Keywords
high-frequency oscillations - intractable epilepsy - intracranial electrography -
epilepsy prognosis - electrocorticogram
Introduction
The aim of an epilepsy surgery is to resect the brain tissue responsible for seizure
generation, known as the epileptogenic zone (EZ). Prior to surgery, the brain tissue
undergoes evaluation to identify the seizure onset zone (SOZ), which refers to the
cortical region that initiates seizures. Accurate identification of the EZ holds the
potential to improve patient prognosis.
In recent years, high-frequency oscillations (HFOs) have emerged as a promising marker
for epileptogenicity. HFOs can be categorized into two frequency ranges: “ripples”
(80–250 Hz) and “fast ripples” (FR; 250–500 Hz).[1] HFOs have demonstrated superior effectiveness in specifically identifying the SOZ
compared with spikes.[2]
[3]
[4]
[5] Previous studies have consistently shown a strong association between HFOs and the
SOZ,[6]
[7]
[8] surpassing that of interictal epileptiform discharges.[9] Significant disparities in HFOs rates and amplitudes have been observed between
the SOZ and non-SOZ regions in various brain lobes.[10]
[11] Hence, HFOs play a crucial role in the detection and delineation of epileptic regions.
Several studies have established a close relationship between HFOs and surgical outcomes.
Complete removal of HFO-producing brain regions has been linked to favorable postoperative
outcomes,[12]
[13]
[14] while incomplete removal correlated with epileptic recurrence.[4]
[15]
[16]
[17]
[18]
However, no significant distinction has been found in the delineation of the SOZ using
HFOs instead of spikes.[19]
[20] Gloss et al's meta-analysis of two prospective studies did not yield a significant
improvement in postoperative seizure outcomes with the use of HFOs.[21] Similarly, Höller et al reported minor differences in their meta-analysis of HFOs
pertaining to epilepsy outcomes.[22] Therefore, the reliability of HFOs as a biomarker for epileptic tissue remains a
subject of ongoing debate.
HFOs can be monitored using scalp electroencephalogram (EEG), electrocorticogram (ECOG),
and stereoelectroencephalography (SEEG). Scalp EEG is prone to artifacts and errors
due to slight movements of the monitored individual, making EZ delineation challenging.
ECOG, consisting of a grid or strip electrode that covers the brain cortex, offers
denser contacts and fewer artifacts compared with scalp EEG.
SEEG, on the other hand, allows for monitoring of electrical activities in deep brain
structures over a wider range. It provides detailed recordings of the origin and transmission
of epileptic discharges, facilitating a better understanding of the transmission network
of the cerebral cortex and accurate EZ delineation.[23]
[24]
[25] In patients in whom noninvasive methods fail to satisfactorily identify epileptic
tissue, intracranial electroencephalography (iEEG) monitoring serves as the gold standard
for EZ delineation.[26]
[27]
[28]
We conducted a meta-analysis to evaluate the association between postoperative efficacy
and HFOs monitored using ECOG and SEEG.
Methods
Inclusion Criteria
The inclusion criteria were the following: prospective or retrospective studies investigating
surgical outcomes in patients with intractable epilepsy using iEEG monitoring and
studies providing information on the proportion of HFO resection and patient prognosis.
Studies were excluded if they involved duplicate patient records or were not written
in English.
Study Search
Two independent reviewers conducted a comprehensive search of PubMed and Cochrane
databases up to January 2023, using the following retrieval strategy:
-
PubMed: ((high frequency oscillations) OR (fast oscillations) OR (ripples)) AND ((epilepsy)
OR (seizure)) AND ((surgery) OR (operation)).
-
Cochrane: (high frequency oscillations) OR (fast oscillations) OR (ripples) AND (epilepsy)
OR (seizure) AND (surgery) OR (operation).
Data Extraction
The following information was extracted by the author: type of iEEG used (ECOG or
SEEG), patient characteristics including mean age at the time of surgery, sample size,
sex ratio, mean follow-up time, and recorded outcomes. Additionally, details regarding
the type of HFOs (ripples or FR), analysis of HFOs (visual or automatic), and the
proportion of HFOs removed were collected.
