Electrographic Seizure Detection by Neuroscience Intensive Care Unit Nurses via Bedside Real-Time Quantitative EEG

Abstract

Objective

Our primary objective was to determine the performance of real-time neuroscience intensive care unit (neuro-ICU) nurse interpretation of quantitative EEG (qEEG) at the bedside for seizure detection. Secondary objectives included determining nurse time to seizure detection and assessing factors that influenced nurse accuracy.

Methods

Nurses caring for neuro-ICU patients undergoing continuous EEG (cEEG) were trained using a 1-hour qEEG panel (rhythmicity spectrogram and amplitude-integrated EEG) bedside display. Nurses' hourly interpretations were compared with post hoc cEEG review by 2 neurophysiologists as the gold standard. Diagnostic performance, time to seizure detection compared with standard of care (SOC), and effects of other factors on nurse accuracy were calculated.

Results

A total of 109 patients and 65 nurses were studied. Eight patients had seizures during the study period (7%). Nurse sensitivity and specificity for the detection of seizures were 74% and 92%, respectively. Mean nurse time to seizure detection was significantly shorter than SOC by 132 minutes (Cox proportional hazard ratio 6.96). Inaccurate nurse interpretation was associated with increased hours monitored and presence of brief rhythmic discharges.

Conclusions

This prospective study of real-time nurse interpretation of qEEG for seizure detection in neuro-ICU patients showed clinically adequate sensitivity and specificity. Time to seizure detection was less than that of SOC.

Trial Registration Information

Clinical trial registration number NCT02082873.

Classification of Evidence

This study provides Class I evidence that neuro-ICU nurse interpretation of qEEG detects seizures in adults with a sensitivity of 74% and a specificity of 92% compared with traditional cEEG review.

Nonconvulsive status epilepticus (NCSE) and nonconvulsive seizures (NCSs) are common in intensive care unit (ICU) patients, particularly in the neuroscience ICU (neuro-ICU), with prevalence reported from 8% to 21%.^1-4 Studies have shown that NCSE/NCS are predictors of functional and cognitive outcomes and that delayed treatment can worsen outcomes.^1,5-9

Delays to treatment are common, often owing to delay in diagnosis.^9-11 The gold standard for diagnosis of NCSE/NCS is continuous EEG (cEEG), which requires interpretation by a neurophysiologist.^12,13 Surveys have shown that neurophysiologist review of these lengthy studies is only completed 2–3 times per day, creating a window as long as 8–12 hours in which a patient could be having unrecognized seizures.^13,14

Quantitative EEG (qEEG) distills cEEG data into its salient features, allowing more rapid review by neurophysiologists and non-neurophysiologists.^14-19 In addition, several retrospective studies have shown promise for ICU nurse interpretation of qEEG as a screening tool for seizure detection in adults.^20-22

A previous prospective neuro-ICU study administered tailored training sessions to nurses caring for patients with known NCSE/NCS.²³ Detection of recurrent seizures was compared with gold standard cEEG review in those patients.²³ Nurse interpretation in that context had excellent sensitivity (85%), specificity (90%), and false alarm rate (0.1/hour).²³ The present study's primary objective was to determine the performance of real-time neuro-ICU nurse interpretation of qEEG in patients with unknown NCSE/NCS presence. Secondary objectives included determining time to seizure detection compared with standard of care (SOC) and factors that affect nurse accuracy.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

This study was approved by the Duke Health Institutional Review Board (IRB). Written informed consent for participation in the study was obtained from neuro-ICU nurses who were caring for patients. A waiver of Health Insurance Portability and Accountability Act authorization and consent was approved by the Duke Health IRB to collect patient and EEG data retrospectively.

Study Design

This is a single-institution prospective cohort study. Adult patients admitted to the Duke neuro-ICU and started on cEEG as a part of usual clinical care between September 2018 and January 2020 were identified prospectively and included based on study team availability.

Patients were selected based on the following inclusion criteria: age 18 years or older, cEEG initiation within the past 12 hours, absence of seizures at the time of study initiation, and indication for cEEG any reason except hypoxic brain injury. The same patient could be studied more than once if cEEG was discontinued and then restarted, and no seizures were appreciated during the previous recording.

