Neuroimaging for Pediatric Non–First-Time Seizures in the Emergency Department

Abstract

Background and Objectives

Neuroimaging is often part of the workup for a pediatric patient presenting with a seizure to an emergency department (ED). We aim to evaluate when neuroimaging in the ED for children with a non–first-time seizure, or nonindex seizure (NIS), is associated with an acute change in management (ACM).

Methods

This is a retrospective cohort study of all pediatric patients presenting to an ED from 2008 to 2018 with a NIS, excluding repeat febrile seizures, who underwent neuroimaging. Clinical characteristics were extracted from the electronic medical record. The primary outcome was new abnormal neuroimaging resulting in an ACM, defined as admission to the hospital, neurosurgical intervention, or new nonseizure medication administration.

Results

We identified 492 encounters. Neuroimaging revealed new findings in 21% of encounters and led to ACMs in 5% of encounters. ACMs included admissions, neurosurgical interventions, and nonseizure medication changes. Factors associated with ACM included new seizure type (odds ratio [OR] 3.3, 95% confidence interval [CI] 1.3–8.0), new focal examination finding (OR 3.0, 95% CI 1.3–7.1), altered mental status (OR 2.9, 95% CI 1.2–7.0), and a history of only provoked seizures (OR 2.8, 95% CI 1.0–7.5). Patients with 2 risk factors had an OR of 6.9 (95% CI 1.8–26.5) for an ACM, and those with 3–4 risk factors had an OR of 45.8 (95% CI 9.8–213.2). The negative predictive value for ACM in a patient with no risk factors was 98.6% (95% CI 95.9–99.5).

Discussion

Patients with a NIS who have abnormal neuroimaging associated with an ACM present with unique risk factors. Prospectively validating these factors may allow for a prediction tool for NIS in EDs where reduced exposure to ionizing radiation, sedation, and resource utilization are critically important.

Seizures are among the most common childhood neurologic conditions evaluated in the emergency department (ED).¹ Seizures in children may be a clinical manifestation of a variety of severe intracranial pathology; therefore, clinical evaluation in the ED often includes neuroimaging with CT or MRI to identify stroke, intracranial mass, infectious processes, hydrocephalus, and head trauma. The American Academy of Neurology and the Child Neurology Society recommend emergent neuroimaging for first-time seizures in children with a postictal focal deficit or those who have not returned to baseline within several hours.²

For the ED provider, presentation for nonindex seizures (NIS) in children, such as those in children with a history of epilepsy, presents a different challenge. Uniform guidelines do not exist with regard to the management of children with a NIS, specifically with respect to obtaining neuroimaging. Neuroimaging is not without risk, such as exposure to ionizing radiation and procedural sedation or anesthetics, especially in the case of MRI. The pediatric population, especially children younger than 1 year, is more radiosensitive than adults and is at increased risk for developing lethal cancers from exposure to ionizing radiation.^3,4 Repeated exposure to anesthesia and intubation carry associated risks and may have adverse effects on neurodevelopment.^5-7 In addition, many institutions do not have 24-hour MRI capability, which often results in prolonged ED encounters or even hospital admissions. Medically complex children, including those with epilepsy, have increased utilization of the ED and are therefore more likely to be exposed to these risks.⁸ Furthermore, obtaining neuroimaging in the ED is costly, is time-consuming, and uses valuable bed space and medical staffing.

For adults presenting to the ED with a NIS, neuroimaging resulted in an acute change in management (ACM) in only 2.9% of cases. Patients presenting with a NIS in the setting of several risk factors, including recent head trauma, alteration of consciousness, or focal neurologic examination findings, were more likely to have abnormal imaging with actionable findings.⁹ In this study, we aim to evaluate the characteristics of pediatric patients presenting to an ED with a NIS who underwent neuroimaging and to identify risk factors of those most likely to have an ACM associated with findings on neuroimaging. As the epidemiology of seizures in children differs from adults, a unique study with a pediatric population is necessary to classify which features of the pediatric NIS presentation have the highest yield of intervenable neuroimaging findings, as well as children who may warrant a more conservative approach.

Methods

This was a retrospective cohort study of children presenting for a NIS to 1 of 4 EDs in a pediatric tertiary care health system. Patients were included if they were younger than 19 years, presented for a NIS, and had neuroimaging obtained in the ED; children admitted for the sole purpose of obtaining neuroimaging were also included. Children were excluded if the current seizure was a nonindex febrile seizure or if a subject's first lifetime seizure was less than 24 hours before presentation for NIS. Investigators met and defined each variable before chart review to optimize uniformity. In addition, every encounter with abnormal imaging was reviewed a second time by a single reviewer to minimize error.

