重型颅脑损伤后意识障碍患者高压氧促醒结局预测模型的构建与验证

Construction and Internal Validation of a Prediction Model for Awakening Outcomes after Hyperbaric Oxygen Therapy in Patients with Disorders of Consciousness following Severe Traumatic Brain Injury

摘要: 目的：构建并内部验证重型颅脑损伤后意识障碍患者高压氧促醒结局的预测模型，为急诊–高压氧–康复衔接阶段的风险分层和治疗沟通提供量化依据。方法：回顾性纳入2021年6月至2025年6月于合肥市第二人民医院急诊科收治并接受高压氧治疗的重型颅脑损伤后意识障碍患者168例，按7:3比例分为建模队列118例和内部验证队列50例。主要结局定义为高压氧治疗启动后3个月促醒结局阳性，依据CRS-R评分变化、病程记录和随访记录综合判定。收集人口学资料、入院GCS评分、基线CRS-R评分、伤后至首次高压氧时间、瞳孔对光反射、Marshall CT分级、弥漫性轴索损伤、血清NSE、血清S100B及高压氧疗程次数。采用单因素分析和多因素Logistic回归筛选变量并建立预测模型，采用受试者工作特征曲线、校准曲线、Brier评分和决策曲线评价模型性能。结果：全组促醒结局发生率为56.5%。建模队列与内部验证队列总体基线特征基本均衡；S100B水平存在统计学差异，但绝对差异较小且未进入最终模型。建模队列中，入院GCS评分、基线CRS-R评分、伤后至首次高压氧时间和血清NSE进入最终模型。模型在建模队列AUC为0.910 (95%CI: 0.851~0.958)，内部验证队列AUC为0.852 (95%CI: 0.731~0.956)。按预测概率分层后，低概率组、中等概率组和高概率组促醒率分别为12.3%、53.8%和87.1%。结论：基于入院GCS评分、基线CRS-R评分、伤后至首次高压氧时间和血清NSE构建的预测模型具有较好的区分度、校准度和临床净获益，可用于重型颅脑损伤后意识障碍患者高压氧促醒结局的早期风险分层。

Abstract: Objective: To construct and internally validate a predictive model for hyperbaric oxygen (HBO) therapy-induced arousal outcomes in patients with disorders of consciousness (DOC) following severe traumatic brain injury (sTBI), providing quantitative evidence for risk stratification and treatment communication during the emergency–HBO–rehabilitation transition phase. Methods: A total of 168 patients with DOC after sTBI who were admitted to the Emergency Department of Hefei Second People’s Hospital and received HBO therapy between June 2021 and June 2025 were retrospectively enrolled. They were divided into a modeling cohort (n = 118) and an internal validation cohort (n = 50) at a ratio of 7:3. The primary outcome was defined as a positive arousal outcome at 3 months after the initiation of HBO therapy, which was comprehensively determined based on changes in the Coma Recovery Scale-Revised (CRS-R) score, medical records, and follow-up data. Demographic data, admission Glasgow Coma Scale (GCS) score, baseline CRS-R score, time from injury to first HBO session, pupillary light reflex, Marshall CT classification, diffuse axonal injury, serum neuron-specific enolase (NSE), serum S100B, and number of HBO sessions were collected. Univariate analysis and multivariate Logistic regression were used to screen variables and establish the predictive model. Model performance was evaluated using receiver operating characteristic (ROC) curve, calibration curve, Brier score, and decision curve analysis (DCA). Results: The overall incidence of positive arousal outcomes in the entire cohort was 56.5%. The baseline characteristics were generally balanced between the modeling cohort and the internal validation cohort; although there was a statistically significant difference in S100B levels, the absolute difference was small and it did not enter the final model. In the modeling cohort, admission GCS score, baseline CRS-R score, time from injury to first HBO session, and serum NSE were included in the final model. The area under the ROC curve (AUC) of the model was 0.910 (95% CI: 0.851~0.958) in the modeling cohort and 0.852 (95% CI: 0.731~0.956) in the internal validation cohort. After stratification by predicted probability, the arousal rates in the low-, moderate-, and high-probability groups were 12.3%, 53.8%, and 87.1%, respectively. Conclusion: The predictive model constructed based on admission GCS score, baseline CRS-R score, time from injury to first HBO session, and serum NSE demonstrated favorable discrimination, calibration, and clinical net benefit, and can be used for early risk stratification of HBO therapy-induced arousal outcomes in patients with DOC following sTBI.