Nurses were consented and then administered a 5–10-minute, standardized, in-person qEEG training session by a member of the research team via printed PowerPoint slide set. The training included background information on qEEG and example qEEG panel printouts. The research team member then inspected the bedside display to ensure that the correct trends were on screen and allowed the nurses to ask any questions about the bedside display. For the remainder of their 12-hour shift, nurses were instructed to log the number of seizures seen on the bedside display based on their interpretation. Nurses were able to defer the hour if the patient was disconnected from EEG for CT, for example, or if nurses had other prioritized clinical responsibilities. Seizures were logged in the following bins: no seizures, 1–2 seizures, 3–5 seizures, 6–10 seizures, and >10 seizures. The training slides were available to nurses for the duration of the shift. Information about the number of years of experience each nurse had in the neuro-ICU was also collected. Nurses were eligible to participate multiple times in the study with the option to go through the training session again.

Data Collection

Digital cEEG recordings (Natus Medical, Pleasanton, CA) were obtained with electrodes placed according to the international 10/20 electrode placement system. All cEEG recordings were processed through Persyst 12 (Persyst Inc., Prescott, AZ) to yield 1-hour qEEG trends displayed in real time at the patient bedside adjacent to a cEEG display. The qEEG panels consisted of rhythmicity spectrogram (Persyst Inc., displayed for both the left and right hemispheres; range 1–25 Hz, 3-second epochs with 1 second step size, EEG time constant = 0.16 second, high-frequency filter = 35 Hz, rhythmicity sampling rate = 128 Hz, epoch duration = 3 seconds, epoch step = 2 seconds, y-axis range 1–25 Hz, y-axis scaling type = square root, z-axis scaling using μV/Hz, z-axis range 0–4 μV/Hz z-axis color palette) and amplitude-integrated EEG (aEEG, displayed for the left and right hemispheres; time constant of 0.5 second with 1 second epochs; downsampled to a rate of 64 samples per second and then filtered using a 60-Hz notch filter and an asymmetrical filter). These trends also underwent automated artifact reduction through the Persyst 12 software, which removes electrode and physiologic artifact (Figure 1).

Figure 1. Sample Bedside Display.

A screen at the patient bedside displayed cEEG on the left (A) (bipolar double banana montage), with a qEEG panel (B) including rhythmicity spectrogram (L/R) and amplitude-integrated EEG (L/R) on the right. Nurses would consult this screen on an hourly basis and log the number of seizures seen. Three seizures are shown on qEEG here along with ictal cEEG. Screenshot taken from Persyst 12. cEEG = continuous EEG; qEEG = quantitative EEG.

Patient demographics and diagnosis information were collected post hoc using the electronic medical record. The corresponding cEEG recordings from each patient were deidentified and reviewed by study authors and board-certified neurophysiologists C.B.S. and C.E.H. post hoc to identify each seizure and describe their characteristics. This is considered the gold standard for NCSE/NCS detection. Electrographic seizures were identified using previously published criteria.²⁴ Interrater reliability was calculated, and for any discrepancies, consensus was reached through deliberation by the 2 cEEG readers. The seizure spatial extent (focal, hemispheric, or generalized/secondary generalized), duration (10–30 seconds, 31–60 seconds, 61 seconds–5 minutes, or >5 minutes), amplitude (low [20–49 μV], medium [50–199 μV], or high [>200 μV]), and background (brief rhythmic discharges [BRDs], periodic discharges [PDs], rhythmic delta activity [RDA], or spike and wave) were determined by study author C.B.S.

Definitions

Rhythmicity spectrogram is a proprietary qEEG algorithm from Persyst Inc. that displays time-compressed frequency in a banded fashion on the y-axis, emphasizing frequency bands with the highest amount of rhythmicity.¹⁵ aEEG is also a time-compressed display and displays filtered, smoothed, and rectified minimum and maximum amplitude for each hemisphere.²⁵ These trends were chosen as they encapsulate multiple dimensions of EEG data and allow for a simplified panel for ease of use. They were also rated to be the most helpful in a previous retrospective study and were used in the previous prospective study completed in the Duke neuro-ICU.^21,23

At our institution, SOC detection is the time of electrographic seizure detection by neurophysiology fellows. This is followed by post hoc verification by a board-certified neurophysiologist.

Statistical Analysis

For our primary objective, standard test characteristics (sensitivity, specificity, positive predictive value, negative predictive value, and false alarm rate) with 95% confidence intervals (CIs) were calculated on a per-hour basis to compare nurse interpretation of qEEG for seizure detection with the gold standard of cEEG interpretation by neurophysiologists. These measures were calculated using a generalized estimating equation using patient number as a random effect to account for clustering on a by-patient basis (Class I Diagnostic Accuracy Criteria).