Outcome Measures

The primary outcome was new abnormal neuroimaging findings that resulted in an ACM. ACMs were defined as hospital admission, neurosurgical intervention, or medication administration or new prescription, excluding adjustments or initiation of antiseizure medications. A patient was considered to have an ACM only if the management change was associated with the acute neuroimaging findings. Admissions to obtain neuroimaging were not included as an ACM.

Data Collection

Patients were identified using a search of the hospital electronic medical record. Specific International Classification of Diseases 9/10 codes were used to identify potentially eligible patients (345, 780.33, 780.39, G40.909, and R56.9). Medical records were reviewed to ensure that patients met the eligibility criteria. For each encounter, data collected included date of visit, patient date of birth, sex, and information regarding prior seizures and neuroimaging across the health care system. Patients were categorized as having a diagnosis of epilepsy (known diagnosis or on antiseizure medications as identified by the ED provider), a history of nonspecific spells (at least 1 unprovoked seizure-like event, not on antiseizure medication, and without a diagnosis of epilepsy), abnormal EEG (without a diagnosis of epilepsy and not on antiseizure medication), or a history of provoked seizures (symptomatic seizures caused by acute pathology, such as after traumatic brain injury, hemorrhage, or infection, that did not reoccur without trigger). For each encounter, historical and examination features included (1) seizure semiology and if there was a change in semiology from previous seizures, (2) seizure duration, (3) presence and duration of altered mentation, (4) focal neurologic deficit on physical examination—classified as a new or chronic finding, (5) concern for acute head trauma, ingestion, or intoxication, (6) presence of ventriculoperitoneal shunt, and (7) type of neuroimaging obtained. Neuroimaging reports were reviewed and compared with previous reports and classified as normal, abnormal but unchanged from previous scans, or new abnormal findings. Any finding reported by the radiologist, except for normal anatomical variants, was classified as abnormal. In the absence of prior imaging, abnormal findings were considered newly abnormal. ACMs based on new abnormal neuroimaging were recorded.

Statistical Analysis

Univariate analysis was used to examine the relationship between encounter variables and ACMs. Variables found to be significant (p ≤ 0.05) were included in a multivariable logistic regression to determine factors most likely to be associated with ACMs. A risk score was calculated assigning equal weight to each risk element. A receiver operator curve (ROC) was created to determine the area under the curve (AUC) for the risk score. Ninety-five percent confidence intervals (CIs) are reported with odds ratios (ORs) and negative predictive values.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was reviewed by our local institutional review board and determined to meet criteria for exemption, including a full waiver of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) authorization.

Data Availability

All anonymized data used in this study are available for review and can be sent to any qualified investigator on request.

Results

The study included 492 ED visits composed of 396 unique patients. Ages ranged from 1 month to 18.9 years old, with a mean age of 8 years (SD ± 5.4 years); 55% were male. For NIS encounters, the majority (68%) of children presented with increased seizure frequency, duration, or new seizure clustering (Table 1). Neuroimaging obtained included head CT (67%), followed by MRI (20%), MRI-abbreviated protocol (15%), and plain radiographic shunt series (14%). Plain radiographic shunt series was obtained along with MRI or CT scan in 70 of 71 encounters. The single patient who underwent only plain radiographic shunt series had a normal study.

Table 1.

Frequency of Risk Factors

Most encounters had abnormal neuroimaging studies (279/492; 57% 95% CI 52.3–61.0). A newly identified abnormal finding was found in 21% (102/492; 95% CI 17.4–24.5) of encounters. New abnormal findings included imaging with abnormal findings when no comparison was available (24/102) or if the finding was a change from prior images (78/102). Among encounters with new abnormal neuroimaging, there was an associated ACM in 25% (26/102; 95% CI 18.0–34.7) (eTable 1, links.lww.com/NXI/A708), representing 5.3% (26/492; 95% CI 3.6–7.6) of all encounters. ACMs included 21 hospital admissions, 6 neurosurgical interventions, 7 changes in medication regimens, and 4 changes classified as “other” (ventriculoperitoneal shunt tap, emergent intubation, hypercoagulability workup, and evaluation for nonaccidental trauma).

Among children with a NIS presenting to the pediatric emergency department, ACMs were associated with a history of provoked seizures, new seizure semiology, focal seizures, a new focal examination finding, and altered mental status on ED presentation (Table 2). In our multivariate regression model, presentation with a focal seizure was not significantly associated with an ACM when controlling for other factors, but all other variables remained statistically significant (Table 3). The study population included 161 of 492 patients with a ventriculoperitoneal (VP) shunt, of which 8 had an associated ACM (5%). In the post hoc analysis of only those with VP shunts, new seizure semiology was the only risk factor associated with an ACM (OR 6.3, 95% CI 1.5–27.1). Of children without VP shunts, the only substantial change in the univariate analysis was that new seizure semiology was not significantly associated with ACMs (OR 2.5, 95% CI 0.9–6.4). Of the 26 ACMs, 9 were identified with MRI or abbreviated MRI and 20 with head CT. Three patients with an ACM underwent both CT and MRI. In all 3 cases, both modalities were abnormal, with the secondary study providing additional details of the identified pathology.

