Epileptiform activity in traumatic brain injury predicts post-traumatic epilepsy

Abstract

We hypothesize that epileptiform abnormalities (EA) in the electroencephalopgram (EEG) during the acute period following traumatic brain injury (TBI) independently predict first-year post-traumatic epilepsy (PTE₁). We analyzed PTE₁ risk factors in two cohorts matched for TBI severity and age (n=50). EA independently predict risk for PTE₁ (OR 3.16[0.99 11.68]); subdural hematoma is another independent risk factor (OR 4.13 [1.18 39.33]). Differences in EA rates are apparent within 5 days following TBI. Our results suggest increased EA prevalence identifies patients at increased risk for PTE₁, and that EA acutely post-TBI can identify patients most likely to benefit from anti-epileptogenesis drug trials.

Introduction

Severe brain trauma is a leading cause of death and disability in adults and children worldwide¹. Post-traumatic epilepsy (PTE) is one of the most disabling complications in survivors and can be difficult to treat². PTE rates are reported in up to 20% of patients, with increased risk based on brain injury severity, surgical intervention, time since traumatic brain injury (TBI) and younger age^3–6.

While some risk factors are known, we need to better stratify patients at highest risk for PTE to better understand anti-epileptogenesis and develop therapeutic agents While there is great interest in interventions to prevent post-TBI epileptogenesis, clinical trials have been plagued by financial and logistical barriers, with estimates upwards of $20M^4,7. Efforts to prevent epileptogenesis would be greatly aided by identification of acute biomarkers that identify patients at high risk for developing PTE, thus enriching the population eligible for clinical trials in a cost effective manner⁸.

Epileptiform abnormalities (EA), which include sporadic epileptiform discharges (spikes and sharp waves), periodic epileptiform discharges, and rhythmic patterns, are common following all types of acute brain injury, including TBI⁹. Recent work from our group suggests that EA predict risk for secondary brain injury (Kim et al. 2017) and acute seizures^10,11. TBI serves as an excellent acute brain injury model in which to investigate the role of EA as a marker of, and possible contributor to, secondary morbidity in the form of PTE. We aimed to determine whether EA could be used as an early biomarker of elevated risk for PTE₁. Such information could be used to specify subpopulations of TBI patients that would benefit from targeted trials of anti-epileptogenic interventions with reduced cost and adverse risk exposures.

Methods

We evaluated EEG reports and medical records from 50 patients with TBI at a tertiary care center (Massachusetts General Hospital Neurosciences and Surgical ICUs) who met study inclusion criteria between 2011 and 2015. Inclusion criteria were: age ≥18 years, TBI on presentation and EEG monitoring during the initial hospital admission for TBI. Retrospective collection and analysis of clinical data were performed under a protocol approved by the local institutional review board. Among patients meeting the inclusion criteria we first evaluated consecutive (based on hospital admission) cases to identify 25 who developed PTE₁ (defined below), and subsequently evaluated consecutive cases to identify 25 controls without PTE₁, matched by age and admission Glasgow Coma Scores (GCS).

EEG recordings and report review

EEG data was recorded using conventional 10–20 scalp electrode placement. EA were classified according to standardized nomenclature¹² as: seizures, sporadic epileptiform discharges (EDs), lateralized or generalized periodic discharges (LPDs and GPDs) and lateralized rhythmic delta activity (LRDA)¹³. We also analyzed generalized rhythmic delta activity (GRDA), polymorphic generalized and focal slowing but consider these separate from EA. The presence (dark bars) or absence (light bars) of these abnormalities, as documented in daily clinical EEG reports, was tallied for each patient with “day of traumatic brain injury” marked as day 0 (Figure 1A). A histogram representing the EEG distribution is shown in Figure 1B.

Figure 1.

