The Role of cEEG as a Predictor of Patient Outcome and Survival in Patients with Intraparenchymal Hemorrhages

Abstract

Purpose:

The objective of this study was to determine if continuous electroencephalography (cEEG) results are associated with functional outcome and survival in critically ill patients with intraparenchymal hemorrhages (IPH).

Methods:

Patients diagnosed with IPH were selected using a Critical Care EEG Monitoring Consortium Database at Brigham and Women’s Hospital in Boston. Functional Outcome in Patients with Primary Intracerebral Hemorrhage (FUNC) scores and Intracerebral Hemorrhage (ICH) scores were calculated as covariates. Outcomes of interest were functional outcome (modified Rankin scale [mRS] <3 vs ≥3) and mortality at hospital discharge. cEEG features, as defined by the American Clinical Neurophysiology Society standard terminology, were assessed for association with outcome after accounting for known clinical covariates.

Results:

A total of 94 patients admitted between March 2013 and December 2015 were selected. Multivariate regression analysis revealed that the presence of Stage II Sleep is independently associated with good functional outcome at discharge after correcting for FUNC (p=0.0080) and ICH (p=0.0088). The absence of anteroposterior (AP) gradient in an EEG is associated with increased likelihood of mortality at discharge after correcting for FUNC (p=0.013) and ICH (p=0.019) scores.

Conclusions:

cEEG measures were significantly associated with functional and mortality outcome measures in patients with IPH even after accounting for known clinical and radiological covariates. Further research is needed to determine whether prediction models are improved by inclusion of cEEG features.

Introduction

Continuous electroencephalography (cEEG) monitoring is frequently utilized during the care of critically ill patients. In addition to being used consistently to provide evidentiary support of diagnoses made primarily from imaging techniques, EEG has also been utilized as a source for assessing the outcome of a patient. However, the relationship between these cEEG features and clinical outcome is not clear and may strongly depend on the underlying pathology. In comatose patients after cardiac arrest, EEG is the strongest predictor of poor outcome [1]. In status epilepticus, the absence of a strong posterior dominant rhythm (PDR) is associated with mortality, and presence of Stage II Sleep is associated with return to baseline after controlling for clinical predictive factors [2]. In intracerebral hemorrhages, periodic discharges have been shown to be independently correlated to poor outcome [3]. In subarachnoid hemorrhage (SAH), the presence of sleep architecture is correlated with patient outcome [4]. In another small study of SAH patients, periodic and rhythmic patterns, though found frequently, did not correlate with outcome [5].

The American Clinical Neurophysiology Society (ACNS) standard terminology for critical care EEG monitoring [6] describes a standard set of nomenclature for rhythmic and periodic patterns as well as features of the EEG background. This allows for standardized assessment of cEEG features in the critically ill population with high interrater reliability [7]. The purpose of this study is to analyze the relationship between cEEG features, as described by the standard terminology, and clinical outcome, as measured by hospital discharge mortality and modified Rankin Scale (mRS) [8], in patients with intraparenchymal hemorrhage (IPH), while controlling for known clinical and radiological features.

Methods

Participants

This study includes all consecutive adult patients admitted to Brigham and Women’s Hospital in Boston between March 2013 to December 2015 who presented with intraparenchymal hemorrhages. The patient pool was comprised of retrospectively identified patients through the Critical Care EEG Monitoring Consortium (CCEMRC) database [9] at the Brigham and Women’s Hospital.

Inclusion criteria for the study consisted of adult patients above 18 years of age with a diagnosis of an IPH that was confirmed by imaging techniques such as CT. All patients included underwent standard cEEG monitoring during admission. Additionally, patients who presented with other major structural abnormalities as seen through their CT or MRI were excluded.

