Spreading Depolarization After Chronic Subdural Hematoma Evacuation: Associated Clinical Risk Factors and Influence on Clinical Outcome

Abstract

Background:

Chronic subdural hematoma (cSDH) is a common neurosurgical condition responsible for excess morbidity, particularly in the geriatric population. Recovery after evacuation is complicated by fluctuating neurological deficits in a high proportion of patients. We previously demonstrated that spreading depolarizations (SD) may be responsible for some of these events. In this study, we aim to determine candidate risk factors for probable SD and assess the influence of probable SD on outcome.

Methods:

We used two cohorts who underwent surgery for cSDH. The first cohort (n=40) had electrocorticographic (ECoG) monitoring to detect SD. In the second cohort (n=345), we retrospectively identified subjects with suspected SD based on the presence of transient neurological symptoms not explained by structural etiology or ictal activity on EEG. We extracted standard demographic and outcome variables for comparisons and modeling.

Results:

80/345 (23%) of subjects were identified in the retrospective cohort as having probable SD. Potential risk factors included history of hypertension, worse clinical presentation on GCS (Glasgow coma scale score), and lower Hounsfield Unit density and volume of pre-operative SDH. Probable SD was associated with multiple worse outcome measures including length of stay and clinical outcomes, but not increased mortality. On multivariable analysis, probable SD was independently associated with worse outcome determined by Glasgow outcome scale at first clinic follow up (OR=1.793, 95%CI 1.022-3.146), and longer hospital length of stay (OR=7.952, 95%CI= 4.062-15.563).

Conclusions:

Unexplained neurological deficits after surgery for cSDH occur in nearly a quarter of patients and may be explained by SD. We identified several potential candidate risk factors. Patients with probable SD have worse outcomes, independent of other baseline risk factors. Further data with gold standard monitoring is needed to evaluate for possible predictors of SD in order to target therapies to a high-risk population.

Introduction:

Chronic Subdural Hematoma (cSDH) is a distinct delayed post-traumatic pathology characterized by fluid and vascularized membrane formation in the subdural space, causing progressive mass effect on the brain and clinical symptoms. cSDH occurs most commonly in the elderly population and is expected to significantly increase in incidence over the next decade as this population continues to grow[1]. Neurosurgical drainage remains the standard therapy[2], but treatments targeting devascularization of the membranes are emerging as potential adjunct or alternative approaches[3].

Clinical symptoms of cSDH do not always directly correlate with the initial radiographic severity of cSDH and are often characterized by a fluctuating course rather than progressive deficits. Fluctuations of consciousness and neurological exam can occur in the post-operative period after cSDH evacuation [4], but many of these events lack structural and EEG correlates. Because these neurologic deficits often result in prolonged workup, increased length of hospital stay, pharmacological treatment, and additional surgical intervention, the identification of underlying mechanisms and possible treatments would have substantial benefit.

Spreading depolarizations (SD) are a plausible explanation for these waves of transient neurologic dysfunction[5],[6]. Levesque and colleagues characterized clinical criteria such as negative symptoms and longer duration or episodes as characteristics more consistent with “non-epileptic, stereotypical, and intermittent symptoms,” (NESIS) which they propose may in fact be clinical manifestation of SD rather than seizure[7] as events did not have an EEG correlate.

In a previously published study, we performed prospective subdural ECoG monitoring (the gold standard technique for assessing SD) in 40 subjects after cSDH evacuation and definitively documented SD in 15% of these patients [8]. In this study, we discovered that patients with documented SD had a higher rate of neurological deterioration and showed that there were clear time-locked focal deficits occurring with repeated SD. These data support our hypothesis that SD may the physiologic cause of the well-characterized fluctuating deficits in cSDH patients. Selective targeting of therapies to prevent or treat SD could have substantial benefit. We recently have presented preliminary data on both ketamine[9] and nimodipine[10] as potential therapeutic agents to target SD in other acute neurological injuries. However, identifying which cSDH patients would benefit from SD-targeted therapies is currently limited by the fact that invasive ECoG monitoring of SD is not feasible in most clinical centers and routine scalp EEG is limited in accurately identifying SD.

In the present study, we aim to determine whether routinely collected clinical data can serve as useful predictors of SD in cSDH patients. We then quantify the independent effect on outcome that probable SD has in order to plan and power future clinical trials.