For our secondary objectives, interrater reliability using Cohen kappa for seizure detection by the 2 neurophysiologists was calculated. Nurse responses that included 1–2 seizures, 3–5 seizures, 6–10 seizures, and >10 seizures were grouped as having seizures present for analyses that assessed for presence or absence of seizures.

Kaplan-Meier curves were used for time to event analysis comparing latency of seizure detection between nurses and SOC detection, using a Cox proportional hazards model to calculate the hazard ratio and associated p value. Finally, hypothesis tests were conducted to determine whether other characteristics of patients, nurses, or EEG affected nurse accuracy. Nurse accuracy was analyzed at the patient level rather than the 1-hour block level. For example, if a nurse noted at any point that a patient had more than 1 seizure, and the neurophysiologists noted at least 1 seizure, this would be considered accurate at the patient level. Analysis of seizure characteristics was performed using only the patients who had seizures during the study period. For continuous variables, the Wilcoxon rank-sum test was used, and for categorical variables, the Fisher exact test was used. A p value of <0.05 was considered statistically significant. Statistical analyses were performed using R (version 3.5.1; The R Foundation for Statistical Computing).

Data Availability

The data that support the findings of this study are available publicly from the Duke Research Data Repository (DOI: 10.7924/r4mp51700).

Results

This study included 65 nurses, 109 patients (106 of whom were unique), and 723 associated 1-hour qEEG/cEEG blocks. Patient demographics and characteristics are summarized in Table 1. Eight patients (7%) had at least 1 seizure during the study period, and 101 did not have any seizures during this period. None of the patients who were included more than once had seizures on more than 1 admission during the study period. The most common diagnosis was intracerebral hemorrhage. The median length of data collection for each patient was 7 hours. Of the 723 one-hour blocks studied, 26 contained seizures (4%). EEG characteristics as determined by the 2 neurophysiologists from these patients are summarized in Table 2. The most common background EEG pattern was slowing, seen in 57 patients (52%) followed by periodic discharges 31 (28%). There were 235 total seizures seen in the data set. Of patients with seizures, the median number of seizures during the study period was 11.5 with range (1–163). Interrater reliability of seizure number between the 2 neurophysiologists grading cEEG was moderate with kappa = 0.78. Interrater reliability for binary seizure presence or absence per hour was strong with unweighted Cohen kappa = 0.81.

Table 1.

Patient Demographics

Table 2.

EEG Characteristics

For our primary objective of determining test characteristics of nurse interpretation of qEEG for seizure detection, overall, sensitivity for seizure detection was 74%, specificity was 92%, and false-positive rate was 0.08/hour. Nurse characteristics are described in Table 3. Nurses deferred logging 42 times total, for various reasons including lack of patient availability due to need for CT and other clinical priorities superseding hourly logs. Nurses accurately identified the correct seizure category bin in 91% of 1-hour blocks. Nurses detected seizures in 6 of 8 (75.0%) seizing patients. The clinical team using SOC detected seizures in 7 of 8 (87.5%) seizing patients. The nurse on study as well as the clinical team failed to recognize a seizure in 1 patient with a single focal seizure during the study period. The other patient with seizures that the nurse on study failed to identify was also a single focal seizure. Figure 2 depicts a qEEG display with 1 seizure that was not identified by the nurse (false negative) as well as a seizure that was identified by the nurse (true positive).

Table 3.

Nurse Performance

Figure 2. Example of qEEG False Negative and True Positive.

qEEG panel with rhythmicity spectrogram (L/R) and amplitude-integrated EEG (L/R) depicting a focal, <30-second, low-amplitude electrographic seizure that was not logged by the nurse on study (red arrow—false negative) and a 61-second–5 minute, generalized, high-amplitude seizure that was logged by the nurse on study (green arrow—true positive). Screenshot from Persyst 12. The false-negative seizure has only a subtle signature on left hemisphere rhythmicity spectrogram and amplitude-integrated EEG (red boxes), whereas the true-positive seizure has the classic flame appearance on rhythmicity spectrogram as well as upward deflections.^19,22 qEEG = quantitative EEG.

