经皮椎体后凸成形术后继发椎体压缩性骨折的危险因素分析及预测模型的建立与验证

Analysis of risk factors associated with subsequent vertebral compression fracture after percutaneous kyphoplasty and the development and validation of a predictive model

中文关键词:  经皮椎体后凸成形术  继发椎体压缩性骨折  危险因素  预测模型

英文关键词:Percutaneous kyphoplasty  Subsequent vertebral compression fracture  Risk factors  Predictive model

中文摘要:

  【摘要】 目的：分析经皮椎体后凸成形术（percutaneous kyphoplasty，PKP）后继发椎体压缩性骨折（subsequent vertebral compression fracture，SVCF）的危险因素，建立预测模型并进行验证。方法：回顾性分析在徐州医科大学附属医院骨科行PKP治疗的415例患者的临床资料，其中有24例（5.8%）患者在术后出现SVCF，所有患者术后随访时间16~30个月，根据术后是否出现SVCF，将患者分为SVCF组与非SVCF组。对两组患者的性别、年龄、体重指数（body mass index，BMI）、骨密度（bone mineral density，BMD）、临床合并症史（高血压病史、糖尿病史）、骨水泥注射量、骨水泥渗漏情况、初始手术节段数及手术前后的Cobb角和椎体前缘高度（anterior vertebral height，AVH）进行单因素分析。根据套索（LASSO）回归，选择与PKP术后SVCF显著相关的危险因素；随后将筛选出来的危险因素纳入多因素Logistic回归分析，最终依据多因素Logistic回归分析结果建立预测模型，并通过绘制列线图对模型进行可视化，以此来预测PKP术后发生 SVCF 的风险概率。使用增强自举法（Bootstrap）进行模型内部验证，绘制校正曲线和受试者工作特征（receiver operating characteristic，ROC）曲线来评估模型性能。使用决策曲线分析（decision curve analysis，DCA）曲线和临床影响曲线（clinical impact curve，CIC）评估该模型的临床效用。结果：SVCF组的年龄、BMI、BMD、骨水泥注射量及骨水泥渗漏情况与非SVCF组比较，差异有统计学意义（P＜0.05）。LASSO回归筛选出与PKP术后SVCF密切相关的危险因素为年龄、BMI、BMD、骨水泥注射量及骨水泥渗漏。多因素Logistic回归分析显示，年龄、BMI、BMD、骨水泥渗漏情况和骨水泥注射量是PKP术后SVCF的独立危险因素。可视化的预测模型（列线图）中的预测因子包括年龄、BMI、BMD、骨水泥注射量和骨水泥渗漏情况。建立的预测模型内部验证结果显示，模型的偏差校正曲线与表观曲线拟合良好，ROC曲线下面积为0.945，95%置信区间为0.897~0.993。DCA曲线显示，在0.1~0.9的阈值区间具有最大效益。CIC表明，预测模型可以在阈值概率范围内有效区分出SVCF的高危患者。结论：年龄、BMD、BMI、骨水泥注射量和骨水泥渗漏情况是SVCF的独立危险因素，基于这5个危险因素所建立的预测模型，可以准确地预测PKP术后发生SVCF的风险概率，以便在术后早期识别SVCF的高危人群并予以提前干预措施。

英文摘要:

  【Abstract】 Objectives: To analyse the risk factors affecting subsequent vertebral compression fractures(SVCF) after percutaneous kyphoplasty(PKP), and to develop and validate a predictive model. Methods: The clinical data of 415 patients treated with PKP in the Department of Orthopedics, the Affiliated Hospital of Xuzhou Medical University were retrospectively analyzed. 24 patients(5.8%) had SVCF after surgery, and the follow-up period was 16-30 months. According to whether the patients had SVCF after operation, the patients were divided into SVCF group and non-SVCF group. Patients′ gender, age, body mass index(BMI), bone mineral density(BMD), history of clinical comorbidities(history of hypertension, history of diabetes), bone cement injection dosage, bone cement leakage, number of initial surgical segments, Cobb angle and anterior vertebral height(AVH) before and after surgery were analyzed by univariate analysis. Based on LASSO regression, risk factors significantly associated with SVCF after PKP were selected. Then, the selected risk factors were included in multivariate Logistic regression analysis. Finally, a prediction model was developed according to the results of multivariate Logistic regression analysis, and the model was visualized by drawing a nomogram to predict the risk probability of SVCF after PKP. The enhanced bootstrap method was used for internal validation of the model. Calibration curve analysis and receiver operating characteristic(ROC) curve were drawn to evaluate the model performance. Decision curve analysis(DCA) and clinical impact curve(CIC) were used to evaluate the clinical utility of the model. Results: There were significant differences in age, BMI, BMD, bone cement injection dosage and bone cement leakage between the SVCF group and the non-SVCF group(P<0.05). LASSO regression analysis showed that age, BMI, BMD, bone cement injection dosage and bone cement leakage were closely related to SVCF after PKP. Multivariate Logistic regression analysis showed that age, BMI, BMD, bone cement leakage and bone cement injection dosage were independent risk factors for SVCF after PKP. The predictive factors in the visual prediction model(nomogram) included age, BMI, BMD, bone cement injection volume and bone cement leakage. The deviation-corrected curve of the model fitted well with the apparent curve after enhanced bootstrap method testing, and the area under the ROC curve was 0.945 with a 95% confidence interval of 0.897-0.993. As shown by the DCA curve, the maximum benefit was obtained in the threshold range of 0.1-0.9. CIC showed that the predictive model can effectively identify high-risk patients with SVCF within the threshold probability. Conclusions: Age, BMD, BMI, bone cement injection dosage and bone cement leakage are independent risk factors for SVCF, and the prediction model developed based on these five risk factors can accurately predict the risk probability of SVCF after PKP and identify the high-risk group in the early postoperative period to focus.