EEG Measures

Recordings were acquired using the international 10–20 system with 21 electrodes (XLTEK; Natus Medical Incorporated, San Carlos, CA, U.S.A.). Low and high frequency filters were set at 1 and 70 Hz respectively, and a notch filter was used as needed. All EEG recordings were reviewed by trained electroencephalographers who had passed a certification exam [7] using the 2012 ACNS Critical Care EEG terminology [6]. All EEGs were assessed for the presence of rhythmic, periodic, and background features. The former two were denoted using main term 1: generalized (G), lateralized (L), bilateral independent (BI), multifocal (M) and main term 2: periodic discharges (PD), rhythmic delta activity (RDA), spike/sharp-and-wave (SW). Background features included an assessment for presence of continuity, AP gradient, posterior dominant rhythm frequency, variability, reactivity. If the patient underwent a prolonged hospitalization with multiple EEGs, the study that was temporally closest to the time of clinical outcome assessment was utilized. A combined variable “Poor EEG Background” was determined to be present if the EEG either lacked continuity, posterior dominant rhythm, or an anteroposterior gradient.

Clinical Measures

Outcome at hospital discharge was assessed using the mRS [8] which assigns a score, 0–6, to patients based on the physical and functional outcome as determined by their hospital records. While it was initially introduced to assess post-stroke functionality, its reliability, feasibility and cost-effective nature has allowed for its integration into many other fields of neurology, and has been previously utilized in other studies to quantify and binarize outcomes [4,5,10]. In our study, mRS outcome was binarized as good outcome vs poor outcome with mRS 0–2 (ranging from no symptoms to slight disability) vs mRS 3–6 (ranging from moderate disability to death), respectively. In order to assign an appropriate mRS score to each patient, we conducted a thorough review of hospital records, discharge summaries, patient-care referral forms, physical therapy notes, and general progress notes.

We utilized validated measures for outcome (the FUNC [11] and ICH [12]) to further adjust for variations in hemorrhage size and other confounding variables. The Functional Outcome in Patients with Primary Intracerebral Hemorrhage (FUNC) score represents probability of attaining functional independence at 90 days after discharge on a scale of 0–11, where 11 denotes functional independence [11]. The intracerebral hemorrhage score (ICH) represents likelihood of mortality at 30 days on a scale of 0–6 where a 6 represents 0% mortality [12]. Outcome predictors FUNC and ICH scores were calculated based on Glasgow Coma Score (GCS), age, volume of hemorrhage, ICH location (lobar, deep, infratentorial), presence of intraventricular extension, and pre-ICH cognitive impairment. The volume of hemorrhage was measured using CT scans and an image processing software (3D Slicer) in which a hemorrhage can be delineated in slices and summed to produce a volume [13,14].

Statistical Analyses

Variables for the final models were selected using a backwards elimination automated selection technique. We included variables in the backwards selection model based on a two-step screening process. First, each potential EEG variable was placed in a univariate logistic regression model with only the functional outcome (“good functional outcome” (mRS < 3) and death, separately). Variables that demonstrated a statistically significant relationship at the α = 0.05 level were then considered in the next step of screening. This second step paired each significant variable with FUNC score and ICH score separately, with the outcome for which statistical significance was already established at the first screening step. From this second step, variables that maintained statistical significance when paired with FUNC score or ICH score were included in the corresponding model for backwards selection, of which there were four: an outcome of “good functional outcome” with predictors of FUNC score and relevant significant variables, an outcome of “good functional outcome” with predictors of ICH score and relevant significant variables, an outcome of death with predictors of FUNC score and relevant significant variables, and an outcome of “death” with predictors of ICH score and relevant significant variables. It should be noted that due to a relatively small sample size and our use of binary variables, we had issues with separation [15]. We corrected for this separation by using the Firth method in all logistic regression models (all models) [16]. Calculations were performed using R 3.3.3 (R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Specifically, the use of the Firth method was implemented using the package logistf (Georg Heinze and Meinhard Ploner (2016); logistf: Firth’s Bias-Reduced Logistic Regression; R package version 1.22).

Results

Study Population

A total of 94 consecutive patients were enrolled in the study (Table 1). No radiographic imaging data was found for five of the 94 patients and thus, this subset consists of the remaining 89 IPH patients. EEG data for these patients categorized by their mRS is shown in Table 2. The FUNC and ICH scores were calculated for each patient using available patient and imaging data. The average ICH score amongst this population was 2 while the average FUNC score was 7. The associations between a) mRS score vs ICH score and b) mRS score vs FUNC score is shown in the Figure and are reflective of their presupposed relationship. The average volume of IPH was found to be 36 cm³. Sixteen patients’ (18.0%) hemorrhages exceeded 60 cm³. Twelve patients’ (13.5%) hemorrhages were between 30–60 cm³. Sixty-one patients (68.5%) had IPH with a volume of less than 30 cm³. Nine patients (10.1%) had a GCS score of 3–4, thirty-five (39.3%) were found to have a score of 5–12, and forty-five (50.6%) were found to have a score of 13–15. Sixty-two (69.7%) patients had a pre-ICH cognitive impairment. Thirty-eight (42.7%) showed presence of IVH as seen in their CT and/or MRI. Four patients (4.5%) had hemorrhages that had an infratentorial origin.