For our secondary objective of time to seizure detection, of the 6 patients whose seizures were detected by SOC as well as the nurse on study, average time to seizure detection by the clinical care team was 185 minutes (median = 149.5) and average time to seizure detection by the nurse on study was 53 minutes (median = 46.5). Compared with SOC time to seizure detection, nurses in the study were able to detect the first seizure 132 minutes faster, on average. Kaplan-Meier curves including all patients with seizures, with nondetections censored, are shown in Figure 3. The proportional hazards assumption was met using the Schoenfeld individual test with p = 0.2899. Cox proportional hazard testing resulted a hazard ratio of 6.96, with 95% CI (1.31–36.9) with a p = 0.023. This hazard ratio suggests that at any particular time, patients were 6.96 times more likely to have their seizure detected by the nurse compared with the SOC.

Figure 3. Kaplan-Meier Plot for Latency Between Seizure Occurrence and First Seizure Detection.

Survival curves for time to latency to seizure detection from onset of first electrographic seizure using nurse interpretation of qEEG compared with SOC. Cox proportional hazard ratio = 6.96 with 95% CI (1.31–36.9) and p = 0.023. CI = confidence interval; Cox PH = Cox proportional hazards; HR = hazard ratio; qEEG = quantitative EEG; SOC = standard of care.

Effects of other factors on nurse accuracy were also analyzed. At the patient level, 80 patients (67%) were accurately found to have seizure or no seizure by nurse interpretation. Using the Wilcoxon rank-sum test, we found that patients who were not accurately categorized by nurses had statistically significantly more hours monitored on study (6 vs 8 hours). We also found using the Fisher exact test that nurse accuracy was not independent from EEG background. Patients with BRD were more likely to be inaccurately categorized, but patients with PD, RDA, or no background abnormality were not. However, this is limited by the presence of only 1 patient with BRD (Table 4). No adjustments for multiple comparisons were made due to low sample size for exploratory analyses.

Table 4.

Characteristics Affecting Patient-Level Accuracy

Classification of Evidence

This study provides Class I evidence that neuro-ICU nurse interpretation of qEEG detects seizures in adults with a sensitivity of 74% and a specificity of 92% as compared to traditional cEEG review.

Discussion

Our primary objective was determining test characteristics of neuro-ICU nurse interpretation of real-time qEEG at the bedside for seizure detection in patients with unknown NCSE/NCS presence. We found that neuro-ICU nurses are able to detect electrographic seizures in patients using a simplified panel of qEEG trends with only a short, standardized training session. We found the sensitivity of seizure detection to be 74% and specificity to be 92%, similar to the previous prospective study in patients with recurrent NCSE/NCS, which reported a sensitivity of 85% and a specificity of 90%.²³ Despite a lower seizure prevalence in the present study, this did not result in significantly poorer sensitivity. We found a false alarm rate of 0.08 per hour, lower than the aforementioned study, which found a false alarm rate of 0.10 per hour. Notably, although the nurses in the present study were not seeking a patient-specific qEEG pattern as they were in the study on recurrent NCSE/NCS, this did not result in a higher false alarm rate.²³ The false alarm rate is, however, biased by the low seizure prevalence in our sample and should be interpreted with caution for that reason. Sensitivity of nonphysician interpretation of qEEG in other studies has ranged from 74%²² to 87%,²¹ and specificity has ranged from 38%²² to 90%.^19,20,23 Thus, we find that our values are fairly consistent with the previous literature, which were done largely in retrospective studies. This sensitivity, specificity, and false alarm rate supports the practice of neuro-ICU nurse interpretation of qEEG in real time at the bedside as an effective screening tool for NCSE/NCS.

One of our secondary objectives was to determine the time to seizure detection by nurse qEEG interpretation compared with SOC. The current gold standard for NCSE/NCS detection in the neuro-ICU is intermittent interpretation of cEEG by a board-certified neurophysiologist, although a recent prospective study trained non-neurophysiologists participants in cEEG for seizure detection.^12,13,26 We find that nurse interpretation leads to clinically and statistically significantly faster time to seizure detection compared with routine cEEG review. The delay in seizure detection for patients with NCSE/NCS is not unique to our institution and has been shown to be significantly longer than in patients having convulsive seizures, which can often be appreciated without cEEG.^9-11,27,28 Factors that can delay detection include time taken to risk stratify the patient, EEG technician availability, time taken to initiate cEEG, high study burden on neurophysiology fellows and attendings, and time needed to review long cEEG studies. Considering the promising results of qEEG in the literature as well as the data from this study, qEEG can be used to mitigate some of the delays NCSE/NCS patients face when it comes to treatment. It has been shown that increased seizure burden correlates with increased patient morbidity and mortality.^6,29 Although it is unclear in which direction causation lies, it is possible that earlier detection of seizures will lead to earlier treatment of seizures, resulting in improved patient outcomes.

