Diagnostic value of stand‐alone videos after inconclusive inpatient EEG–video monitoring

Abstract

Inpatient electroencephalographic (EEG)‐video monitoring occasionally fails to capture the events in question, leaving patients without a definitive diagnosis. Management after an inconclusive admission is generally limited to repeating monitoring or empirical adjustment of antiseizure medications. Although stand‐alone videos are increasingly used in clinical practice, their diagnostic value in this setting has not been well established. We aimed to assess the diagnostic contribution, submission rate, and interrater reliability of stand‐alone videos obtained after inconclusive inpatient EEG‐video monitoring. Over a 26‐month period, we prospectively identified patients with inconclusive epilepsy monitoring unit (EMU) admissions who were instructed to submit videos of typical events not captured during their stay. Submitted videos were independently reviewed by two epileptologists, with adjudication by a third reviewer when consensus was not achieved. Among 619 EMU admissions, 165 (27%) were inconclusive. Twenty‐seven (16%) patients submitted a stand‐alone video. A definitive conclusion was reached in 23 (85%) patients, and in 20 (74%) patients, videos either established, confirmed, or revealed an additional diagnosis. Substantial interrater reliability was observed (κ = .690, 95% confidence interval = .456–.893), with perfect agreement for nonepileptic and physiologic events. This study provides preliminary evidence that stand‐alone videos have the potential to be a valuable adjunct in the evaluation of seizures and seizurelike events after inconclusive EMU admissions. Early and systematic reinforcement of video collection may improve diagnostic yield and reduce health care utilization when inpatient studies are unrevealing.

Key points.

• EEG‐video monitoring often fails to capture events, affecting approximately 1 in 4 patients admitted for evaluation.

• Although stand‐alone videos are infrequently submitted, when available, they led to a diagnosis and changes in management in 85% of cases.

• Specialists showed substantial agreement when reviewing videos, with complete agreement for nonepileptic and physiologic events.

• Stand‐alone videos have the potential to be a valuable adjunct in the evaluation of seizures and seizurelike events.

1. INTRODUCTION

Inpatient electroencephalographic (EEG)‐video monitoring is considered the gold standard for seizure diagnosis, primarily through recording the events in question. However, up to 23% of inpatient EEG‐video monitoring admissions fail to capture these events. ¹ Stand‐alone videos, such as those captured by smartphones and security cameras, have been shown to be a valuable tool and are increasingly used in clinical practice to assist in the diagnosis. ² , ³ , ⁴ , ⁵ Despite this evidence, the frequency of stand‐alone video submission and their contribution to diagnosis after inconclusive EEG‐video monitoring remain poorly characterized.

Our objective was to assess the diagnostic value, submission rate, and interrater reliability of stand‐alone videos obtained after inconclusive inpatient EEG‐video monitoring.

2. MATERIALS AND METHODS

We prospectively identified patients between September 2023 and November 2025 whose epilepsy monitoring unit (EMU) evaluation was inconclusive. EEG‐video monitoring was conducted in accordance with the American Clinical Neurophysiology Society guidelines. ⁶ Standard procedures were implemented to maximize the likelihood of capturing the events of interest, including antiseizure medication reduction, sleep deprivation, exercise, hyperventilation, and photic stimulation. As part of routine clinical care, patients were asked prior to discharge to provide a video of any typical event not captured during their admission. Inconclusive studies were defined as failure to record the events in question, and were subdivided as follows:

1. Completely inconclusive: No interictal epileptiform discharges (IEDs) and no events recorded

2. Partially inconclusive: No IEDs, some events recorded, but not all event types

3. Partially inconclusive: IEDs present, but no events recorded

4. Partially inconclusive: IEDs present, some events recorded, but not all event types

2.1. Review and classification process

Irrespective of the date of recording, videos were deemed eligible for analysis as long as they had not been submitted before the EMU evaluation and had not been previously shared with or evaluated by a physician. Submitted videos were deidentified and independently reviewed by two fellowship‐trained epileptologists who were not involved in the patient's care and were blinded to the patient's history. Reviewers were provided with all videos submitted by the patient, and the diagnostic process involved three stages:

1. Video‐level assessment: Reviewers classified each video independently.

2. Overall patient‐level diagnosis: Based on all submitted videos, reviewers provided a final diagnosis (or diagnoses) for the patient.