Table 1.

Patient Demographics

	IPH Patients (n=89)
Women, n (%)	40 (44.9)
Mean Age (std)	67.83 (14.6)
Mean mRS (std)	4.3 (1.4)
mRS, n (%):
Score 0: No symptoms at all	0 (0.0)
Score 1: No significant disability despite symptoms; able to carry out all usual duties and activities	6 (6.7)
Score 2: Slight disability; unable to carry out all previous activities, but able to look after own affairs without assistance	3 (3.4)
Score 3: Moderate disability; requiring some help, but able to walk without assistance	15 (16.9)
Score 4: Moderately severe disability; unable to walk without assistance and unable to attend to own bodily needs without assistance	16 (18.0)
Score 5: Severe disability; bedridden, incontinent and requiring constant nursing care and attention	33 (37.1)
Score 6: Dead	16 (18.0)
Duration cEEG (hrs) (std)	37.02 (35.5)
Seizures on cEEG, n (%)	9 (10.1)

IPH = intraparenchymal hemorrhage

Table 2.

EEG characteristic of interest per mRS group

	mRS 0 (n=0)	mRS 1 (n=6)	mRS 2 (n=3)	mRS 3 (n=15)	mRS 4 (n=16)	mRS 5 (n=33)	mRS 6 (n=16)
EEG Features
AP Gradient, n (%)	0 (0)	5 (83)	2 (67)	10 (67)	8 (50)	18 (55)	3 (19)
PDR, n (%)	0 (0)	5 (83)	1 (33)	2 (13)	4 (25)	6 (18)	0 (0)
Continuity, n (%)	0 (0)	6 (100)	3 (100)	12 (80)	12 (75)	20 (61)	9 (56)
Stage II Sleep, n (%)	0 (0)	6 (100)	3 (100)	8 (53)	10 (63)	12 (36)	3 (19)
LPD, n (%)	0 (0)	0 (0)	0 (0)	2 (13)	2 (13)	3 (9)	4 (25)
GPD, n (%)	0 (0)	0 (0)	0 (0)	0 (0)	5 (31)	8 (24)	4 (25)
LRDA, n (%)	0 (0)	0 (0)	1 (33)	1 (7)	2 (13)	3 (9)	1 (6)
GRDA, n (%)	0 (0)	0 (0)	0 (0)	2 (13)	4 (25)	1 (3)	1 (6)
LSW, n (%)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
GSW, n (%)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
Seizure, n (%)	0 (0)	0 (0)	0 (0)	0 (0)	3 (19)	1 (3)	2 (13)
Median ICH Score	0	0.5	0	1	1	2	3
Median FUNC Score	0	9.5	11	8	4	6	4

AP gradient = anteroposterior gradient; PDR = posterior dominant rhythm; LPD = lateralized periodic discharges; GPD = generalized periodic discharges; LRDA = lateralized rhythmic delta activity; GRDA = generalized rhythmic delta activity; LSW = lateralized spike/wave; GSW = generalized spike/wave

Figure:

Scatterplot of a) mRS versus ICH; b) mRS versus FUNC, with size and color of the point indicating larger populations.

EEG

The median duration of cEEG was 24.0 hours for each patient. Forty-six patients underwent 12–24 hours of recording, thirty-one patients underwent cEEG recording for 24–48 hours, and thirteen patients carried on with cEEG for more than 48 hours. Only one patient showed epileptiform discharges on their cEEG who later died during hospitalization (mRS 6). Only one patient had experienced seizures prior to their cEEG who also died during their hospital course and was deemed to have an mRS score of 6.

Univariate and Multivariate Results.

