Transcranial Doppler combined with quantitative EEG brain function monitoring and outcome prediction in patients with severe acute intracerebral hemorrhage

In this prospective study, we consecutively enrolled patients with ICH with coma who were admitted to the Department of Neurology, First Hospital of Jilin University, China, between June 2015 and December 2016. The patients were included in our study if they met the following criteria: admission time ≤ 72 h after onset; presence of supratentorial hemorrhage; and Glasgow Coma Scale (GCS) score ≤ 8 points on admission, as assessed by an experienced neurologist. Exclusion criteria were the following: ICH secondary to aneurysm, vascular malformation, tumor, and cerebral infarction; deficient temporal acoustic bone window; scheduled surgical treatments, including ventricular drainage, clot removal, and craniotomies; middle cerebral artery or other intracranial and extracranial major vascular stenosis/occlusion; previous ischemic or hemorrhagic cerebrovascular disease; pathological changes that affect intracranial EEG activity, such as intracranial infection, ischemia anoxic encephalopathy, cerebral trauma, and others; presence of marked environmental turbulence, such as low temperature (core body temperature < 32 °C), hypoglycemia (<50 mg/dl), hyponatremia (<116 mg/dl), and others; and central nervous system depressant use, including sedative, narcotic, antidepressants, antipsychotics, antiepileptic drugs, and others. Fifteen age and sex-matched healthy controls (64.3 ± 13.5 years old, eight men) were recruited. Written informed consent was obtained from each patient’s immediate family members before the beginning of the study. The study was approved by the Ethics Committee of the First Hospital of Jilin University, China, and conformed to the tenets of the Declaration of Helsinki.

All patients received routine monitoring of vital signs and intensive nursing care in the Neurological Intensive Care Unit. We recorded and analyzed the following variables: demographics (age and sex); stroke risk factors (hypertension, diabetes mellitus, dyslipidemia, atrial fibrillation, previous acute coronary event, smoking, excessive drinking); admission GCS score; time from ICH onset until the time monitoring commenced; whether blood pressure and glucose management were in accordance with current ICH management guidelines [18]; other admission laboratory tests (serum potassium, calcium, and sodium, white blood cell count, platelet count, activated partial thromboplastin time (APTT), International Normalized Ratio (INR)); neuroimaging variables (for regular hematoma, ICH volume was measured using length × width × depth/2; for irregular hematoma, length × width × depth/3 was adopted [19]—head CT with hematoma location and volume on admission was evaluated by a neuroradiologist blinded to the clinical and brain function data); and clinical outcome assessed using the 5-point Glasgow Outcome Scale score 90 days after ictus.

Patients were in the supine position during quantitative brain function monitoring. TCD was performed using 2-MHz pulsed-wave Doppler probes fixed to each temporal window with a helmet. The depth of acquiring optimal middle cerebral artery signals was 50–60 mm from both sides. Simultaneous EEG was obtained with standard 16-channel electroencephalography with silver chloride scalp electrodes placed in accordance with the 10–20 system; electrode impedance was maintained below 10 KΩ. The healthy control group remained with their eyes closed and awake during the process. Data were recorded for over 30 min until a stable recording was established; the recorded data were stored for further analysis.

Blinded analysis was performed for each patient. All clearly readable TCD waveforms were used in the calculations. The following variables were analyzed: systolic flow velocity (Vs), diastolic flow velocity (Vd), mean velocity (Vm), and PI from the affected and unaffected hemispheres. Vm was calculated using the following equation:

and PI was calculated using the following equation:

Offline QEEG analysis was performed using MATLAB (MathWorks, Natick, MA, USA). After data filtering (high pass 0.3 Hz, low pass 30 Hz) and all segments of artifact-free EEG were analyzed, spectral power was calculated using Fast Fourier transform (FFT) for each electrode over the 1–30 Hz range. The relative power of the delta (1–3 Hz, RDP), theta (4–7 Hz, RTP), alpha (8–13 Hz, RAP), and beta (14–30 Hz, RBP) frequency bands over all channels were used to calculate the global delta/alpha ratio (DAR) and the (delta + theta)/(alpha + beta) ratio (DTABR). The brain symmetry index (BSI) was calculated according to the following formula:

with the power of the signal obtained from a particular hemispheric bipolar channel pair i (with i = 1, 2, …, N) at frequency j (or Fourier coefficient, with index j = 1, 2, …, M), N being the number of channel pairs, M being the number of Fourier coefficients, Rij(t) and Lij(t) for the right and left hemisphere, respectively [20].