3. Adjudication: If consensus was not achieved between the two reviewers, a third reviewer adjudicated to determine the final classification.

2.2. Classification scheme

The following categories were used to classify both individual videos and patient‐level diagnoses:

1. Epileptic seizure

2. Psychogenic nonepileptic seizure (PNES)

3. Nonepileptic physiologic

4. Epileptic and PNES

5. Cannot be determined

2.3. Data analysis

Descriptive statistics were used to describe video submissions and diagnostic outcomes. Interrater agreement was assessed using Cohen kappa, which measures agreement beyond chance. ⁷ For ordinal ratings, weighted kappa with quadratic weights was applied to account for partial disagreement between adjacent categories. ⁸ , ⁹ Ninety‐five percent confidence intervals (CIs) were estimated using bootstrap resampling (1000 iterations).

2.4. Standard protocol approvals, registrations, and patient consents

This study was considered exempt from informed consent due to the study involving prospective observational data collected as part of routine clinical care, with no intervention or alteration of patient management.

3. RESULTS

3.1. EMU admissions overview

Over the 26‐month study period, 619 patients were admitted to the EMU. EEG‐video monitoring was inconclusive in 165 patients (27%), and completely inconclusive in half of these (13.5%), as shown in Figure 1. The average EMU length of stay was 3.5 days overall, compared to 4.4 days for patients with inconclusive results.

FIGURE 1.

Epilepsy monitoring unit (EMU) admission distribution. Flowchart depicts the categorization of patients (N = 619) based on diagnostic outcomes. Admissions are primarily classified as conclusive or inconclusive. Inconclusive cases are further stratified into subgroups based on the presence or absence of interictal epileptiform discharges (IEDs) and recorded clinical events. Bottom tiers indicate diagnostic yields for specific stand‐alone video subsets.

3.2. Video submission rate

Following an inconclusive admission, 27 patients (16%) submitted at least one video. On average, patients submitted 1–2 videos, with a median time of 25 days (range = 1–634) from EMU discharge. Of the 27 patients, 20 (74%) had undergone one or more additional EEG‐video monitoring studies prior to submitting a video.

3.3. Diagnostic value

Stand‐alone videos yielded a definitive conclusion in 23 patients (85%), defined as a finding that prompted change in patient management (Figure 2). The clinical contribution for each patient is summarized below and in Figure 1.

1. Established a new diagnosis in eight (30%)

2. Confirmed a suspected diagnosis in 10 (37%)

3. Clarified an event as physiologic in three (11%)

4. Identified a concomitant diagnosis in two (7%)

5. Were inconclusive in four (15%)

FIGURE 2.

Video breakdown of patient‐level diagnosis, showing percentage of clinical diagnoses established through the analysis of stand‐alone video recordings (n = 27).

3.4. Interrater reliability

A third reviewer was required for six videos (22%) due to diagnostic discrepancies between the initial two reviewers. For these videos, the discrepant classifications were between the following diagnostic categories:

• Two videos (33%): epileptic versus PNES

• Two videos (33%): epileptic and PNES versus epileptic

• One video (17%): cannot be determined versus physiologic

• One video (17%): cannot be determined versus PNES

Cohen kappa demonstrated substantial interrater agreement (κ = .690, 95% CI = .456–.893). Weighted kappa, which more strongly penalizes larger categorical discrepancies, yielded a comparable level of agreement (κ = .634, 95% CI = .216–.910), supporting reliable scoring across all five ordinal categories. ⁹ Agreement distribution and classification criteria for interrater agreement are detailed in Table S1 and Figure S1.

When examining agreement by event type and based on the final diagnoses, events classified as nonepileptic psychogenic or nonepileptic physiologic showed perfect interrater agreement, whereas interpretations resulting in epileptic seizures or “cannot be determined” accounted for all initial disagreements, as illustrated in Table S2.

