ICU卒中患者动态GCS轨迹与死亡风险:多中心年龄分层研究
这项回顾性多中心队列研究纳入MIMIC-IV、eICU和NSICU三个数据库共7876例非创伤性卒中ICU患者,用潜类别增长模型识别出4种GCS轨迹:稳定高(35.7%)、快速改善(38.0%)、持续中等(18.8%)、持续低(7.5%)。与稳定高组相比,全调整模型下快速改善、持续中等、持续低组的院内死亡HR分别为2.94(95%CI 2.24-3.84)、8.11(95%CI 6.18-10.66)、17.72(95%CI 13.30-23.59)。将GCS轨迹加入Boruta筛选的基线模型后,AUC从0.800升至0.860(IDI 0.085,连续NRI 0.283,均P<0.001)。
要点
- 非创伤性卒中ICU患者中,持续低GCS轨迹组院内死亡HR为17.72(95%CI 13.30-23.59),风险最高。
- 快速改善轨迹组死亡风险仍高于稳定高组,HR 2.94(95%CI 2.24-3.84),不能因GCS改善而放松警惕。
- 持续中等轨迹组死亡HR 8.11(95%CI 6.18-10.66),介于快速改善与持续低之间,呈阶梯式升高。
- 将GCS轨迹加入Boruta筛选基线模型,AUC从0.800提至0.860(IDI 0.085,连续NRI 0.283,均P<0.001)。
- 老年患者(≥65岁)中观察到一种独特的逐渐恶化轨迹,需对该人群动态监测GCS变化。
适用人群与场景
- 纳入:ICU收治的非创伤性卒中(NTS)患者,来自MIMIC-IV、eICU和NSICU三个数据库,共7876例。
- 年龄分层分析:老年患者定义为≥65岁。
- 排除:原文未提及具体排除标准。
诊断标准与关键阈值
- 采用潜类别增长模型(LCGM)识别GCS轨迹,分为稳定高、快速改善、持续中等、持续低四类。
- 基于阈值的平均GCS AUC指标与院内死亡相关,但原文未给出具体阈值界值。
处置流程与药物剂量
- 原文未提及
红旗征与不能漏的情况
- 持续低GCS轨迹:院内死亡HR 17.72(95%CI 13.30-23.59),需高度警惕。
- 持续中等GCS轨迹:院内死亡HR 8.11(95%CI 6.18-10.66)。
- 快速改善轨迹:死亡风险仍高于稳定高组(HR 2.94,95%CI 2.24-3.84),不应因GCS改善而误判为低危。
- 老年患者(≥65岁)出现逐渐恶化轨迹,需及时识别并处理。
研究设计与关键结果
- 研究类型:回顾性多中心队列研究,纳入MIMIC-IV、eICU、NSICU三个数据库共7876例非创伤性卒中ICU患者。
- 干预与对照:以稳定高GCS轨迹组为参照,比较快速改善、持续中等、持续低轨迹组的院内死亡风险。
- 主要终点:全调整模型下,快速改善、持续中等、持续低组的院内死亡HR分别为2.94(95%CI 2.24-3.84)、8.11(95%CI 6.18-10.66)、17.72(95%CI 13.30-23.59)。
- 预测性能:GCS轨迹加入Boruta筛选基线模型后,AUC从0.800升至0.860,IDI 0.085,连续NRI 0.283(均P<0.001)。
- 关键次要终点:基于阈值的平均GCS AUC指标与院内死亡相关;老年患者(≥65岁)中观察到独特的逐渐恶化轨迹。
证据类型、等级与局限
- 证据类型:回顾性多中心队列研究。
- 局限:回顾性设计,原文未报告具体证据等级或推荐强度;数据来自三个数据库,可能存在选择偏倚和信息偏倚。
对床旁实践的改变
- 对非创伤性卒中ICU患者,应动态监测GCS轨迹而非仅看单次评分,持续低或持续中等轨迹提示死亡风险显著升高。
- 快速改善轨迹患者死亡风险仍高于稳定高组(HR 2.94),不宜过早降级监护或放松治疗。
- 老年(≥65岁)非创伤性卒中患者需警惕逐渐恶化的GCS轨迹,及时识别并干预。
- GCS轨迹可作为神经重症风险分层的补充工具,但原文未给出具体干预阈值或治疗推荐,不能替代临床决策。
指南或共识出处
- 原文未提及
以上提炼依据论文摘要(全文未开放获取,或许可不允许翻译转载);全文请点「阅读原文」。
英文原文(摘要)· 点击展开
OBJECTIVE: The purpose of this study was to determine whether dynamic Glasgow Coma Scale (GCS) trajectories are associated with mortality risk and provide incremental prognostic information beyond selected baseline clinical variables in intensive care unit (ICU)-admitted patients with non-traumatic stroke (NTS).
METHODS: In this retrospective cohort study of 7876 patients from the MIMIC-IV, eICU, and NSICU databases, latent class growth modeling (LCGM) was used to identify GCS trajectories. The associations of trajectory groups and threshold-based GCS metrics with in-hospital mortality were evaluated using multivariable Cox regression. The incremental value of trajectories was assessed using variable sets selected by Boruta, least absolute shrinkage and selection operator (LASSO), and best subset selection (BSS).
RESULTS: Four distinct GCS trajectories were identified: stable high (35.7%), rapid improvement (38.0%), persistent moderate (18.8%), and persistent low (7.5%). Compared with the stable high group, mortality risk increased stepwise in fully adjusted models, with HRs of 2.94 (95% CI 2.24-3.84), 8.11 (95% CI 6.18-10.66), and 17.72 (95% CI 13.30-23.59) for rapid improvement, persistent moderate, and persistent low, respectively. Threshold-based mean GCS AUC metrics were also associated with in-hospital mortality. A distinct gradual deterioration trajectory was observed in elderly patients (≥ 65 years). Adding GCS trajectories to the Boruta-selected baseline model improved predictive performance, increasing the AUC from 0.800 to 0.860, with an IDI of 0.085 and a continuous NRI of 0.283 (all P < 0.001).
CONCLUSIONS: Trajectory-based and threshold-based longitudinal GCS assessments were strongly associated with in-hospital mortality in critically ill patients with NTS, supporting the value of dynamic GCS monitoring as a complementary tool for risk stratification in neurocritical care.
参考来源
- Wang J, Dai W, Xu MM, et al. Dynamic GCS Trajectories and Mortality Risk in ICU Patients with Stroke: A Multicenter, Age-Stratified Study. Neurocrit Care. 2026. PMID: 42477259. doi:10.1007/s12028-026-02583-3
供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。