Quality Assessment
The quality of the included studies was evaluated by two reviewers using Cochrane's
risk of bias tool.[29] In cases of disagreement, a third author was consulted to reach a resolution. The
following aspects were assessed for risk of bias, categorized as low, high, or unclear:
age at surgery, iEEG methodology, detection and thresholding of HFOs, sample size,
representation of females, follow-up duration, and recorded outcomes.
Data Classification and Analyses
The patients from the included studies were categorized into four groups based on
the extent of HFOS removal: completely removed FR (C-FR), completely removed ripples
(C-Ripples), mostly removed FR (P-FR), and mostly removed ripples (P-Ripples). A meta-analysis
was conducted to investigate the relationship between seizure outcomes and the proportion
of removed HFOs. However, some studies in the P-Ripples and P-FR groups did not provide
specific criteria for determining the majority of HFOs removed.
To define the majority of HFOs removed, we applied the equation listed below.[13] The criteria were as follows: if the value of RatioChanns (ev) was ≥0 and ≤1, it
was classified as the majority of HFOs removed; otherwise, it was classified as the
majority of HFOs untouched.
Here, ev represented HFOs (FR or ripples), ChannRem represented the channels of events
that had been removed, ChannNonRem represented the channels of events that had not
been removed, and RatioChanns(ev) represented the difference between the number of
removed channels (ev) and the number of unremoved channels (ev), divided by the total
number of channels (ev). The value of RatioChanns(ev) fell between 1 and –1. A value
between 0 and 1 (including 0) indicated that the majority of events were removed,
while a value between –1 and 0 (excluding 0) indicated that the majority of events
were untouched.
Statistical Analyses
Review Manager 5.4 software was used to analyze the collected data. Patients without
either ripples or FR were excluded from their respective studies. The assumption of
heterogeneity was assessed using the chi-square-based Q test. A p value of ≥0.10 indicated a lack of heterogeneity among the studies. The fixed-effects
model was used to calculate the odds ratio (OR) and 95% confidence interval (CI).
Results
Study Screening Results
A total of 216 records were retrieved from PubMed and 8 records were obtained from
the Cochrane Library, all of which met the inclusion criteria. After reviewing the
titles and abstracts, 20 studies were selected for further evaluation ([Fig. 1]). However, only 10 studies provided the necessary data upon full-text review. It
should be noted that one study by Hussain et al[30] was excluded due to patient overlap with their previous study,[31] which we already included. Consequently, a total of nine studies were included in
this meta-analysis.
Fig. 1 Flow diagram of study selection. HFOs, high-frequency oscillations.
Quality Assessment
The Cochrane risk of bias assessment[32]
[33] revealed that none of the included studies reported random allocation, allocation
concealment, or implementation of blinding methods.
Clinical Characteristics
Most of the included studies provided demographic information, such as the age and
gender of the patients (see [Table 1]). Four studies exclusively involved minors.[31]
[34] One study did not specify the follow-up time,[34] while another study reported a small number of patients lost to follow-up.[35] Additionally, one study did not mention the criteria for prognostic assessment,[35] but defined seizure freedom at each follow-up visit as the absence of breakthrough
seizures since the last clinical visit. For our analysis, we defined seizure freedom
based on Engel I or International League Against Epilepsy I (ILAE I) classification.
A study conducted by Fedele et al[36] evaluated the efficacy of C-FR and ripples, providing data on the removed HFOs for
each patient. Hence, we were able to classify their patients into the P-Ripples or
P-FR group.
