梯度提升机模型对腰椎间盘突出症经皮内镜切除术近期疗效的预测作用

Gradient boosting machine model in predicting the short-term clinical effects of PELD in treating lumbar disc herniation

目的探讨梯度提升机(gradient boosting machine,GBM)模型对经皮内镜腰椎间盘切除术(percutaneous endoscopic lumbar discectomy,PELD)治疗腰椎间盘突出症近期疗效的预测效果。方法回顾性分析2016年10月至2018年3月因单节段腰椎间盘突出症行PELD手术的475例患者的临床资料,疗效评价采用腰椎JOA评分,评分改善率≥50%为疗效良好、<50%为疗效不佳。采用GBM模型及多因素Logistic回归分析模型筛选影响手术疗效的因素,比较二者的受试者工作特征(receiver operator characteristic,ROC)曲线,计算敏感度、特异度、约登指数,评价GBM模型的预测效果。结果395例患者获得随访,随访时间24个月。疗效优良者347例(87.8%)、疗效不佳者48例(12.2%)。不同年龄、椎间盘突出部位及突出类型、手术节段椎间盘及关节突关节退变水平、突出物矢状径及有无钙化、初次症状至手术时间、邻近节段椎间盘退变水平组疗效优良率的差异有统计学意义(P<0.05)。多因素Logistic回归分析显示,年龄≥60岁[OR=9.15,95%CI(4.04,20.73),P<0.001]、突出物矢状径大[OR=1.37,95%CI(1.18,1.58),P<0.001]是术后2年疗效不佳的危险因素;椎间盘突出单侧型相对于极外侧型疗效不佳的发生概率更小[OR=0.17,95%CI(0.06,0.55),P=0.003];手术节段椎间盘退变Ⅱ级者相对于Ⅲ级者疗效不佳的发生概率更小[OR=0.17,95%CI(0.04,0.70),P=0.014];邻近节段椎间盘退变2级者相对于3级者疗效不佳的发生概率更小[OR=0.29,95%CI(0.10,0.81),P=0.018]。GBM模型预测预后的ROC曲线下面积为0.92[95%CI(0.77,0.96)],灵敏度、特异度和约登指数分别为93.46%、83.33%和0.77,均高于Logistic回归模型的0.86[95%CI(0.73,0.90)]、88.79%、66.67%和0.55。GBM模型和Logistic回归分析模型的预测效应均有统计学意义(P<0.001),两者ROC曲线下面积值的差异有统计学意义(Z=0.11,P<0.001)。结论在预测PELD治疗腰椎间盘突出症的近期疗效方面,GBM模型的预测效能高于多因素Logistic回归分析模型。

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 gradeⅢintervertebral 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.