Our other secondary objective was to determine whether there were any nurse, patient, or EEG characteristics that affected test characteristics. The previously mentioned prospective nurse study that trained nurses on patients who already demonstrated NCSE/NCS noted that neuro-ICU nurses were statistically significantly more likely to detect a generalized seizure compared with a focal seizure.²³ This was also observed in a retrospective qEEG study sample of EEG technologists, but not neuro-ICU nurses.²¹ We did not find this effect in our sample, but our study is underpowered to detect this kind of difference. In addition, choice of qEEG trend can have an effect on seizure types that are more easily detected. For example, asymmetry spectrogram is excellent at highlighting and emphasizing short, focal seizures, whereas most other trends are not.^19,30 Subsequent studies may consider including asymmetry spectrogram if this trend is borne out in clinical practice and future investigations. We also did not find that nurses with more neuro-ICU experience had statistically significantly better or worse accuracy, a question that was brought up in the previous prospective nurse study.²³

We did find some statistically significant differences in nurse accuracy when it came to hours monitored as well as EEG background abnormality on the patient level. We found that patients with more hours on study were more likely to be inaccurately assessed by nurses. The difference in hours monitored between accurately categorized patients and inaccurately categorized patients is not something we found to be clinically significant however, with a difference of only 2 hours. In addition, we found that EEG background abnormality was not independent of nurse accuracy, with BRD having the most inaccurate cases and no specified background abnormality having the most accurate cases. However, we are limited in our analysis of EEG background as only 1 patient had BRD on EEG. For the sake of this exploratory analysis, no adjustment was made for multiple comparisons. This, along with our low sample size, prompts us to interpret these results with caution. Further studies with a higher prevalence of seizures are needed to more fully assess these factors.

The results of this study have several implications for eventual implementation of qEEG in the neuro-ICU. Our false alarm rate of 0.08 per hour is one of the lowest in the current literature, although as mentioned previously, this number is biased by the low seizure prevalence in our sample. Regardless, as cEEG volume increases, the number of additional false alarms may be disruptive. This is of particular concern to centers with high cEEG volumes. For example, if a given neuro-ICU has 5 cEEG studies running, nurse interpretation based on this study would lead to about 10 additional false alarms over a 24-hour period. If institutions decide to implement nurse interpretation of qEEG to flag areas of concern, a workflow would need to be instituted to assess these areas. One scenario is that flags from nurses could alert the EEG reader (neurophysiology fellow or attending) directly. Another option is having an initial review by the neuro-ICU team, which could then be routed to the EEG reader if deemed of high concern. Choice of workflow would vary on a variety of institutional factors including EEG reader workload, the neuro-ICU team's makeup and comfort with EEG, and others. Although these questions have not yet been studied, the latter pathway could decrease the number of false alarm calls to the EEG reader, minimizing alarm fatigue. Implementation studies are needed to determine the most optimal strategy.

For the purposes of reducing bias in this study, nurses did not receive feedback on their previous performance before their subsequent participation. In reality, if implemented clinically, nurses would receive feedback, perhaps even in real time, and would theoretically perform better with each EEG patient encounter.

We found that nurses deferred hourly logging 42 times in aggregate, which is low considering that 109 patients were included in the study, resulting in 723 total logged 1-hour blocks. Nurses deferred hourly logging for patient-related reasons as well as reasons relating to their other clinical priorities. We did not collect information from nurses on reasons for deferral, impression of qEEG ease of use, or time required to interpret. Although those variables were not collected, it appears that nurses are able to log a large proportion of hours on study even with their other clinical responsibilities, which is important when considering widespread implementation.