Methods:

All patients included in this study underwent surgical evacuation for cSDH at our study institution. We used two cohorts for the current study. For the first cohort, we identified consecutive subjects undergoing surgery for cSDH entered into a registry collected between 2009 and 2016. For the second cohort, we then collected additional demographic and cSDH characteristic data from our previously published ECoG monitoring study called “CSD2”[8] (UNM HRPO# 16-352) in order to allow for comparison to the larger group. Demographic, radiographic, and outcome data were recorded. (UNM HRPO# 15-564). Subjects in both cohorts were compared between those with no SD and definite SD, as measured by ECoG recording, and probable SD, meaning no seizures detected on EEG[7,8]. Per chart review, EEG was ordered for patients who had a new neurological exam finding or failed to improve as expected post-operatively. Most subjects are treated with prophylactic antiseizure medications at our institution. For SD versus no SD comparisons, continuous variables were compared using t-tests, ordinal variables were compared using the Mann-Whitney test, and dichotomous variables were compared using Chi-square tests. These descriptive comparisons were done for demographics in both groups and for overall outcomes in the registry.

In order to determine the potential influence of probable SD on outcome in the larger cohort, we performed univariable logistic regression to identify factors associated with poor outcome as measured by GOS at first clinic visit and hospital length of stay. GOS was dichotomized into good recovery (5) versus anything less than good recovery. LOS was dichotomized at the median value (7 days). Significant factors on the univariable analysis were incorporated into a multivariable analysis. Significance was set as p<0.05 or 95% confidence intervals not overlapping 1.0. Analysis was performed using SAS 9.4 (Carey, NC) and GraphPad Prism (San Diego, CA). Subjects with incomplete data for a specific analysis were excluded from that analysis.

Results:

Demographics and outcomes of the two cohorts are described in table 1. There were 345 subjects in the registry. Probable SD occurred in 23% (n=80) of these patients. To confirm that these represented subjects with transient neurological deficits suspicious for SD, we reviewed the indication for EEG ordering. Overall indications for EEG included encephalopathy (52.8%), aphasia (17%), suspected seizure activity determined by new onset abnormal movements(19.8%), focal motor deficit (7.5%), and cranial nerve deficit (2.8%). All subjects either were given prophylactic levetiracetam (93%) or no medication (7%). The most common dose was 1000mg BID (51.6%). Either routine or long term EEG was used. Compared to the rate of definite SD in the CSD2 study (15%) the overall rate was slightly higher.

Table 1:

Baseline admission demographics of the two study cohorts, CSD2 (n=40) and cSDH registry (n=365). Four admission characteristics were found to be significantly associated with increased risk of definitive SD or suspected SD, hypertension, GCS, mean SDH density in Housefeld units, and volume of SDH.

Characteristics	cSDH registry (n=345)			CSD2 prospective study (n=40)
	No SD (n=265)	Probable SD (n=80)	p	No SD (n=34)	SD (n=6)	p
Age (mean)	70.20	72.24	0.214	71.50	73.33	0.739
Female Sex	67 (25.28%)	22 (27.50%)	0.691	10 (29.41%)	2 (33.33%)	0.847
Comorbidities
Diabetes n(%)	69 (26.04%)	28 (35.44%)	0.103	11 (32.35%)	1 (16.67%)	0.440
Hypertension n(%)	138 (52.08%)	54 (68.35%)	0.011*	18 (52.94%)	4 (66.67%)	0.533
Liver disease n(%)	14 (5.28%)	79 (10.13%)	0.123	0	0	-
EtOH abuse	30 (11.36%)	11 (13.75%)	0.564	3 (8.2%)	0	0.449
GCS (median)	15	14	0.015*	15.00	14.50	0.307
MGS (median)	1	1	0.137	2	1.5	>0.999
Antiplatelet use	79 (30.38%)	27 (34.62)	0.480	-	-
Anticoagulant use	36 (13.79%)	16 (20.51%)	0.149	9 (26.47%)	3 (50%)	0.246
Seizure at admission	12 (4.53%)	4 (5.06%)	0.843	1 (2.94%)	0	0.671
SDH characteristics
Bilateral hematoma	62 (23.4%)	20 (25%)	0.768	11 (32.35%)	2 (33.33%)	0.962
Density in Hounsfeld units (mean)	29.90	26.7	0.047*	38.53	39.12	0.886
Volume in mL (mean)	104.8	89.96	0.040*	229.2	190.8	0.7220
Membranes	151 (67.41%)	45 (68.18%)	0.906	1 (2.94%)	0	0.671
Midline shift in mm	6.216	6.012	0.716	5.748	4.417	0.344
Burrhole (versus craniotomy)	42 (15.85%)	12 (15%)	0.855	3 (8.82%)	0	0.449

