Chatelle Camille, Rosenthal Eric S, Bodien Yelena G, Spencer-Salmon Camille A, Giacino Joseph T, Edlow Brian L
GIGA Consciousness, Coma Science Group, University of Liège, Avenue de l'Hôpital, 11, 4000, Liège, Belgium.
Center for Neurotechnology and Neurorecovery, Massachusetts General Hospital, Boston, MA, USA.
Neurocrit Care. 2020 Oct;33(2):449-457. doi: 10.1007/s12028-019-00904-3.
BACKGROUND/OBJECTIVE: Behavioral examinations may fail to detect language function in patients with severe traumatic brain injury (TBI) due to confounds such as having an endotracheal tube. We investigated whether resting and stimulus-evoked electroencephalography (EEG) methods detect the presence of language function in patients with severe TBI.
Four EEG measures were assessed: (1) resting background (applying Forgacs' criteria), (2) reactivity to speech, (3) background and reactivity (applying Synek's criteria); and (4) an automated support vector machine (classifier for speech versus rest). Cohen's kappa measured agreement between the four EEG measures and evidence of language function on a behavioral coma recovery scale-revised (CRS-R) and composite (CRS-R or functional MRI) reference standard. Sensitivity and specificity of each EEG measure were calculated against the reference standards.
We enrolled 17 adult patients with severe TBI (mean ± SD age 27.0 ± 7.0 years; median [range] 11.5 [2-1173] days post-injury) and 16 healthy subjects (age 28.5 ± 7.8 years). The classifier, followed by Forgacs' criteria for resting background, demonstrated the highest agreement with the behavioral reference standard. Only Synek's criteria for background and reactivity showed significant agreement with the composite reference standard. The classifier and resting background showed balanced sensitivity and specificity for behavioral (sensitivity = 84.6% and 80.8%; specificity = 57.1% for both) and composite reference standards (sensitivity = 79.3% and 75.9%, specificity = 50% for both).
Methods applying an automated classifier, resting background, or resting background with reactivity may identify severe TBI patients with preserved language function. Automated classifier methods may enable unbiased and efficient assessment of larger populations or serial timepoints, while qualitative visual methods may be practical in community settings.
背景/目的：由于诸如气管插管等混杂因素，行为检查可能无法检测出重度创伤性脑损伤（TBI）患者的语言功能。我们研究了静息和刺激诱发脑电图（EEG）方法能否检测出重度TBI患者的语言功能是否存在。
评估了四种EEG测量方法：（1）静息背景（应用福尔加克斯标准），（2）对言语的反应性，（3）背景和反应性（应用西内克标准）；以及（4）自动支持向量机（用于区分言语与静息的分类器）。科恩kappa系数用于衡量这四种EEG测量方法与行为昏迷恢复量表修订版（CRS-R）和综合版（CRS-R或功能磁共振成像）参考标准上语言功能证据之间的一致性。针对参考标准计算每种EEG测量方法的敏感性和特异性。
我们纳入了17例成年重度TBI患者（平均±标准差年龄27.0±7.0岁；中位数[范围]受伤后11.5[2 - 1173]天）和16名健康受试者（年龄28.5±7.8岁）。该分类器，其次是静息背景的福尔加克斯标准，与行为参考标准的一致性最高。只有背景和反应性的西内克标准与综合参考标准显示出显著一致性。该分类器和静息背景对行为参考标准（敏感性=84.6%和80.8%；特异性均为57.1%）和综合参考标准（敏感性=79.3%和75.9%，特异性均为50%）显示出平衡的敏感性和特异性。
应用自动分类器、静息背景或具有反应性的静息背景的方法可能识别出具有保留语言功能的重度TBI患者。自动分类器方法可能能够对更大规模人群或连续时间点进行无偏且高效的评估，而定性视觉方法在社区环境中可能更实用。
