基于局部脑氧饱和度和经颅多普勒超声监测参数的中重型创伤性脑损伤预后预测模型的构建与验证

Construction and validation of prediction model on prognosis of moderate to severe traumatic brain injury based on regional cerebral oxygen saturation and transcranial Doppler ultrasound monitoring parameters

目的构建中重型创伤性脑损伤(msTBI)患者基于局部脑氧饱和度(rScO_(2))和经颅多普勒超声(TCD)监测参数的预后预测模型,并验证其效能。方法采用回顾性队列研究分析2021年1月至2022年12月河南省人民医院治疗的161例msTBI患者的临床资料,其中男104例,女57例;年龄19~76岁[(53.1±12.8)岁];格拉斯哥昏迷评分(GCS)3~12分[(7.0±1.9)分]。均行rScO_(2)和TCD监测。根据出院90 d改良Rankin量表(mRS)评分评估的患者预后,将患者分为预后良好组(mRS评分≤3分,88例)和不良预后组(mRS评分4~6分,73例)。收集两组一般资料、临床资料、rScO_(2)监测参数和TCD监测参数。采用单因素分析进行预后相关指标的差异性比较。采用多因素Logistic逐步回归分析确定msTBI患者不良预后结局的预测因子并构建回归方程。应用R语言绘制基于回归方程的列线图预测模型。绘制受试者工作特征(ROC)曲线,计算模型的曲线下面积(AUC)、灵敏度、特异度、约登指数,同时计算一致性指数(C指数),以衡量模型的区分度。采用Hosmer‑Lemeshow(H‑L)拟合优度检验评价模型的拟合性,以及校准曲线评价模型的校准度。采用决策曲线分析(DCA)评价模型的临床获益和临床适用性。结果两组临床资料(合并脑疝、入院时GCS、入院时急性生理学与慢性健康状况评估Ⅱ(APACHEⅡ)评分、入院时鹿特丹CT评分、入院时氧合指数、入院时平均动脉压)、rScO_(2)监测参数(rScO_(2)平均值、rScO_(2)最大值、rScO_(2)变异度)、TCD监测参数[收缩期峰血流速度(Vs)、平均血流速度(Vm)、搏动指数(PI)]差异均有统计学意义(P<0.05或0.01)。多因素Logistic逐步回归分析结果表明,合并脑疝(OR=9.28,95%CI 3.40,25.33,P<0.01)、入院时鹿特丹CT评分(OR=1.92,95%CI 1.32,2.78,P<0.01)、rScO_(2)变异度(OR=4.66,95%CI 1.74,12.43,P<0.01)、Vs(OR=0.66,95%CI 0.61,0.75,P<0.01)及PI(OR=20.07,95%CI 4.17,16.50,P<0.01)是msTBI患者不良预后结局的预测因子。依据上述5个变量构建回归方程:Logit[P(/1-P)]=2.23ד合并脑疝”+0.65ד入院时鹿特丹CT评分”+1.54דrScO_(2)变异度”-0.42דVs”+3.00דPI”-6.75。该预测模型对msTBI患者预后结局预测的AUC为0.90(95%CI 0.85,0.95),预测概率的灵敏度和特异度分别为86.3%和78.4%,约登指数为0.65,C指数为0.90。H‑L拟合优度检验显示,预测模型的拟合性较好(χ^(2)=12.58,P>0.05)。校准曲线平均绝对误差为0.025,提示该模型的校准度良好。DCA结果显示,该模型在风险阈值概率范围(20%~100%)内的净收益率高于参考模型,对预测msTBI患者不良预后风险有较好的临床应用价值。结论基于rScO_(2)与TCD监测参数(rScO_(2)变异度、Vs、PI)联合临床多指标特征(合并脑疝、入院时鹿特丹CT评分)构建的模型对msTBI患者预后具有良好的预测效能。