4. DISCUSSION

In this prospective observational study, EEG‐video monitoring was inconclusive in 27% of all admissions. When compared to a study ¹ from 22 years ago (23%), the likelihood of nondiagnostic admissions has remained relatively consistent over time. These findings highlight the need for complementary diagnostic approaches when EEG‐video monitoring remains inconclusive.

Despite high diagnostic yield and substantial interrater agreement, obtaining stand‐alone videos after inconclusive EMU admissions proved challenging, with only 16% of patients submitting recordings. During follow‐up visits, patients reported barriers such as difficulty capturing acute events and skepticism about the value of videos compared to conventional diagnostic tools such as EEG and MRI. This perception is also shared by neurologists and other health care professionals, with the clinical value of videos often underemphasized relative to EEG. ⁴ , ⁵ , ¹⁰

Among patients who submitted a stand‐alone video, 74% had undergone one or more additional inconclusive EEG studies, further increasing health care utilization and cost. Although repeat EEG‐video monitoring, whether inpatient or ambulatory, can occasionally yield a diagnosis, inconclusive results often lead physicians to repeat the same studies or consider empiric medication adjustments based on clinical history alone. In such situations, stand‐alone videos offer an inexpensive and accessible alternative and should be incorporated into the diagnostic strategy. We suspect that repeated reinforcement of instructions, and the rationale for video collection at multiple points during care, may improve compliance with video submission.

Despite their value, stand‐alone videos have limitations. It is often difficult for witnesses to record an event, and the technical quality of the videos is variable. ¹⁰ Recordings that capture only the later part of an event may be insufficient for diagnosis, and brief clinical signs might need EEG for confirmation. While the diagnostic yield highlights the clinical utility of stand‐alone videos, this figure reflects a self‐selected cohort and should be interpreted as the tool's potential yield when used in this specific clinical context.

Interrater reliability in video analysis also presents challenges. ⁵ , ¹⁰ , ¹¹ , ¹² , ¹³ In this study, the two initially assigned epileptologists disagreed on 22% of videos, reflecting the inherent interrater variability also reported in EEG‐video ¹⁴ and routine EEG ¹⁵ interpretation. As in prior EEG‐video studies, most discrepancies arose from a small subset of videos with subtle or brief clinical signs. In contrast, clear PNES and clear epileptic seizures (i.e., bilateral tonic–clonic seizures) demonstrated strong interrater agreement. Additionally, once all results were finalized, and solely for the purpose of internal education, all epileptologists convened to review the discrepant cases. A common observation was that clinicians may have felt compelled to provide a definitive diagnosis rather than choose “cannot be determined.” Recognizing that video findings may be inconclusive is just as important as reaching a definitive diagnosis, as it justifies the need for continued evaluation. Some cases should be acknowledged as “cannot be determined,” underscoring the necessity of keeping an open mind about the differential diagnoses.

5. CONCLUSIONS

This study provides preliminary evidence that stand‐alone videos are a valuable diagnostic adjunct for seizures and seizurelike events following an inconclusive EMU admission. Although challenging to obtain, they have the potential to reduce unnecessary repeat monitoring and empiric medication modifications, thereby decreasing both patient burden and health care costs. As digital tools become increasingly incorporated into clinical care, ongoing efforts are needed to further emphasize the role of videos in epilepsy clinical evaluation.

AUTHOR CONTRIBUTIONS

Marina O. Azevedo: Conceptualization (lead); methodology (lead); investigation (lead); visualization (lead); writing − original draft (lead); writing − review and editing (equal). Ushtar Amin: Writing − review and editing (supporting); investigation (supporting). Angelica Rivera‐Cruz: Investigation (supporting). Kristen Zemina: Investigation (supporting). Mark Armanious: Investigation (supporting). Ana Omura: Investigation (supporting); visualization (supporting). Nicholas Tenney: Investigation (supporting). Yangxin Huang: Formal analysis (lead). Selim R. Benbadis: Conceptualization (lead); methodology (lead); investigation (lead); writing − original draft (lead); writing − review and editing (equal).

CONFLICT OF INTEREST STATEMENT

None of the authors has any conflict of interest pertinent to this research to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Appendix S1.

DATA AVAILABILITY STATEMENT

Deidentified data and analyses are available upon request.