EEG recording distribution and prevalence in PTE₁ (red) and non-PTE₁ (blue) patients. A) EEG recording days (colored boxes) plotted for individual patients plotted along y-axis based on TBI severity. Shading based upon presence (dark) or absence (light) of EA during that day’s recording. B) Histogram summarizing the proportion of EEGs during each 5 day time-period. C) Prevalence of EA in PTE₁ and non-PTE₁ groups. D) Prevalence of EA subtypes in PTE₁ and non-PTE₁ groups. E) Cumulative probability of the first appearance EDs in recordings up to the first 10 days post-TBI. F) Models of possible causal relationship between SDH, EA and PTE₁. Model (1) with dotted box outline is the most likely model based upon logistic regression.

PTE₁ definition

Patients with at least one seizure 2–12 months post-TBI, based on medical record review. Control subjects were patients meeting the inclusion criteria who had TBI without any documented seizures in the same period, matched for age and admission GCS (Table 1). Patients were excluded if there were insufficient follow up visits in the electronic health record to determine PTE₁ status. For practicality, we analyzed up to 12 months, the highest risk period¹⁴, while acknowledging this does not fully capture eventual PTE development.

Table 1.

Demographic predictors of PTE₁ development.

Univariate Analysis	PTE (n=25)	No PTE (n=25)	OR ([95% CI])	P-value
Age	52.5±20.4	49.6±25.8	1.01 [0.98 1.03]	0.36
Gender: F/M	10/15	8/17	1.42 [0.41 4.85]	0.25
IPH	8 (32%)	3(12%)	3.45 [0.77 >100]	0.06
SDH	21 (84%)	15(60%)	4.13 [1.18 39.33]	0.02*
SAH	14 (56%)	12(48%)	1.38 [0.40 4.66]	0.27
EDH	5 (20%)	2(8%)	2.88 [0.41 >100]	0.10
Skull Fracture	14 (56%)	10(40%)	1.91 [0.62 6.73]	0.12
Admission GCS	6±4.66	8±4.53	0.94 [0.81 1.07]	0.25

Key: IPH-intraparenchymal hemorrhage, SDH- subdural hemorrhage, SAH-subarachnoid hemorrhage, EDH-epidural hemorrhage, GCS-Glasgow coma scale

Data analysis

For data analysis we used Matlab, including the Matlab Statistics Toolbox (MathWorks; Natick, MA). We employed univariate and multivariate logistic regression to calculate odds ratios of the reported demographic or EEG features (candidate predictor variables for PTE₁). In addition to evaluating EA as a group, we analyzed individual EA subtypes (seizures, EDs, LPDs, GPDs and LRDA). Bootstrapping was used to determine 95% confidence intervals and p-values, with a significance threshold of p≤0.05.

Results

Demographic predictors

We calculated associations between demographic variables and PTE₁, including age, gender, admission Glasgow coma scale (GCS), presence of intraparenchymal hemorrhage (IPH), subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH), or epidural hemorrhage (EDH) (Table 1). The only demographic variable significantly associated with PTE₁ development is subdural hemorrhage (p=0.02, Table 1).

EEG Distribution

EEG acquisition days are shown for each individual and summarized for each cohort (Figure 1A,B). The PTE₁ group has more days of EEG monitoring overall, possibly attributable to the continuation of EEG monitoring when epileptiform abnormalities were found.

EEG predictors

EA are more common in patients with PTE₁ compared to patients without PTE₁ (64% vs 36%; p = 0.04) (Figure 1C). The prevalence of each EA subtype is shown in Figure 1D. EA are significant predictors of PTE₁ with an odds ratio of 3.16 [0.99 11.68] (p=0.04) (Table 2).

Table 2.

EEG predictors of PTE₁ development.

Univariate Analysis	OR ([95% CI])	P-value
Epileptiform Abnormalities	3.16 [0.99 11.68]	0.042*
Early Seizures	2.95 [0.80 24.42]	0.06
EDs	4.57 [1.60 21]	0.007*
LPDs	2.85 [0.71 >100]	0.10
GPDs	2.09 [<0.1 >100]	0.25
LRDA	1.88 [0.48 8.80]	0.19
GRDA	0.43 [<0.1 2.30]	0.87
Generalized slowing	3.27 [<0.1 >100]	0.15
Focal slowing	2.67 [0.97 10.1]	0.04*
Multivariate Analysis
EAs adjusted for SDH	2.97 [0.91 14.18]	0.03*
EDs adjusted for SDH	3.8 [1.18 18.96]	0.016*