Good Outcome vs Poor Outcome

In univariate analysis, the presence of PDR (p=0.0011), continuity (p=0.032), Stage II Sleep (p = 0.00067), and lack of poor EEG background (p=0.0012) were independently associated with better functional outcome as per mRS in ICH patients (Table 3). Additionally, FUNC and ICH scores were used in regression models to verify their presupposed correlation with outcome as per mRS, and were found to be significant: ICH score (p<0.0001) and FUNC score (p=0.0013). When controlling only for known non-EEG confounders with FUNC score, PDR (p=0.0095), Stage II Sleep (p=0.0080), and Poor EEG Background (p=0.0083) remained significant. The same three components also remained statistically significant when utilizing ICH score as a covariate: PDR (p=0.022), Stage II Sleep (p=0.0088), and Poor EEG Background (p = 0.022). A multivariate analysis was performed including with PDR, Stage II Sleep, and Poor EEG Background, and with either FUNC or ICH scores. In these models, Stage II Sleep remained statistically significant for good outcome of all other EEG parameters in both the FUNC (p = 0.0080) and ICH (p = 0.0088) scores while PDR and Poor EEG Background were no longer significant for either model.

Table 3.

Correlating EEG Features to Binarized Outcome

	Good Outcome: mRS 0–2 (n=9)	Poor Outcome: mRS 3–6 (n=80)	OR (95% CI)	Univariate p value	OR (95% CI)	Multivariate p value
EEG Features
AP Gradient, n (%)	7 (78)	39 (49)
PDR, n (%)	6 (67)	12 (15)	10.2 (2.5, 47.9)	0.0011
Continuity, n (%)	9 (100)	53 (66)	9.8 (1.2, 1276.3)	0.032
Stage II Sleep, n (%)	9 (100)	33 (41)	25.6 (3.1, 33343.0)	0.00067	FUNC: 17.0 (1.8, 2295.6) ICH: 16.1 (1.8, 2126.5)	FUNC:0.0080 ICH: 0.0088
LPD, n (%)	0 (0)	11 (14)
GPD, n (%)	0 (0)	17 (21)
LRDA, n (%)	1 (11)	7 (9)
GRDA, n (%)	0 (0)	8 (10)
LSW, n (%)	0 (0)	0 (0)
GSW, n (%)	0 (0)	0 (0)
Seizure, n (%)	0 (0)	6 (8)
Median ICH Score	0	2	0.2 (0.1, 0.5)	<0.0001	0.2 (0.1, 0.6)	0.017
Median FUNC Score	10	7	1.8 (1.3, 3.0)	0.0013	1.6 (1.1, 2.8)	0.0010

AP gradient = anteroposterior gradient; PDR = posterior dominant rhythm; LPD = lateralized periodic discharges; GPD = generalized periodic discharges; LRDA = lateralized rhythmic delta activity; GRDA = generalized rhythmic delta activity; LSW = lateralized spike/wave; GSW = generalized spike/wave

Mortality

When outcome was binarized to death (mRS = 6) vs survival at hospital discharge (mRS 0–5), AP gradient (p=0.0036), the presence of a PDR (p=0.019), the presence of Stage II Sleep (p=0.0089), and Poor EEG Background (p=0.025) were significant in a univariate logistic analysis (Table 4). AP gradient remained significantly associated with survival at discharge when corrected for known non-EEG confounders with FUNC score (p=0.013) or ICH score (p=0.019). Stage II Sleep only remained significant for survival after correcting for FUNC (p=0.043) but fell out of the model when ICH score was utilized as a covariate. Multivariate analysis revealed that AP gradient remained statistically significant for mortality in both the FUNC (p=0.013) and ICH (p=0.019) scores as compared to the remainder of EEG parameters.

Table 4.