Objective To construct a prognostic predictive model for patients with moderate to severe traumatic brain injury(msTBI)based on regional cerebral oxygen saturation(rScO_(2))and transcranial Doppler ultrasound(TCD)monitoring parameters and validate its effectiveness.Methods A retrospective cohort study was conducted to analyze the clinical data of 161 patients with msTBI who were treated at Henan Provincial People′s Hospital from January 2021 to December 2022,including 104 males and 57 females,aged 19‑76 years[(53.1±12.8)years].Glasgow coma scale(GCS)score was 3‑12 points[(7.0±1.9)points].Both rScO_(2) and TCD monitoring were performed.Based on the results of prognostic evaluation of patients with the modified Rankin scale(mRS)score at 90 days after discharge,the patients were divided into good prognosis group(mRS score≤3 points,n=88)and poor prognosis group(mRS score of 4‑6 points,n=73).The following data of the two groups were collected:the general data,clinical data,rScO_(2) monitoring parameters and TCD monitoring parameters.Univariate analysis was employed to compare the differences in the relevant prognostic indicators.Multivariate Logistic stepwise regression analysis was conducted to determine the predictors of poor prognostic outcomes in msTBI patients and regression equations were constructed.A nomogram predictive model based on regression equations was drawn with R language.The discriminability of the model was evaluated by drawing the receiver operating characteristic(ROC)curve,to calculate the area under the curve(AUC),sensitivity,specificity,and Jordan index of the model,and measuring the consistency index(C index).Hosmer‑Lemeshow(H‑L)goodness of fit test was conducted to evaluate the fit of the model,and the calibration curve was used to evaluate the calibration degree of the model.Decision curve analysis(DCA)was employed to evaluate the clinical benefit and applicability of the model.Results There were significant differences between the two groups in the clinical data(cerebral hernia formation,GCS on admission,acute physiology and chronic health evaluation II(APACHE II)score on admission,Rotterdam CT score on admission,oxygenation index on admission,mean arterial pressure on admission),rScO_(2) monitoring parameters(mean rScO_(2),maximum rScO_(2),rScO_(2) variability),TCD monitoring parameters[peak systolic blood flow velocity(Vs),average blood flow velocity(Vm),pulse index(PI)](P<0.05 or 0.01).The results of multivariate Logistic stepwise regression analysis showed that cerebral hernia formation(OR=9.28,95%CI 3.40,25.33,P<0.01),Rotterdam CT score on admission(OR=1.92,95%CI 1.32,2.78,P<0.01),rScO_(2) variability(OR=4.66,95%CI 1.74,12.43,P<0.01),Vs(OR=0.66,95%CI 0.61,0.75,P<0.01)and PI(OR=20.07,95%CI 4.17,16.50,P<0.01)were predictive factors for poor prognosis in patients with msTBI.The regression equation was constructed with the forementioned 5 variables:Logit[P/(1-P)]=2.23×"brain hernia formation"+0.65×"Rotterdam CT score on admission"+1.54×"rScO_(2) variability"-0.42×"Vs"+3.00×"PI"-6.75.The AUC of prognostic predictive model of msTBI patients was 0.90(95%CI 0.85,0.95),with the sensitivity and specificity of 86.3%and 78.4%,Youden index of 0.65 and C index of 0.90.H‑L goodness of fit test showed that the calibration degree of the predictive model was accurate(χ^(2)=12.58,P>0.05).The average absolute error of the calibration curve was 0.025,showing that the calibration of the model was good.DCA results showed that this model had higher net return rate than the reference model within the probability range of risk threshold(20%‑100%),with good clinical application value in evaluating the risk of poor prognosis of msTBI patients.Conclusion The model constructed based on the combination of rScO_(2) and TCD monitoring parameters(rScO_(2) variability,Vs and PI)with multiple clinical indicators(cerebral hernia formation and Rotterdam CT score on admission)has good predictive performance for the prognosis of msTBI.