Key: EDs-epileptiform discharges, LPDs-lateralized periodic discharges, GPDs-generalized periodic discharges, LRDA-lateralized rhythmic delta activity, GRDA- generalized rhythmic delta activity, EAs-epileptiform abnormalities

When evaluating individual EA subtypes, only EDs (p=0.01) are significantly associated with PTE₁ (Table 2). While not classified as an EA, focal slowing (p=0.04) is also significantly associated with PTE₁ (Table2). Early seizures and LPDs show positive associations with PTE₁ but did not reach significance (p=0.06 and p=0.10, respectively), probably due to small sample size.

The difference in EDs is observed early, ≤5 days after TBI, with 50% occurring on day 0 after TBI (OR 3.67 [1.02 18.76], p=0.04; Figure 1E).

Multivariate analysis

Controlling for SDH, acute EA remains significantly associated with subsequent PTE₁ (OR 2.97 [0.91 14.18], p=0.03; Table 2). EDs alone, after adjusting for SDH, have an even stronger effect with an adjusted odds ratio of 3.8 [1.18 18.96] (p=0.016, Table 2).

By comparing multivariate logistic regression models of SDH + EA with the univariate regressions of EA and SDH independently, we find that SDH and EA independently contribute to increased PTE₁ risk (p=0.05 and p=0.03, respectively) without any direct relationship to each other (p=0.17), suggesting that model (1) is the most likely relationship between the variables: SDH and ED are independent causal factors for PTE₁ (Figure 1F)

Discussion

Our results provide novel evidence that EA may be a useful marker in identifying patients at high risk for PTE₁.

SDH and PTE₁

SDH is significantly associated with PTE₁ in our study, in concordance with multiple other studies^15,16. Prior studies also find associations with other variables, such as post-TBI amnesia, alcohol and midline shift, which we did not assess¹⁶. Intraparenchymal hemorrhage and skull fractures are also associated with PTE₁ in other studies¹⁵, and while neither odds ratio in our cohort is significant (p=0.06 and 0.12 respectively), we are underpowered to detect such associations.

EA and PTE₁

While the presence and prevalence of EA after TBI has been described¹⁷, the association with PTE₁ has not been reported. Our results demonstrate that the presence of EA following TBI signals increases risk for the development of PTE₁. More specifically, EDs are associated with PTE₁ development. Other subtypes of EA in our data, including early seizures and LPDs, show weak associations with PTE₁ but do not reach statistical significance, potentially due to small sample size. Interestingly, focal polymorphic slowing is also significantly associated with PTE₁. While often considered a non-specific EEG pattern, focal slowing has been observed in PTE previously¹⁸ and a recent study showed focal slowing in areas corresponding to blood-brain barrier (BBB) disruption after TBI, which correlated with PTE₁¹⁹.

Our results also show that EA occur early (<5 days) after TBI, suggesting early EEG could be a useful diagnostic tool to assess TBI patients for PTE₁ risk. TBI is a defined time-point event in which patients are known to be at risk for epileptogenesis, thus making this group prime for anti-epileptogenesis trials. However, the large patient numbers needed to test interventions and unnecessary exposure to potential adverse effects in low-risk patients has been prohibitive. For example, for an anti-epileptogenesis drug trial that enrolled severe TBI patients with an estimated incidence of PTE₁ at 7.1%²⁰, the sample size required to detect a 50% treatment effect is 1364 patients (Fisher’s 2-sided exact test, power 0.8, alpha 0.05). By contrast, if we enroll severe TBI patients with early EAs on EEG, according to our data the incidence of PTE₁ rises to 12%, and the required sample size is only 778, a decrease of 43%²⁰. While our sample size is small, retrospective in design and needs further confirmation, our results suggest by using EA as a biomarker to identify the subset of TBI patients at highest risk for PTE₁ development, anti-epileptogenesis interventions could be feasible in a cost-effective manner.