discectomy (PELD) in the treatment of lumbar disc herniation.Methods:Clinical data and outcomes of 475 patients who underwent PELD surgery for single-segment lumbar disc herniation from October 2016 to March 2018 were retrospectively collected. The lumbar JOA score was used as a reference for the evaluation of curative effects. The improvement rate ≥50% was considered as good curative effects, while <50% was considered as poor curative effects. GBM model and multivariate Logistic regression model were utilized to screen out the influencing factors of the short-term clinical effects of PELD. Prognostic models were established, receiver operating characteristic (ROC) curves were drawn and compared. Sensitivity, specificity and Youden index were compared to evaluate the predictive performance of GBM model.Results:A total of 395 patients were followed up effectively for 24 months. There were 347 patients (87.8%) with good curative effects. However, forty-eight patients (12.2%) had poor curative effects. There were statistically differences in the lumbar JOA score improvement rates between the groups in regards to age, location and type of herniated disc, degeneration level of intervertebral disc and facet joint in surgical segment, sagittal diameter of the protrusion and whether or not there was calcification, onset time to the surgery time period and degeneration level of intervertebral disc in adjacent segment ( P<0.05). The results of multivariate analysis showed that patients with age ≥60 [ OR=9.15, 95% CI(4.04, 20.73), P<0.001] and with larger sagittal diameter of the protrusion [ OR=1.37, 95% CI(1.18, 1.58), P<0.001] were more likely to have a poor prognosis. Patients with unilateral disc herniation had a better prognosis than the extreme lateral type [ OR=0.17, 95% CI(0.06, 0.55), P=0.003]. The prognoses of patients with grade Ⅲ intervertebral disc degeneration in surgical segment were worse than those with grade Ⅱ [ OR=0.17, 95% CI(0.04, 0.70), P=0.014]. The prognoses of patients with grade Ⅲ intervertebral disc degeneration in adjacent segment were worse than those with grade Ⅱ [ OR=0.29, 95% CI(0.10, 0.81), P=0.018]. The AUC predicted by GBM model was 0.92 [95% CI(0.77, 0.96)] with 93.46% sensitivity, 83.33% specificity and 0.77 Youden index. The above parameters were higher than those by the Logistic regression model. The predictive effects of the two models were both statistically significant ( P<0.001). The AUC values of the two models were also statistically significant ( Z=0.11, P<0.001). Conclusion:GBM model is better than multivariate logistic regression analysis model in predicting the short-term clinical effects of PELD in treating lumbar disc herniation.