The role of EEG and brain stem auditory evoked potentials (BAEPS) as predictors of outcome in severe brain injury

Objective

To determine the role of EEG and BAEPs to predict the outcome and to build a predictive model in patients suffering from severe head injury (SHJ).

Materials and methods

The prospective cohort study includes 102 patients admitted to a university ICU. The first EEG and BAEPs recordings were obtained within 48 hours of the trauma, followed by recordings after 2 days and later on the basis of the clinical evolution. For every patient the first Glasgow Coma Scale (GCS) and the worst CT scan during the first week were evaluated following Marshall classification. The patient's outcome was classified on the basis of the Glasgow Outcome Scale (GOS): 1) death or vegetative state, and 2) recovery (with different degrees of impairment). The EEG was classified on the basis of reactivity in three categories: 1) flat, 2) reactivity, and 3) no reactivity [1]. The BAEPs were classified on the basis of the Greenberg classification criteria and subsequently for statistical evaluation compacted on two classes: 1) present, and 2) absent [2]. First, each clinical and instrumental parameter was tested for discrimination, calculating the area under the receiving operator curve (ROC). Then, a predictive model was made using Stepwise Logistic Regression (SPSS 10.1). EEG, BAEPs, GCS, CT scan and age were the parameters tested to enter the model. Calibration was evaluated with the Goodness-of-Fit Hosmer–Lemeshow test and discrimination with the ROC curve.

Results

The mean age of the patients was 40.5 ± 20.3 and GCS was 7.4 ± 2.7. Seventy-eight of the 102 patients had a GOS = 2 (recovery). Parameters with good discrimination were: EEG (area = 0.888; ES = 0.051; P < 0.0001); GCS (area = 0.828; ES = 0.052; P < 0.0001); BAEPs (area = 0.765; ES = 0.065; P < 0.0001). Parameters that entered the model were: GCS (β_G = -0.404); EEG (β_El = 6.83004; β_E2 = -1.426608; β_E3 = -5.386608);BAEPs (β_Bl = 0.827; β_B2 = -0.827); constant = 5.640. This model was well calibrated (Hosmer–Lemeshow test = 8.91; P = 0.350) and had a good discrimination (area under the ROC = 0.981; P < 0.0001).

Conclusions

The different parameters considered have good discrimination ability to predict the outcome of the patients. The EEG seems to be the best parameter to predict the patient's outcome, even if all the patients were sedated. In our case studies, this did not seem to significantly influence the parameter of reactivity to sensory stimulus. Age seems not to be a reliable parameter to predict outcome.