目的 分析创伤性颈脊髓损伤（TCSCI）5年死亡率的危险因素，探讨据此构建的预测模型对TCSCI 5年死亡率的预测效能。 方法 回顾性队列研究。纳入2010年1月—2017年12月美国脊髓损伤数据库（NSCID）中TCSCI患者2 540例，其中男2 030例、女510例，年龄15~89岁。2 540例患者以3∶1比例按照随机抽样法分为开发队列1 927例和验证队列613例，其中开发队列用于分析危险因素、建立5年死亡率预测模型，验证队列用于预测模型的内部验证。根据开发队列患者伤后5年生存与否分为存活组1 554例和死亡组373例。观察指标：（1）对比开发队列和验证队列TCSCI患者的临床病理特征。（2）对开发队列中2组TCSCI患者临床病理特征、出院后是否因呼吸或泌尿系统疾病再入院、出院后是否需转医疗机构继续治疗等观察指标进行单因素分析和多因素logistic回归分析，分析TCSCI后5年死亡率的独立危险因素。根据危险因素构建预测TCSCI 5年死亡率的列线图模型。（3）对验证队列采用受试者操作特征（ROC）曲线、校准曲线、决策曲线评估列线图模型的性能。 结果 （1）开发队列和验证队列TCSCI患者性别、年龄、合并伤、治疗方式、脊髓损伤程度等临床病理特征比较，差异均无统计学意义（P值均>0.05）。（2）开发队列单因素分析显示：年龄、损伤节段、美国脊髓损伤委员会（ASIA）脊髓损伤分级、椎体骨折、手术、住院期间机械通气、出院后因呼吸/泌尿系统疾病再入院、出院后是否需转医疗机构继续治疗是TCSCI 5年死亡率的影响因素（P值均<0.05）。多因素logistic回归分析显示：年龄、ASIA脊髓损伤分级、住院期间机械通气、出院后因呼吸或泌尿系统疾病再入院、出院后是否需转医疗机构继续治疗是TCSCI 5年死亡率的独立危险因素（P值均<0.05）。根据危险因素构建列线图预测模型，验证队列对列线图预测模型进行内部验证：ROC曲线分析显示，曲线下面积为0.841，95%可信区间 0.817~0.864，阈值为0.219，灵敏度为82.5%，特异度为70.0%，列线图预测模型诊断性能较好；校准曲线分析显示，列线图预测模型预测死亡率和实际死亡率有良好的一致性；决策曲线分析显示，列线图预测模型有较高的临床应用价值。 结论 高龄、入院时高ASIA脊髓损伤分级、住院期间机械通气、出院后因呼吸或泌尿系统疾病再入院、出院后需要去医疗机构继续治疗是TCSCI 5年死亡率的独立危险因素。根据危险因素构建的列线图预测模型对TCSCI患者的5年死亡率具有较好的预测效能。 Objective This study aims to explore the clinical risk factors of the five-year mortality of traumatic cervical spinal cord injury (TCSCI) and the predictive efficacy of the prediction model. Methods In this retrospective cohort study, 2 540 patients were selected from the National Spinal Cord Injury Database from January 2010 to December 2017, including 2 030 males and 510 females with age in the range of 15-89 years. The 2 540 patients were randomly divided into the development cohort (1 927 patients) and the validation cohort (613 patients) in a 3∶1 ratio. The development cohort was used for analyzing of risk factors and establishing a five-year mortality prediction model. The validation cohort was used for the internal validation of the model. According to whether the patients in the development cohort died five years after TCSCI, they were divided into death group (373 patients) and survival group (1 554 patients). Outcome measures: (1) The clinicopathological features of patients with TCSCI in the development cohort and the validation cohort were compared. (2) The clinical baseline data of the two groups of patients with TCSCI in the development cohort, as well as observation indicators such as whether they were readmitted to hospital due to respiratory or urinary diseases after discharge and whether patients should go to a medical institution for further treatment after discharge, were analyzed by univariate analysis and multivariate logistic regression analysis to determine the independent risk factors for the five-year mortality of TCSCI. A nomogram model was constructed to predict the five-year mortality according to the risk factors. (3) In the validation cohort, the receiver operating characteristic (ROC) curve, the calibration curve, and the decision curve were used to evaluate the predictive performance of the nomogram model. Results (1) No significant difference existed between the clinicopathological features of patients with TCSCI in the development cohort and the validation cohort in gender, age, combined injury and degree of spinal cord injury (all P values >0.05). (2) Univariate analysis of the development cohort showed age, injury segment, American Spinal Injury Association (ASIA) spinal cord injury grade, vertebral fracture, surgery, mechanical ventilation during hospitalization, readmission to hospital for respiratory or urinary diseases after discharge, and patients should go to a medical institution for further treatment after discharge were the influencing factors for the five-year mortality of TCSCI (all P values <0.05). Multivariate logistic regression analysis showed age, ASIA spinal cord injury grade, mechanical ventilation during hospitalization, readmission for respiratory or urinary diseases after discharge, and patients should go to a medical institution for further treatment after discharge were independent risk factors for the five-year mortality of TCSCI (all P values <0.05). A nomogram prediction model was constructed according to the risk factors, and an internal verification of the nomogram model was conducted on the validation cohort. The ROC curve showed the area under the curve was 0.841, the 95% confidence interval was 0.817-0.864, the threshold was 0.219, the sensitivity was 82.5%, the specificity was 70.0%, and the diagnostic performance of the nomogram model was satisfactory. The calibration curves revealed the predicted and actual five-year mortality probabilities were fitted well. The decision curve analysis demonstrated the clinical value of this nomogram. The calibration curve showed the nomogram prediction model exhibited good consistency between the predicted mortality rate and the actual mortality rate. The analysis of decision curve showed the nomogram prediction model had a high clinical application value. Conclusion The results show advanced age, high ASIA spinal cord injury grade, mechanical ventilation during hospitalization, readmission due to respiratory or urinary system diseases, and patients should go to a medical institution for further treatment after discharge are independent risk factors for the five-year mortality of TCSCI. The nomogram prediction model constructed according to the risk factors displays good prediction performance and which can provide a reference for predicting the five-year mortality of TCSCI.