Table 1
Basic information of the included research literature
Study
|
Age at surgery (y)
|
iEEG
|
Detection
|
Threshold
|
HFOs
|
Sample
|
Female
|
Follow-up
|
Outcome
|
Akiyama et al[41]
|
10.68 ± 5.0
|
SEEG/ECoG
|
Auto
|
Bootstrapping
|
R/FR
|
28
|
–
|
≥2 y
|
LIAE
|
Fujiwara, 2012
|
–
|
ECoG
|
Auto
|
1/min
|
FR
|
41
|
–
|
≥12 mo
|
–
|
van Klink, 2014
|
18.43 ± 10.39
|
ECoG
|
Auto
|
1/min
|
R/FR
|
14
|
7
|
≥12 mo
|
Engel
|
Okanishi et al[40]
|
9.35 ± 5.26
|
ECoG
|
Auto
|
Bootstrapping
|
R/FR
|
10
|
4
|
≥19 mo
|
Engel
|
Fujiwara et al[34]
|
5.96 ± 4.43
|
ECoG
|
–
|
–
|
R/FR
|
14
|
7
|
–
|
LIAE
|
Hussain et al[31]
|
8.93 ± 5.34
|
ECoG
|
Visually
|
–
|
FR
|
30
|
15
|
52.84 ± 26.29 mo
|
–
|
Fedele et al[36]
|
32.1 ± 11.5
|
SEEG/ECoG
|
Auto
|
95%
|
R/FR
|
20
|
6
|
25.1 ± 12.5 mo
|
ILAE
|
Jacobs et al[39]
|
23.2 ± 17.2
|
SEEG/ECoG
|
Visually/auto
|
–
|
R/FR
|
52
|
34
|
≥12 mo
|
Engel
|
Nariai et al[35]
|
13.1 ± 5.4
|
SEEG/ECoG
|
Visually
|
–
|
FR
|
19
|
10
|
–
|
–
|
Abbreviations: ECoG, electrocorticogram; FR, fast ripples; HFO, high-frequency oscillation;
iEEG, intracranial electroencephalography; ILAE, International League Against Epilepsy;
R, ripples; SEEG, stereoelectroencephalography.
Group Classification Based on Removed Ratio
The distribution of studies among the groups was as follows: 8 studies in the C-FR
group, 5 studies in the C-Ripples group, 5 studies in the majority removed FR group,
and 5 studies in the majority removed ripples group.
Completely Removed FR and Prognosis
[Fig. 2] illustrates the meta-analysis results regarding the relationship between C-FR and
patient prognosis. The fixed-effects model was utilized, and the heterogeneity test
indicated no significant heterogeneity (I
2 = 0%; p = 0.68). The estimated model yielded a significant result, indicating that patients
in the C-FR group had a significantly higher prognosis compared with those with incomplete
removal (OR = 6.62; 95% CI: 3.10–14.15; p ≤ 0.00001).
Fig. 2 Completely removed fast ripples (FR) and outcome. In seven studies, the confidence
intervals (CIs) overlap with 0, indicating no clear difference in outcome between
patients based on these seven studies alone. However, the pooled meta-analysis demonstrates
a significant positive difference that does not overlap with 0. Therefore, it can
be concluded that patients with completely removed FR have a better prognosis compared
with those with incompletely removed FR (odds ratio [OR] = 6.62; 95% confidence interval
[CI]: 3.10–14.15; p < 0.00001).
The funnel plot, displayed in [Fig. 3], exhibits symmetrical distribution of points along the center line and both sides,
suggesting no noticeable publication bias among the included studies.
Fig. 3 Funnel plot. OR, odds ratio; SE, standard error.
Completely Removed Ripples and Prognosis
[Fig. 4] presents the meta-analysis results concerning the association between C-Ripples
and patient prognosis in the C-Ripples group. The fixed-effects model was employed,
and the heterogeneity test indicated no significant heterogeneity (I
2 = 0%; p = 0.59). In the C-Ripples group, the estimated model yielded a significant result
(OR = 4.45; 95% CI: 1.33–14.89; p = 0.02). Patients with C-Ripples demonstrated a significantly better prognosis compared
with those with incomplete removal.
