基于评估学信息的意识障碍患者6个月结局预测列线图模型的建立

关键词：意识障碍  预后预测  评估  列线图

Keywords:disorders of consciousness  prognostic prediction  assessment  nomogram

摘要：

      摘要 目的：基于意识障碍患者的行为学量表和神经电生理学评估建立一种稳定可靠预测意识障碍患者6个月预后结局的列线图模型。 方法：回顾性收集2020年2月1日至2021年7月1日在南昌大学第一附属医院康复医学科首诊并住院治疗的意识障碍患者临床资料（20个预测因子），结局指标采用格拉斯哥结局扩展量表评分。对缺失值进行多重插补后使用LASSO回归法筛选优化变量，将所得预测因子构建多因素Logistic回归预测模型，并绘制出列线图，最后采用受试者工作特征曲线和校准曲线对模型的性能进行评估。 结果：根据纳入及排除标准本研究共纳入144例患者，其中79例预后良好，65例预后不良。通过LASSO回归分析法最终筛选出5个预测变量，当患者FOUR评分高、无瞳孔散大、昏迷时间＜4周、体感诱发电位Ⅰ—Ⅱ级、肌张力≤1+级时6个月预后良好的可能性越高。该模型的ROC曲线下面积为0.98（95%CI：0.97—0.99）＞0.75，校准曲线的预测概率与实际结果吻合度较好，说明该模型具有良好的区分度和校准度。 结论：本文构建的基于意识障碍患者行为学量表和神经电生理学技术及临床特征的多领域预后模型可能为临床早期预测意识障碍患者预后结局提供部分依据。

Abstract:

      Abstract Objective： To establish a a robust and reliable nomogram for predicting the awake outcome of patients with consciousness disorder caused by brain injury at six months based on behavioral scale scores and neuroelectrophysiological techniques. Method： The clinical data (20 predictors) of patients with consciousness disorder who were first diagnosed and hospitalized in the Department of Rehabilitation Medicine of the First Affiliated Hospital of Nanchang University from February 1，2020 to July1， 2021 were collected retrospectively. Outcome indicators were scored by the Glasgow Outcome Scale-Extended. After multiple imputation of missing values， the LASSO regression method was used to screen the optimized variables, then, the prediction model was constructed by Multivariable Logistic regression analysis， and the nomogram was drawn. Finally， the performance of the model was evaluated by the receiver operating characteristic curve and calibration curves. Result： A total of 144 patients were included in this study according to inclusion and exclusion criteria，of which 79 had a good prognosis and 65 had a poor prognosis. Five predictors were screened out by LASSO logistic regression analysis. Patients with high FOUR score， no pupil dilation， coma time <4 weeks， SEP grade Ⅰ —Ⅱ， muscle tension ≤1+ had a higher probability of good prognosis at 6 months. The AUC of the model is 0.98（95%CI： 0.97—0.99） >0.75， and the prediction probability of calibration curve is in good agreement with the actual result， indicating that the model has good discrimination and calibration degree. Conclusion： The multi-domain prognostic model based on behavioral scale scores and neuroelectrophysiological techniques and clinical characteristics of patients with consciousness disorders has certain accuracy， which can provide partial reference for predicting the prognosis of patients consciousness disorders in early clinical stage.