All statistical analyses were performed with SPSS version 17.0 (SPSS Inc., Chicago, IL, USA) and MedCalc version 11.4.4 (MedCalc Software, Mariakerke, Belgium). In the univariate analysis, data were reported as mean and SD for normally distributed variables and as median and interquartile range (IQR) for nonnormally distributed variables. Categorical variables were presented as percentages. Student’s t tests and median two-sample tests were used for normally distributed variables. Nonparametric Wilcoxon (Kruskal–Wallis) analysis of variance was used for nonnormally distributed variables. The comparison of categorical variables was performed with the chi-squared test. From these analyses, we could select the variables with some prognostic significance (p ≤ 0.001). Then, we performed a backward stepwise logistic regression analysis with death at 90 days as the dependent variable. If a TCD or QEEG measure was statistically associated with survival, we calculated its separated and united sensitivity, specificity, and area under the curve with the aid of receiver operator characteristic (ROC) curves. ROC curves were compared by applying DeLong’s test. Calculated two-tailed p < 0.05 was considered statistically significant.

During a 1.5-year period, 76 consecutive patients were diagnosed with severe, acute, spontaneous supratentorial ICH. The following patients were excluded: admission beyond 72 h from onset of symptoms (n = 4), surgical procedure (n = 2), macrovascular stenosis (n = 7), previous cerebrovascular disease (n = 5), deficient temporal window (n = 6), signal artifacts (n = 3), and loss to follow-up (n = 2). Finally, we enrolled 47 patients, of which 26 (55.3%) died during the 90-day follow-up period. The median age was 67.3 ± 12.6 years, and 23 (48.9%) of the patients were male. The first TCD-QEEG was performed a mean of 31.0 (19.0–59.0) h after the onset of symptoms. No statistically significant differences between survivors and nonsurvivors were noted for clinical baseline data, including age, sex, risk factors, blood pressure, serum glucose, serum potassium, calcium, and sodium, white blood cell count, platelet count, APTT, INR, and hematoma location. Only larger hematoma volume (p < 0.0001) and higher Glasgow Coma Scale score (p = 0.001) were associated with mortality (Table 1).

Figure 1 shows the CT and TCD-QEEG findings of representative patients.

Regarding TCD relevant indicators, Vd of unaffected hemispheres was lower in nonsurvivors (UVd p = 0.018) and the PI of both hemispheres was higher in nonsurvivors than survivors in patients with SAS-ICH (API p < 0.0001, UPI p < 0.0001) (Table 2 and Fig. 2b, c). Vm in the affected hemisphere and Vd and PI in both hemispheres showed significant differences (all p < 0.0001) between patients with ICH and healthy controls (Table 2 and Fig. 2a–c).

Regarding QEEG relevant indicators, higher relative delta power (RDP p < 0.0001), lower relative alpha power (RAP p < 0.0001), higher delta/alpha ratio (DAR p < 0.0001), and higher (delta + theta)/(alpha + beta) ratio (DTABR p = 0.002) were associated with mortality in patients with SAS-ICH. All indicators except BSI showed significant differences (all p < 0.0001) between patients with ICH and healthy controls (Table 2 and Fig. 2d, e). There were no significant differences in BSI between survivors and nonsurvivors or between patients and controls (all p > 0.05) (Table 2 and Fig. 2f).

All p ≤ 0.001 variables in the univariate analysis were entered into a logistic regression model with mortality at 90 days as the dependent variable. The result showed that only DAR (OR 5.306, CI 1.533–18.360, p = 0.008) and UPI (adjusted OR 2.373, CI 1.299–4.335, p = 0.005) were identified as independent predictors for mortality at 90 days after SAS-ICH. The results of the logistic regression analysis remained unchanged when we excluded the GCS score and hematoma volume to verify model stability.

To determine whether the combination of TCD and QEEG variations in the model improved outcome prediction, we compared the ROC curves of five models: the first model was obtained by the GCS score, the second model contained the hematoma volume, the third model was obtained by the independent predictors of TCD with UPI, the fourth model was obtained by the independent predictors of QEEG with DAR, and the final model included both UPI and DAR. All models could predict mortality in patients with ICH at 90 days. The final model attained a high prognostic power, as the area under the ROC curve was 0.949 when UPI and DAR were combined. The contribution of the final model was significant, while each variable alone did not seem to be significant (Fig. 3).