Fig. 4 Completely removed ripples and outcome. In four studies, the confidence intervals
(CIs) overlap with 0, suggesting no clear difference in outcome between patients based
on these four studies individually. However, the pooled meta-analysis reveals a positive
difference without overlap with 0, indicating that the prognosis of patients with
completely removed ripples is better than that of patients with incompletely removed
ripples (odds ratio [OR] = 4.45; 95% CI: 1.33–14.89; p = 0.02).
Majority of FR Removed and Prognosis
Data regarding the majority of FR removed was obtained from five studies. Since Fedele
et al[36] did not specify the criteria for the majority removed FR, we applied the equation
listed earlier[13] to define the majority removed FR. [Fig. 5] presents the meta-analysis results regarding the relationship between the majority
of FR removed and patient prognosis in the P-FR group.
Fig. 5 Majority of fast ripples (FR) removed and outcome. In three studies, the confidence
intervals (CIs) overlap with 0, indicating no clear difference in outcome between
patients based on these three studies alone. Nevertheless, the pooled meta-analysis
demonstrates a significant positive difference that does not overlap with 0. Thus,
it can be concluded that the prognosis of patients with the majority of FR removed
is better than that of patients with the majority of FR untouched (odds ratio [OR] = 6.23;
95% CI: 2.04–19.06; p = 0.001).
The fixed-effects model was utilized, and the heterogeneity test indicated no significant
heterogeneity (I
2 = 0%; p = 0.68). The estimated model yielded a significant result, indicating that patients
with the majority of FR removed had a significantly higher prognosis compared with
those with the majority of FR untouched (OR = 6.23; 95% CI: 2.04–19.06; p = 0.001).
Majority of Ripples Removed and Prognosis
Similar to the previous analysis, data on the majority of ripples removed were obtained
from the aforementioned study. The same method was applied to the data from Fedele
et al[36] to define the criteria for the majority of ripples removed. [Fig. 6] depicts the meta-analysis results for the majority of ripples removed and patient
prognosis in the P-Ripples group. The fixed-effects model was used, and the heterogeneity
test indicated no significant heterogeneity (I
2 = 0%; p = 0.67). The estimated model yielded a significant result, showing that patients
with the majority of ripples removed had a significantly better prognosis compared
with those with the majority of ripples untouched (OR = 8.14; 95% CI: 2.62–25.33;
p = 0.0003). The seizure-free ratio was higher in the majority of ripples removed groups
than in the majority of ripples untouched groups.
Fig. 6 Majority of ripples removed and outcome. In three studies, the confidence intervals
(CIs) overlap with 0, suggesting no clear difference in outcome between patients based
on these three studies individually. However, the pooled meta-analysis reveals a positive
difference without overlap with 0, indicating that the prognosis of patients with
the majority of ripples removed is better than that of patients with the majority
of ripples untouched (odds ratio [OR] = 8.14; 95% CI: 2.62–25.33; p = 0.0003).
Discussion
This study included a total of nine studies involving 228 patients, which were categorized
into four groups based on the removal of HFOs: C-FR, C-Ripples, P-FR, and P-Ripples.
The estimated model for all groups revealed that removing a larger region of HFOs
could significantly enhance patient prognosis.
Prospective Studies
Currently, there is no standardized method for HFO detection, and interrater reliability
remains a major challenge.[37]
[38] Notably, a previous meta-analysis by Gloss et al[21] that included two prospective studies reported no evidence supporting the use of
HFOs to improve the efficacy of epileptic surgery. However, these findings are inconsistent
with previous studies, likely due to variations in evaluation criteria and analysis
methods for HFOs. In a recent prospective multicenter study by Jacobs et al,[39] which was one of the studies included in our analysis, complete removal of HFOs
did not demonstrate a significant association with the efficacy of epileptic surgery.
Nevertheless, their data from a single center suggested that completely removing HFOs
was more likely to improve patient outcomes, consistent with their previous studies.
The consistent positive results from this single-center study may be attributed to
Jacobs et al[39] using uniform criteria for recording and analyzing HFOs. Therefore, conducting prospective
studies using consistent methods could yield more reliable and significant results.