Our study has a variety of limitations. The literature reports prevalence of NCSE/NCS in neuro-ICU populations to be between 8% and 21% of patients.^1-4 Although the prevalence in our sample, 7%, barely falls outside this range, a study conducted at our institution found retrospective prevalence to be 21%.² It is likely that our sample is not completely representative of the neuro-ICU patient census. We hypothesize that this is due to a variety of factors. We were only able to enroll patients when the research team was available to consent nurses, most often during usual weekday working hours. We also only enrolled patients who had been on EEG for less than 12 hours to (1) ensure reasonable duration of cEEG monitoring for patients in this study and (2) reduce and/or avoid nurse bias by enrollment before verbal or EMR feedback on the results of cEEG monitoring. This likely limited our study size. In addition, our inclusion and exclusion criteria prevented us from including a patient who already had 1 or more seizures noted by the clinical team and/or neurophysiology team to avoid biasing nurses by enrolling a group of patients more likely to seize. Thus, patients who demonstrated seizures very close to time of EEG initiation were excluded. This was done not only to avoid biasing the nurses on study, but also to test nurse interpretation of qEEG in a novel setting, as the prior prospective study was conducted solely on patients with known NCSE/NCS.²³ The present study considered a mutually exclusive set of patients from the prior study, allowing our results to be considered in conjunction with the prior data. With a low number of seizing patients on study, true sensitivity is more difficult to infer (95% CI 29%–90%), whereas specificity is more easily studied (95% CI 88%–95%), and the results should be interpreted with this limitation in mind.

In addition, we studied a limited number of hours as we were limited by the length of neuro-ICU nurse shifts, which are 12 hours at most; therefore, no 1-hour block studied was past the 24-hour mark of EEG monitoring for any given patient. Previous studies show that 88% of patients who will have seizures have them in the first 24 hours, and 92%–93% of them will have seizures in the first 48 hours, so any patients who eventually had NCSE/NCS were likely to have had a seizure during their study period.^2,12 However, this time frame does not adequately account for degrading of EEG signal over time. Although we did find that the median number of hours monitored of patients inaccurately categorized by nurses was 2 hours higher than those who were accurately categorized, this difference is small on the scale of days-long studies. However, this may represent some degradation of EEG signal. On the other hand, given that most seizures present in the first 24–48 hours, and our use of automated artifact reduction, we believe that this effect has been minimized. Future study is needed regarding qEEG interpretation for longer-term monitoring in the neuro-ICU and associated EEG signal changes.

Finally, different institutions use various EEG acquisition software and qEEG algorithms. Some have also begun to use rapid response or sonified EEG rather than qEEG to decrease time to seizure detection.^31,32 We utilized Persyst and used their proprietary rhythmicity spectrogram, limiting our generalizability. Other similar studies have used combinations of other trends including compressed spectral array and asymmetry spectrogram. Further study is required to determine the best collection of trends for bedside seizure detection. There is also no standardized training protocol or module for seizure detection using qEEG. In this study, we used a short training session for the convenience of our nurses who have many other clinical responsibilities, but other studies have used longer and more in-depth sessions, albeit with comparable results.^20-22

Conclusions

We find real-time neuro-ICU nurse interpretation of bedside qEEG trends in patients with unknown NCSE/NCS presence to be sensitive and specific with low false alarm rate. Nurse interpretation significantly decreases time to seizure detection compared with current clinical practice at our institution. Characteristics of patients, nurses, and EEG that improve performance of nurse interpretation, alternative panels of qEEG trends, and a standardized training module require further study.

TAKE-HOME POINTS

• → Quantitative EEG as interpreted by neuro-ICU nurses has clinically useful sensitivity and specificity for seizure detection in adult neuro-ICU patients.

• → Although less accurate, neuro-ICU nurse time to seizure detection was faster than that of the current standard of care seizure detection at our institution.

• → Further study is needed on a representative sample of neuro-ICU patients to further elucidate the best strategies for clinical implementation.

Appendix. Authors

Study Funding

The role of the first author in this study was supported in part by a Pfizer Foundation grant and the Duke Translational Science Institute (CTSI). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Pfizer Foundation of the Duke CTSI.

Disclosure

S. Kaleem, J.H. Kang, A. Sahgal, and C.E. Hernandez report no disclosures relevant to the manuscript. S.R. Sinha has received research support from Eisai Pharmaceuticals, Monteris Inc., and UCB Pharmaceuticals; has received consulting fees from Monteris Inc., UCB Pharmaceuticals, and Basilea Pharmaceuticals; and has received royalties from Springer International Publishing. C.B. Swisher has received speaker's honorarium from UCB Pharmaceuticals and Eisai Pharmaceuticals; has served on an advisory board for Eisai; and receives consulting fees from Minnetronix and Marinus. Full disclosure form information provided by the authors is available with the full text of this article at Neurology.org/cp.