A blinded comparison of continuous versus sampled review of video-EEG monitoring data

https://doi.org/10.1016/j.clinph.2010.10.048Get rights and content

Abstract

Objective

While there are well-established guidelines for optimum video-EEG monitoring (VEM), the process of reviewing VEM data varies amongst centres. In this study, we compared continuous with sampled reviewing of VEM data to assess whether their diagnostic yield differs.

Methods

VEM data acquired from 50 consecutive patients (31 females) admitted for VEM were reviewed by two independent electroencephalographers, one using the continuous review method, and the other sampling the first five minutes of each hour together with events identified by push buttons and automated spike detection software. Overall agreement between reviewers was calculated using the Kappa statistic. Comparison between the total number of clinical events detected by the two methods was done by Pearson’s correlation coefficient.

Results

A substantial number of events were missed using sampled review. Despite this, there was excellent agreement between the two methods on the final electro-clinical diagnosis for each patient (Kappa = 0.89).

Conclusion

In our laboratory, continuous VEM more comprehensively captured information of interest, but it did not substantially alter the final electro-clinical diagnosis.

Significance

Sampled review of VEM data captures sufficient data to reliably make accurate clinical decisions. It may be considered as a more cost and labor efficient alternative to continuous review.

Highlights

► We compared continuous to sampled reviewing of video-EEG monitoring data to assess whether their diagnostic yield differs. ► A substantial number of events were missed using sampled review. ► Despite this, sampled review of video-EEG monitoring data did not alter the final electro-clinical diagnosis.

Introduction

Long term video-EEG monitoring (VEM) is an established method for the diagnosis and characterization of epileptic seizures and syndromes (American Electroencephalographic Society, 1994, Gates and Hemmes, 1990, Ghougassian et al., 2004, Gotman et al., 1985, Gumnit, 1987, Lagerlund et al., 1996, Panych et al., 1985). The considerable diagnostic and economic implications of differentiating epileptic from non-epileptic seizures as well as precise seizure localization in potential surgical candidates led to the development of well-established guidelines for optimum EEG recording and video telemetry by the American Electroencephalographic Society (1994) and most laboratories follow standardized protocols to ensure accurate data acquisition and clinical interpretation. However, the process of reviewing VEM data varies from centre to centre (American Electroencephalographic Society, 1994, Gates and Hemmes, 1990, Ghougassian et al., 2004, Gotman et al., 1985, Gumnit, 1987, Lagerlund et al., 1996, Panych et al., 1985). Many centres, including ours, review the entire VEM dataset, while other centres use a sampled approach whereby a set amount of VEM data is reviewed per hour of recording (typically 5–10 min), together with the data recorded in relation to events identified by button presses or by seizure or spike detection programs. These samples of VEM data are then subjected to close review. To date, there is no evidence to support the superiority or diagnostic yield of one method versus the other for electro-clinical diagnosis.

One unanswered question is whether scrutiny of all VEM data offers an advantage in terms of detecting the number and type of interictal and ictal (especially subclinical) events over the sampled review method. More importantly, it is not known whether the electro-clinical diagnosis for a given patient may differ depending on the method used to review VEM data. This could have important implications for diagnostic accuracy and quality of care at the individual patient level, and for workforce requirements and efficiency at a systems level. To our knowledge, no study directly comparing these two data analysis protocols has been performed. Thus, the objective of this study was to compare whether the electro-clinical diagnostic yield of continuous reviewing of VEM data was superior to sampled reviewing of VEM data. Our hypothesis was that there would be no significant difference between the two methods.

Section snippets

Methods

VEM data acquired from 50 consecutive patients admitted to the seizure monitoring unit at our centre between August 2005 and May 2006 was reviewed independently using two review methods. The first was the continuous review method which was performed by three of the study co-investigators (PF, NJ, NP). The second method sampled the first five minutes of each hour together with events identified by button presses (patient, relative, or health care provider) and by the automated spike detection

Results

Of the 50 patients studied, 31 were female and 44 were right handed. The mean age was 36.6 years (range 20–62 years) and the mean age of seizure onset was 19.7 years (range from birth to 47 years). The patients were monitored for periods ranging between 3 and 21 days. The average amount of time spent to review all interictal EEG data over a 24 h period was 45.8 min per patient, whereas the average amount of time spent to review the sampled recording for the same duration of EEG data was 15.0 min per

Discussion

Although much work has been done to substantiate the value of long-term VEM and enhanced computer assisted methods of data reduction and analysis (American Electroencephalographic Society, 1994, Gates and Hemmes, 1990, Ghougassian et al., 2004, Gotman et al., 1985, Gumnit, 1987, Lagerlund et al., 1996, Panych et al., 1985), no study has investigated whether sampled review of VEM data yields comparable electro-diagnostic results to a complete and exhaustive review of the same data. It is

Disclosures

None of the authors has any conflict of interest to disclose.

Acknowledgements

We are thankful for the expert assistance of Ms. Vicky Stagg and the Neuroscience Clinical Research Unit of the Hotchkiss Brain Institute in preparing the database and data analysis. This work was supported by the Alberta Heritage Foundation for Medical Research (AHFMR) and the Canadian Institutes of Health Research (CIHR).

References (9)

There are more references available in the full text version of this article.

Cited by (9)

  • Using sampled visual EEG review in combination with automated detection software at the EMU

    2020, Seizure
    Citation Excerpt :

    One study suggested that the first hour of sleep reliably predicts the occurrence of interictal epileptiform activity for whole recording [3]. Another study showed that sampled review was non-inferior regarding final electro-clinical diagnosis, although a substantial number of events was missed [4]. Another approach is automated EEG analysis, using detection software.

  • Long-term EEG in adults: Sleep-deprived EEG (SDE), ambulatory EEG (Amb-EEG) and long-term video-EEG recording (LTVER)

    2015, Neurophysiologie Clinique
    Citation Excerpt :

    On study compared the reading of LTVER by two independent operators, one using the continuous review method, and the other sampling the first five minutes of each hour together with events identified by push buttons and automated spike detection software. A substantial number of events were missed by the sampled review, yet in spite of this there was excellent agreement between the two methods on final diagnosis for each patient and on the impact on care management [8]. Furthermore, some studies showed some intra-observer discordance during the analysis of events recorded in video-EEG, especially for PNES [14].

  • French Guidelines on electroencephalogram

    2014, Neurophysiologie Clinique
  • Video-electroencephalography investigation of ictal alterations of consciousness in epilepsy and nonepileptic attack disorder: Practical considerations

    2014, Epilepsy and Behavior
    Citation Excerpt :

    Although time to initial event after electrode placement does not differ between diagnoses, clinical events prior to or during placement were mainly nonepileptic attacks, while events at night were mainly epileptic seizures [56]. These findings also have practical implications for the development of internal policies on the level of monitoring required [57–59] in order to optimize allocation of health-care resources. From the clinical point of view, the management of these two patient populations should comply with different principles, as in patients where the alteration of consciousness has been related to psychological factors, reinforcement of the sick role has been shown to be detrimental with regard to long-term outcome [60].

  • Performance-power consumption tradeoff in wearable epilepsy monitoring systems

    2015, IEEE Journal of Biomedical and Health Informatics
View all citing articles on Scopus
View full text