A study by Nariai et al,[35] also included in our analysis, was the first prospective study to provide evidence
linking complete removal of visually recognized FR in ECoG with postoperative prognosis
in epilepsy patients, supporting the findings of our meta-analysis. However, it is
worth mentioning that similar prospective studies have included a limited number of
cases, which reduces the credibility of their results. This limitation mainly stems
from the high technical requirements and economic costs associated with these studies.
Therefore, in the future, conducting more prospective studies with consistent criteria,
either through a multicenter collaboration or systematic evaluations, would be beneficial
in reducing heterogeneity among studies.
Heterogeneity
Studies by Okanishi et al[40] and Fujiwara et al[34] focused on minors with multiple tuberous sclerosis, while Akiyama et al[41] and Hussain et al[31] also exclusively included minors. Subgroup analysis of these four studies could
be conducted if necessary to explore the relationship between HFOs and surgical efficacy
in patients with multiple tuberous sclerosis complex or intractable epilepsy among
minors. In a prospective multicenter study by Jacobs et al,[39] artificial visual recognition of HFOs was employed to test the effectiveness of
automatic HFO detection. However, visual recognition is subjective, and the reliability
of automatic HFO detection remains a subject of debate. Therefore, further studies
are warranted to investigate this aspect.
Automated Detection
In recent years, many researchers have shifted toward the use of newly developed automatic
detection methods instead of visual detection. Höller et al[22] discovered that automatic detection outperforms visual detection. Moreover, when
it comes to detecting HFOs in the operating room immediately after resection, automatic
detection becomes essential. However, it should be noted that many detectors require
offline processing of recorded signals to apply automatic or semiautomatic artifact
elimination stages for the removal of events incorrectly classified as HFOs. Therefore,
the development of automatic detection has become increasingly significant.
To comprehend the superiority of automatic detection over visual detection, researchers
are increasingly emphasizing the need to compare the relationship between HFO removal
and outcomes to verify the efficiency of the detectors in reverse. Previous studies
have not established a clear relationship between the two. However, Burelo et al[18] proposed a detector based on a spiking neural network (SNN) and spectral analysis
to explore this relationship, offering the possibility of real-time detection of HFOs
during epilepsy surgery.
Predicting Individual Outcome
Although the analysis at the group level revealed a correlation between the removal
of HFO-generating tissue and seizure freedom, it should be noted that accurate prediction
of patient outcomes based on HFO removal is limited to a subset of patients.[13]
[14] In contrast to population-level results, it has been observed that HFOs are not
consistently effective in identifying seizures at an individual patient level, as
they exhibit similarities to spikes.[19] This discrepancy may arise due to the challenge of differentiating physiological
HFOs from pathological HFOs. There is significant frequency overlap between these
two types of HFOs,[42]
[43]
[44]
[45] and relying solely on frequency and amplitude analysis is inadequate for distinguishing
between them.[46]
[47] Therefore, achieving a better distinction between physiological and pathological
HFOs may enhance the ability to predict the prognosis of individual patients.
Postoperative residual HFOs also hold importance in predicting the prognosis of individual
patients. Many researchers have emphasized the significance of using residual HFOs
to predict epilepsy outcomes. They have found that residual HFOs observed in ECoG
recordings may provide more reliable predictions of seizure outcomes compared with
preoperative HFOs.[4]
[48]
Conclusion
Our study reveals a significant correlation between the removal of HFOs and achieving
seizure freedom. Specifically, the removal of a larger region containing HFOs is associated
with favorable postoperative seizure outcomes. This meta-analysis highlights the crucial
role of HFOs in identifying the EZ and guiding surgical strategies. However, to validate
these findings, it is imperative to conduct prospective multicenter studies with standardized
protocols and methods. Furthermore, additional research is needed to investigate the
impact of residual HFOs on seizure outcomes. The efficacy of residual HFOs relies
on the development of more advanced techniques for automatic intraoperative detection
of HFOs. Hence, the future advancement of automatic intraoperative detectors holds
great importance in this regard.