Correlating EEG Features to Binarized Mortality

	Alive (n = 73)	Dead (n = 16)	OR (95% CI)	Univariate p value	OR (95% CI)	Multivariate p value
EEG Features
AP Gradient, n (%)	43 (59)	3 (19)	0.2 (<0.1, 0.6)	0.0036	FUNC: 0.2 (<0.1, 0.7) ICH: 0.2 (0.1, 0.8)	0.013 (FUNC) 0.019 (ICH)
PDR, n (%)	18 (25)	0 (0)	0.1 (<0.1, 0.7)	0.019
Continuity, n (%)	53 (73)	9 (56)
Stage II Sleep, n (%)	40 (55)	3 (19)	0.2 (0.1, 0.7)	0.0089
LPD, n (%)	7 (10)	4 (25)
GPD, n (%)	13 (18)	4 (25)
LRDA, n (%)	7 (10)	1 (6)
GRDA, n (%)	7 (10)	1 (6)
LSW, n (%)	0 (0)	0 (0)
GSW, n (%)	0 (0)	0 (0)
Seizure, n (%)	4 (5)	2 (13)
Median ICH Score	2	3	2.1 (1.3, 3.7)	0.0036	1.8 (1.1, 3.2)	0.021
Median FUNC Score	8	4	0.7 (0.6, 0.9)	0.0014	0.7 (0.6, 0.9)	0.0049

AP gradient = anteroposterior gradient; PDR = posterior dominant rhythm; LPD = lateralized periodic discharges; GPD = generalized periodic discharges; LRDA = lateralized rhythmic delta activity; GRDA = generalized rhythmic delta activity; LSW = lateralized spike/wave; GSW = generalized spike/wave

Discussion

The main findings of this study are that of all cEEG characteristics in IPH patients, 1) the presence of Stage II Sleep is independently associated with good outcome of a patient at discharge; 2) the presence of an AP gradient was independently associated with mortality at discharge, after correcting for non-EEG confounders through either the ICH or FUNC scores. Patients exhibiting EEGs with strong Stage II Sleep features had an increased likelihood of lower mRS scores and better functional recovery at discharge. This has been shown to be true in previous studies assessing functional outcome in other neurological conditions such as subarachnoid hemorrhage and status epilepticus [2,4]. It appears to hold significant value as an EEG feature associated with clinical outcome, even more so than PDR, poor EEG background, and continuity which lose their association value after adjusting for other known confounders.

Additionally, it was found that the presence of AP gradient, a background feature, in EEGs of IPH patients is independently associated with lower mortality after correcting for known confounders. A previous study in status epilepticus found that the presence of a posterior dominant rhythm, another background feature, was associated with lower mortality [2].

EEG oscillations, including the posterior dominant rhythm [17] and the sleep spindles [18] demonstrate the integrity of the thalamocortical system by representing the robust connections between the neocortex and the thalamus. Additionally, a previous study [19] found that EEG background organization and sleep architecture are strongly associated with preserved cognition assessed and brain metabolism in patients with disorders relating to consciousness. Those findings, in conjunction with those of our present study, suggest that the presence of a robust background and sleep architecture serve as clinical surrogates of the thalamocortical system integrity and validate EEG as a useful biomarker in the field of epilepsy [20].

Our results are in agreement with Alvarez et al.’s study of patients presenting with status epilepticus in that the background features are critical in determining mortality at time of hospital discharge [2]. In their study, the presence of a PDR was associated with decreased mortality; in our study, the presence of an AP gradient, another feature of the EEG background, was also associated with decreased mortality. This associative trend between AP gradient and mortality was consistent in our cohort which adjusted for clinical and radiological variables through the FUNC and ICH scores. Additionally, Alvarez et al. found return to previous baseline to be associated with the presence of Stage II Sleep. This was also found in the current study, where the presence of Stage II Sleep was associated with better functional outcome as measured by mRS.

Both studies found that the presence of rhythmic or periodic patterns were not associated with either mortality or functional outcome, after correcting for clinical confounders [2]. However, other studies have found that the presence of periodic lateralized periodic discharges (PLEDS) was strongly correlated to poor outcome in patients with subarachnoid hemorrhage [4]. Jadeja et al. found that patients with generalized periodic discharges (GPDs) were associated with a constellation of poor clinical outcomes such as dementia, poor mental status during EEG, focal radiological abnormalities that were chronic, cardiac arrest, and chronic obstructive pulmonary disease [21].

Other studies have reported that rhythmic and periodic patterns (RPPs) or seizures may be useful in outcome prediction. A study by Jaitly et al. correlated the presence of a burst-suppression pattern and lateralized periodic discharges (LPDs) after status epilepticus ictal discharges with increased morbidity and mortality [22]. However, for the purposes of their study, EEG features were divided up into seven different groups for assessment instead of the detailed classification utilized in this study [7] that allowed for the analysis of each unique feature of the cEEG.