There was a significantly higher rate of hypertension (p=0.011), and worse admission GCS (p=0.015) in those with probable SD. The volume of the cSDH was significantly smaller (p=0.04) and the Hounsfeld unit density was significantly lower (less acute component, p=0.05) in the probable SD group. We performed similar analyses using the data from the CSD2 study, to determine if these trends would be supported. Interestingly, the demographic data for these significant factors (hypertension, and GCS) trended in the same direction, but did not reach significance, likely due to small numbers. Similarly, volume was smaller in the definite SD group, though also non-significant.

Outcomes (Table 2, Figure) were worse in the subjects with probable SD across multiple measures including length of stay, ICU length of stay, discharge neurologic function, and GOS at the first clinic follow up. 30 day mortality, 2-3%, was not significantly affected.

Table 2:

Outcome measurements in the cSDH registry cohort (n=345) in patients with probable SD and no SD. Outcome measurements include total length of hospital stay, length of ICU stay, discharge Glasgow Coma Scale (GCS), discharge Markwalder Grading Scale (MGS), Glasgow Outcome Score (GOS) at first clinic follow up visit, and 30-day mortality. Probably SD is associated with statistically significant increase in mean hospital and ICU length of stay as well as significantly decreased functional outcome measures determined by GCS, MGS, and GOS. Suspected SD had no statistically significant effect on 30-day mortality.

Outcome	Retrospective cSDH cohort (n=345)
	No SD (n=265)	Probable SD (n=80)	p
Total stay duration (mean days)	8.545, n=264	14.00, n=80	<0.001*
ICU length of stay (mean days)	4.466, n=262	6.550, n=80	0.002*
Discharge GCS (mean)	14.62, n=256	14.01, n=78	0.001*
Discharge MGS (mean)	0.7287, n=258	1.128, n=78	<0.001*
GOS at first clinic followup (median)	5, n=240	4, n=70	0.001*
30 day mortality	6 (2.38%), n=246	3 (3.85%), n=75	0.488

Figure:

Comparisons between subjects with no SD and probable SD from the registry population in terms of Glasgow outcome score at first clinic visit (GOS) and overall hospital length of stay. Red dots are individual subject data and the black lines represent the means used for comparison between groups. Both comparisons demonstrated significantly worse outcomes in subjects with probable SD.

We focused on length of stay and GOS for logistic regression in order to determine the effect of baseline variables and probable SD on outcome. (Table 3). Probable SD was a significant predictor of worse outcome while controlling for other risk factors. Probable SD was associated with nearly double the length of stay (adjusted OR 7.952 [95%CI= 4.062- 15.567]) compared to baseline GCS (adjusted OR 3.771 [95%CI= 2.307- 6.163]).

Table 3: Univariable and Multivariable predictors of clinical outcome.

Multivariable and univariable analysis for logistic regression to determine the effect of baseline variables on two measures of clinical outcome, Glasgow Outcome Scale (GOS) and hospital length of stay. Variables associated with Glasgow Outcome Scale (GOS) at first follow up clinic visit are described in the top part of the table, with admission Glasgow Coma Scale (GCS) (p<0.0001, 95% CI 1.022-3.146), liver disease (p= 0.0207, 95% CI 1.213-10.377), and seizure on admission (p=0.0487, 95% CI 1.007-11.823) are significantly associated with worse GOS at first clinic visit. Probable SD was significantly associated with worse GOS at first clinic visit, (p=0.0416, 95% CI 1.022-3.146). Two variables were associated with increased length of hospital stay, probable SD (p<0.0001, 95% CI 4.062-15.567), and admission Glasgow Coma Scale (GCS) (p<0.0001, 95% CI 2.307-6.163).