Analysis of risk factors and construction of prediction model for five-year mortality of traumatic cervical spinal cord injury
Objective This study aims to explore the clinical risk factors of the five-year mortality of traumatic cervical spinal cord injury (TCSCI) and the predictive efficacy of the prediction model. Methods In this retrospective cohort study, 2 540 patients were selected from the National Spinal Cord Injury Database from January 2010 to December 2017, including 2 030 males and 510 females with age in the range of 15-89 years. The 2 540 patients were randomly divided into the development cohort (1 927 patients) and the validation cohort (613 patients) in a 3∶1 ratio. The development cohort was used for analyzing of risk factors and establishing a five-year mortality prediction model. The validation cohort was used for the internal validation of the model. According to whether the patients in the development cohort died five years after TCSCI, they were divided into death group (373 patients) and survival group (1 554 patients). Outcome measures: (1) The clinicopathological features of patients with TCSCI in the development cohort and the validation cohort were compared. (2) The clinical baseline data of the two groups of patients with TCSCI in the development cohort, as well as observation indicators such as whether they were readmitted to hospital due to respiratory or urinary diseases after discharge and whether patients should go to a medical institution for further treatment after discharge, were analyzed by univariate analysis and multivariate logistic regression analysis to determine the independent risk factors for the five-year mortality of TCSCI. A nomogram model was constructed to predict the five-year mortality according to the risk factors. (3) In the validation cohort, the receiver operating characteristic (ROC) curve, the calibration curve, and the decision curve were used to evaluate the predictive performance of the nomogram model. Results (1) No significant difference existed between the clinicopathological features of patients with TCSCI in the development cohort and the validation cohort in gender, age, combined injury and degree of spinal cord injury (all P values >0.05). (2) Univariate analysis of the development cohort showed age, injury segment, American Spinal Injury Association (ASIA) spinal cord injury grade, vertebral fracture, surgery, mechanical ventilation during hospitalization, readmission to hospital for respiratory or urinary diseases after discharge, and patients should go to a medical institution for further treatment after discharge were the influencing factors for the five-year mortality of TCSCI (all P values <0.05). Multivariate logistic regression analysis showed age, ASIA spinal cord injury grade, mechanical ventilation during hospitalization, readmission for respiratory or urinary diseases after discharge, and patients should go to a medical institution for further treatment after discharge were independent risk factors for the five-year mortality of TCSCI (all P values <0.05). A nomogram prediction model was constructed according to the risk factors, and an internal verification of the nomogram model was conducted on the validation cohort. The ROC curve showed the area under the curve was 0.841, the 95% confidence interval was 0.817-0.864, the threshold was 0.219, the sensitivity was 82.5%, the specificity was 70.0%, and the diagnostic performance of the nomogram model was satisfactory. The calibration curves revealed the predicted and actual five-year mortality probabilities were fitted well. The decision curve analysis demonstrated the clinical value of this nomogram. The calibration curve showed the nomogram prediction model exhibited good consistency between the predicted mortality rate and the actual mortality rate. The analysis of decision curve showed the nomogram prediction model had a high clinical application value. Conclusion The results show advanced age, high ASIA spinal cord injury grade, mechanical ventilation during hospitalization, readmission due to respiratory or urinary system diseases, and patients should go to a medical institution for further treatment after discharge are independent risk factors for the five-year mortality of TCSCI. The nomogram prediction model constructed according to the risk factors displays good prediction performance and which can provide a reference for predicting the five-year mortality of TCSCI.
Spinal cord injuriesMortalityRisk factorsNomogram model
万方数据