Another retrospective analysis of 50 SE cases [23] showed that periodic discharges (PDs) were the only EEG feature associated with outcome. That said, their findings were based on univariate analysis that did not adjust for etiology or clinical characteristics. Moreover, this study focused on RPPs without assessing EEG background.

There is also much variation in the results regarding the role and value of RPPs in outcome prediction that depends on the patient population. Amongst comatose patients, there was no association found between PD and mortality [24], even when PDs were prolonged [25]. That said, the presence of PDs was a strong predictor of patient outcome in the setting of CNS infection [26] or intracerebral hemorrhage (ICH) [3]. The fact that PDs have been found to impact outcome in cohorts with specific brain pathology, such as CNS infection or ICH, and insignificant when assessing critically ill patients in general, reinforces the importance of accounting for the underlying etiology.

This study has multiple strengths. We utilized the American Clinical Neurophysiology Society’s (ACNS) Standardized Critical Care EEG Terminology [6] to assess cEEG features in a uniform and consistent manner. This method has repeatedly been shown to have high inter-rater agreement for major cEEG features, thus making EEG interpretations easily replicable [2,5,7]. Similarly, we adhered to the standards of the Modified Rankin Scale (mRS) to quantify functional outcome of patients at discharge. Its scores are easily understood and create little room for bias when interpreting global disability. This research tool has been used by others [8,27] to assess post stroke functionality and has also been reported to have high inter-rater agreement due to its rigorously defined scores; there is a great deal of evidence attesting to its reliability and clinical relevance. Our study performed careful neuroimaging analysis and incorporated known independent variables for outcome. We utilized image processing software to measure hematoma volumes to a great degree of precision as compared to other methods [13,14,28] that utilized the ABC/2 formula. Calculating ICH volume using the ABC/2 formula has been shown to have a significantly high estimation error of ICH volume as compared to similar image processing utilized in this study [28]. This volumetric assessment was then used in conjunction with a number of other demographic, clinical, and radiological confounding variables to determine FUNC and ICH scores which have been shown to reliably predict functional capacity and stratify risk at 90 and 30 days, respectively [11,12]. Including FUNC and ICH to correct for known variables further increases the validity of the mRS and cEEG results.

The limitations to this study included the relatively small population size from a single center. Further research will have to be done on a larger scale to confirm the findings that this study presents. Additionally, there was an inevitable time gap between EEG recording and patient discharge from where we extrapolated functional outcome. This interval varied slightly on a case by case basis. Nevertheless, we did our best to temporally match the EEG to mRS at discharge. Neither FUNC nor ICH were validated for functional outcome or mortality at hospital discharge. However, these represent the best validated assessments of outcome, and the fact that our results were duplicated by using either of these measures is reassuring. Finally, mRS was dichotomized as 0–2 (no symptoms to slight disability) vs 3–6 (moderate disability to death). Other studies have dichotomized to 0–3 vs 4–6 [4]. However, we feel that separation between excellent versus suboptimal outcome is better characterized by our binarization method. We repeated the analysis utilizing 0–3 vs 4–6 and obtained comparable results, with the same significant variables (not shown). Patients with small IPH may not have undergone cEEG monitoring. However, it is likely that the conclusions from this study may be strengthened further by the addition of such patients.

Conclusion

This study reports that cEEG features, particularly AP gradient and Stage II Sleep, is associated with mortality and functional outcome at hospital discharge in patients with intraparenchymal hemorrhage, even after accounting for other known clinical and radiographic features. Further studies are needed to support these findings. Our results suggest that a prediction model incorporating EEG characteristics may outperform models that do not incorporate electrophysiological data. Several studies now have reported that Stage II sleep has strong association with functional outcome. Although the causative role has not yet been determined, regulation of sleep in-hospital may be an intervention to improve functional outcome.

Highlights.

• EEG data could predict functional outcome in IPH patients at hospital discharge

• Stage II Sleep is associated with better functional outcome at hospital discharge

• Absence of AP gradient is associated with increased mortality at hospital discharge

• Validated clinical measures like FUNC and ICH may be improved by including EEG data