Glasgow Outcome Scale at First Follow-up Clinic Visit
Covariates	Univariable Analysis			Multivariable Analysis
	OR	95% Conf. Interval		OR	95% Conf. Interval		P value
Age	2.236	0.914	5.468
Sex	1.040	0.635	1.702
Probable SD	2.359	1.415	3.933	1.793	1.022	3.146	0.0416
Admission GCS	4.301	2.731	6.774	3.999	2.487	6.430	<.0001
Diabetes	1.170	0.730	1.876
Hypertension	1.815	1.173	2.807	1.434	0.887	2.317	0.1415
Liver Disease	2.968	1.178	7.478	3.548	1.213	10.377	0.0207
EtOH Abuse	1.747	0.905	3.372
Anticoagulant Use	1.143	0.631	2.068
Antiplatelet Use	1.112	0.700	1.766
Seizure at Admission	2.991	1.016	8.804	3.450	1.007	11.823	0.0487
Bilateral Hematoma	0.632	0.378	1.057
Burrhole versus Craniotomy	0.471	0.160	1.388
Membranes on CT	1.300	0.791	2.136
Hospital Length of Stay
Covariates	Univariable Analysis			Multivariable Analysis
	OR	95% Conf. Interval		OR	95% Conf. Interval		P value
Age	2.236	0.914	5.468
Sex	1.071	0.657	1.744
Probable SD	8.164	4.355	15.306	7.952	4.062	15.567	<.0001
Admission GCS	4.301	4.076	2.595	3.771	2.307	6.163	<.0001
Diabetes	1.500	0.936	2.405
Hypertension	1.611	1.048	2.477	1.149	0.702	1.881	0.5814
Liver Disease	2.032	0.830	4.977
EtOH Abuse	1.191	0.620	2.287
Anticoagulant Use	1.480	0.817	2.682
Antiplatelet Use	0.973	0.614	1.542
Seizure at Admission	1.116	0.409	3.046
Bilateral Hematoma	1.084	0.660	1.781
Burrhole versus Craniotomy	1.929	0.602	6.178
Membranes on CT	0.883	0.540	1.445

Discussion:

Spreading depolarization has been well described for many years in the pre-clinical neurophysiology literature, but its relevance to human brain physiology has been poorly understood until recently. This is primarily because SD can only be definitively measured currently using invasive subdural DC or near DC ECoG recordings[11,12] since sensitivity of scalp EEG in detecting SD is poor. Recent data in traumatic brain injury and neurosurgical patients have definitively demonstrated that SD is a relatively common occurrence in acute neurological injury and is associated with worse outcomes[13,14]. SD has long been hypothesized to be the underlying mechanism of neurological deterioration in a number of other neurological conditions, such as slowly progressive visual scintillating scotoma in migraine. Blood flow imaging during migraine has demonstrated the characteristic spreading oligemia followed by compensatory hyperemia expected in SD[15,16]. In acute neurologic injury it is more difficult to assess the direct clinical correlates of SD due to the severity of the injury, however deterioration events have been closely temporally linked to SD[14].

Delayed fluctuating neurologic deficits associated with cSDH surgery are a clinically common and vexing problem[6]. These transient deficits typically are characterized by neurological deficits rather than activation-type events. These events are often loosely classified and treated as seizures for lack of an accurate, definitive diagnosis and are treated with antiseizure medication despite lack of EEG correlate[7]. Such events frequently result in additional workup with multiple repeated imaging studies, prolonged EEG, and occasionally, repeat surgery.

Building on our experience with SD monitoring after acute neurological injuries[17] and during neurosurgical procedures[18], we hypothesized that some of these events in cSDH could be related to SD and detected these events in 1-2 days of monitoring in 15% of subjects, though this may be an under-estimation of the true incidence[8]. In our current study, we identified 23% of subjects with deficits severe enough to prompt EEG assessment and we hypothesize that these subjects may be having SD as there were no electrographic seizure correlates on EEG. Unfortunately, we were not able to identify consistent risk factors for definite SD and probable SD across both groups, though lower GCS and hypertension (for unknown reasons) were identified as risk factors in the probable SD group and need further validation. Interestingly, the SDH was smaller and more chronic appearing in the subjects with probable SD. We hypothesize that there may be a lower threshold to take patients to surgery with a neurologic deficit even with smaller cSDH. This is very preliminary, but raises the possibility that some of the pre-op neurologic deficits could be related to SD and that such patients could potentially benefit from less invasive therapies such as targeted medication and embolization.