Table 2.

Risk Factors for Acute Change in Management

Table 3.

Multivariate Analysis of Risk Factors for Acute Change in Management

Risk factors identified in the multivariate model were used to create a risk score for each encounter. The presence of multiple risk factors was associated with a dose-dependent increased risk of abnormal imaging with an ACM (Table 4). Presentation for a NIS with 2 risk factors (OR 6.9, 95% CI 1.8–26.5) or ≥3 risk factors (OR 45.8, 95% CI 9.8–213.2) was associated with increased odds of imaging leading to an ACM (p < 0.01). The negative predictive value for ACM in a patient with no risk factors was 98.6% (95% CI 95.9–99.5). The positive predictive value of the presence of 2 risk factors was 9.1% (95% CI 4.7–16.9) or ≥3 risk factors was 40.0% (95% CI 20.4–63.4). A ROC, demonstrating the sensitivity and specificity of the risk score, yielded an AUC of 0.75 (95% CI 0.65–0.85).

Table 4.

Odds Ratio of Acute Change in Management Based on the Number of Risk Factors

Of the new abnormal scans, the majority (76/102; 75% 95% CI 65.2–82.0) did not lead to an ACM. The new imaging findings not associated with an ACM were broadly classified as follows: ventricle size change not found to be clinically significant (21%), encephalomalacia or volume loss from prior injury (13%), structural anomalies such as arachnoid cyst and mesial temporal sclerosis (9%), interval change of chronic pathology including tubers in tuberous sclerosis and growth of a glioma (10%), and postsurgical changes (7%). Other findings (8%) included extraventricular shunt location, nonspecific white matter changes, acute sinusitis, MRI perfusion changes suggestive of spreading depolarization, and 2 children with artifact vs pathology. The 2 children in our study with abnormal imaging due to artifact underwent additional monitoring and follow-up imaging with resolution of the findings. New abnormal imaging findings led to a non-acute change in management in 15% (16/102; 95% CI 9.9–24.0) of encounters: 3 new diagnoses, 4 referrals to a new specialty clinic, 6 interval neurosurgical follow-ups, 2 children requiring follow-up imaging, and 1 study used to help prognosticate for a child with high-grade glioma.

Discussion

In this study, we found that neuroimaging in children presenting with a NIS led to an ACM in 5% of encounters. We were able to identify risk factors associated with ACMs in our cohort of children, which has not been previously described. New abnormal neuroimaging was relatively common in children with a NIS, present in 1 in 3 children. Many of the new imaging findings were expected for a patient's given history, such as new tubers in a child with tuberous sclerosis or encephalomalacia from an old brain injury. Neuroimaging did not acutely change management in most encounters; 1 in 20 children presenting for a NIS had an associated ACM. We identified 4 risk factors that were significantly associated with ACMs: having a history of provoked seizures alone, presentation with new seizure semiology, a new focal examination finding, and altered mental status on ED presentation. Our findings support that children presenting with a NIS in the setting of any of these risk factors, especially if multiple risk factors are present, should prompt consideration for neuroimaging. When 3 or 4 risk factors were present, neuroimaging led to an ACM in 2 of every 5 patients. Conversely, in the absence of these risk factors, there was a low (1.4%) chance of an actionable imaging finding and a high negative predictive value (98.6%) for an ACM. The choice of imaging modality should be made with the consideration of several factors: imaging modality availability, urgency of obtaining neuroimaging, modality most likely to demonstrate the pathology of the highest concern, limiting lifetime radiation exposure, and risk of IV contrast agents.

Seizures are a common reason for a child to present to the ED, accounting to 1% of all pediatric ED encounters.¹ Although ED providers have guidelines for the evaluation and treatment of febrile seizures and first-time seizures, they are left without direction for the child presenting with a recurrent seizure. The overall low rate of actionable imaging findings for children with a NIS poses a challenge to the ED provider who does not want to miss intervenable findings (intracranial bleed, brain tumor, and acute hydrocephalus) but wants to limit the harm associated with unnecessary emergent imaging. With validation in a prospective cohort, our findings may help risk-stratify patients with readily identified factors present on presentation, allowing ED clinicians to minimize exposure to ionizing radiation and sedation as well as decrease the time and cost associated with obtaining emergent MRIs. Our study resulted in the identification of different risk factors than those in adults.⁹ This is likely due to a number of factors that vary between adults and children including the causes of trauma, developmental stages and disability, comorbid illnesses, and the underlying etiology of seizures. Critically, the presence of a new focal examination finding was a risk factor for an ACM in both adults and children, highlighting the importance and sensitivity of a newly abnormal neurologic examination. It is essential to determine whether abnormal examination findings are consistent with prior examinations from caregivers or the medical chart available.