The data we present here demonstrate that in a large cohort of cSDH patients, those with probable SD had overall longer length of stay and worse overall outcome independent of admission demographics and radiographic severity of initial cSDH. Chart review of our cohort indicated that deleterious effects of SD in the study population were not related to new-onset post-operative structural damage such as ischemia, rebleed, or herniation. In elderly populations, however, which represent the majority of the cSDH population, the additional burden of being bedbound for several extra days, additional workup, unnecessary antiseizure medication administration, and additional surgical interventions can have clinically meaningful effects on outcomes. Our data here support that the strongest effect of probable SD is on increased length of stay (even more than admission severity) with an OR of nearly 8. Though overall outcomes were affected more by admission severity, as would be expected, probable SD still had an additional effect on GOS at first follow-up clinic visit. For these reasons, targeted effective treatments would be highly useful for this population.

The current primary targets for targeted therapy for acute SD are NMDA-R antagonists and L-type calcium channel antagonists. Ketamine is the strongest supported agent, with data ranging from “all or none” case reports[20] to retrospective series[21,17], to a pilot clinical trial[9]. Nimodipine also has been investigated with promising early results in both pre-clinical settings[22,23] and a pilot clinical trial[10]. It has been hypothesized that the beneficial effect of nimodipine in subarachnoid hemorrhage patients could in fact have been related to the effects on SD dependent ischemia[24]. Of note, medications targeting migraine aura may have some long-term efficacy against SD, but require prolonged treatment, so are not ideal candidates for acute conditions[25].

The data from this retrospective study have provided valuable background information to allow us to design an upcoming clinical study and trial with the goals of 1) validating and refining potential risk factors for SD using gold standard ECoG monitoring in a larger group of subjects; 2) exploring pilot feasibility, safety, and efficacy data using the NMDA-R antagonist memantine (IND# 1530765) (NCT# 04966546). Memantine is an attractive agent in this population since it is primarily used in the elderly dementia population with a proven safety record. In addition, there are preliminary pre-clinical data[26] and anecdotal clinical data supporting its efficacy for SD and neurological recovery in trauma[27].

This study has several limitations. First, we cannot yet confirm that the neurologic deficits we noted retrospectively were from SD. We believe that coupled with data from our gold standard monitoring group and other published data hypothesizing the clinical syndrome of SDm however that it is highly probable. Prospective observational data is strongly needed. In addition, we were not able to identify consistent risk factors for SD across the cohorts, likely due to the small sample size of subjects with definite SD; this offers potential for future study. Additionally, due to the small sample size in the definitive SD cohort, no significant differences were found in baseline admission characteristics between SD and no SD cohorts. Second, traditional scalp EEG is limited in the identification of possible seizures, due to temporal limitations between witnessed symptoms and EEG placement as well as a low sensitivity to detect SD. We additionally did not distinguish between routine EEG and long term EEG, the latter of which would increase the time to detect seizures. Additional EEG variables, such as interictal epileptiform discharges, were not considered as seizure category, and thus included in the “probable SD” cohort, which may afford a degree of bias, though technically, no definitive seizures were identified. Our definition for probable SD may or may not reflect true cases of SD, but is similar to the group identified by Levesque[7] with neurological deficits prompting EEG monitoring, but no correlating seizure activity. Due to the overlap of SD and seizure, it is also possible that some patients with seizure also had SD and some patients with interictal activity may not have had SD (cite Dreier, Brain, spreading convulsions article, 2012). More detailed neurological exam was not available to determine if the subjects met criteria for the “NESIS” syndrome suggested to be related to SD. Finally, the relationship between seizures and SD also may be inter-related and it is possible that some of the poor outcome we identified is related to effects from seizures that were not detected at the time of EEG. Our goal, however, is primarily to provide a foundational framework for ongoing and future trials.

Conclusions:

Unexplained neurological deficits after surgery for cSDH occur in nearly a quarter of patients and may be explained by SD. Admission characteristics such as neurological and radiographic severity may be associated with risk of SD and need to be validated in subjects with gold standard monitoring. Patients with probable SD have worse outcomes, independent of other baseline risk factors. These data provide important physiological insight into a common clinical problem. Further work focused on identifying this high-risk population and developing targeted therapies is needed for this large and growing condition.