Three patients in our study had an ACM without identified risk factors. Two of these patients had trauma associated with their seizures: a patient with fall proceeding increased seizures was found to have a small nonsurgical subarachnoid hemorrhage, and a patient with an atonic seizure with head strike found had a temporal skull fracture. It is highly likely that these children would have been imaged based on head trauma guidelines, independent of NIS presentation. Although head trauma was not predictive of ACM in our logistical regression model, trauma is still an appropriate indication for neuroimaging in certain circumstances.¹⁰ The final patient without identified risk factors was a 3-month-old with cortical dysplasia admitted for additional imaging and EEG monitoring. Although this patient met ACM criteria because she was admitted for her CT findings, retrospectively, it is difficult to determine whether other factors, e.g., age, influenced the decision for admission independent of neuroimaging.

Neuroimaging is not benign, and the risks and benefits of obtaining neuroimaging must be critically evaluated for every patient. Among our study population, 34 were children younger than 1 year, 24 of whom (71%) were imaged using CT of the head. It is important that these patients are at highest risk for the development of lethal malignancies from exposure to ionizing radiation.³ This concern is heightened in the children in our study because children with epilepsy often have multiple ED encounters with several neuroimaging studies, therefore increasing the cumulative risks from ionizing radiation.⁸ Admission for the sole purpose of obtaining MRI was needed in 46 of 492 (9%) encounters (admissions for imaging were not considered ACMs without further action). Many pediatric hospitals have shifted away from CT scans and increased MRI use to minimize radiation exposure.¹¹ However, as MRI scans take longer and require skilled technicians, this shift places scheduling demands on hospitals and requires prioritization of the patients most likely to benefit from an in-hospital MRI. Furthermore, inpatient MRIs are costly to hospitals, lead to prolonged patient length of stay, and use valuable resources including bed space, nursing, and physicians.^12,13 Finally, incidental findings and artifacts on imaging are unavoidable in diagnostic imaging, but limiting unnecessary neuroimaging decreases the chance of these false positives and the associated further workup. We identified 2 patients who underwent additional workup of artifacts leading to the unintentional consequences of distress to families and utilization of medical resources.

There are limitations to this study. The study was conducted at a single tertiary care pediatric hospital system. Neuroimaging practices likely vary widely based on the patient populations, access to neuroimaging, financial barriers, and local practice. Data were collected using retrospective chart review, which is subject to the quality of documentation and variability in identifying relevant outcomes. We used radiology reports, rather than direct review of the neuroimaging. However, we felt this method of chart review best matched how an ED provider would view and process information when evaluating a child. We were limited to neuroimaging available in our electronic health record, likely leading to misclassification of chronic findings as new in the absence of prior imaging. Similarly, when comparing a new MRI with a prior brain CT, the higher sensitivity MRI may pick up findings classified as new but could have been present and not seen on prior CT scan. We recognize that there is value in negative neuroimaging when ruling out acute pathology, such as a stroke or intracranial hemorrhage. Because of the retrospective nature of this study, it would be difficult to determine with certainty when a negative study was the key factor associated with a management change. This study does help determine the incidence of negative imaging in our cohort, which may help providers balance risks and benefits better based on the clinical scenario at the time. We did not determine a “false-negative” rate wherein imaging was obtained, there was no ACM, but patients required interventions at a later time. We also did not evaluate the outcomes of patients with a NIS who did not have neuroimaging. Finally, the retrospective nature of this study and low rate of encounters in which neuroimaging led to a change in management limited the number of risk factors or risk factor combinations that we were able to explore and our ability to perform subanalyses on specific patient groups. There may be children with unique sets of risk factors, such as those with VP shunts, that could be better understood with future prospective studies.

Neuroimaging in the absence of identified risk factors was not associated with ACMs in children presenting to the ED for a NIS. Specifically, patients without a new seizure semiology, only prior provoked seizures, altered mental status, or a new focal examination finding had a low risk of neuroimaging leading to ACMs. Neuroimaging plays an important role in the workup of children with recurrent seizures; however, limiting low-yield imaging in the acute setting can decrease cumulative radiation and anesthesia exposure in many children. Additional studies to validate a decision-making tool for the ED clinician caring for a child with a NIS could prove valuable.

Appendix. Authors

Study Funding

The authors report no targeted funding.

Disclosure

The authors report no disclosures relevant to the manuscript. Full disclosure form information provided by the authors is available with the full text of this article at Neurology.org/cp.