| |

Received: 19 February 2021    Revised: 15 June 2021    Accepted: 15 July 2021

DOI: 10.1111/epi.17024

F U L L - ­L E N G T H O R I G I N A L R E S E A R C H

Improving the prediction of epilepsy surgery outcomes

using basic scalp EEG findings

Zachary Fitzgerald1   | Marcia Morita-­Sherman1   | Olivia Hogue2  |

Boney Joseph3   | Marina K. M. Alvim4   | Clarissa L. Yasuda4  | Deborah Vegh1  |
Dileep Nair1  | Richard Burgess1  | William Bingaman1  | Imad Najm1  | ­
Michael W. Kattan2  | Ingmar Blumcke5   | Gregory Worrell3  | ­
Benjamin H. Brinkmann3   | Fernando Cendes4   | Lara Jehi1

1
Epilepsy Center, Cleveland Clinic
Foundation, Cleveland, Ohio, USA Abstract
2
Quantitative Health Sciences, Objective: This study aims to evaluate the role of scalp electroencephalography
Cleveland Clinic Foundation, (EEG; ictal and interictal patterns) in predicting resective epilepsy surgery out-
Cleveland, Ohio, USA
3
comes. We use the data to further develop a nomogram to predict seizure freedom.
Department of Neurology, Mayo
Clinic, Rochester, Minnesota, USA Methods: We retrospectively reviewed the scalp EEG findings and clinical data
4
Department of Neurology, University of patients who underwent surgical resection at three epilepsy centers. Using
of Campinas, Campinas, Brazil both EEG and clinical variables categorized into 13 isolated candidate predic-
5
Institute of Neuropathology, University tors and 6 interaction terms, we built a multivariable Cox proportional hazards
Hospitals Erlangen, Erlangen, Germany
model to predict seizure freedom 2 years after surgery. Harrell's step-­down proce-
Correspondence dure was used to sequentially eliminate the least-­informative variables from the
Lara Jehi, Cleveland Clinic Epilepsy model until the change in the concordance index (c-­index) with variable removal
Center, S-­51, 9500 Euclid Ave,
Cleveland, OH, 44195, USA. was less than 0.01. We created a separate model using only clinical variables.
Email: [email protected] Discrimination of the two models was compared to evaluate the role of scalp EEG
in seizure-­freedom prediction.
Funding information
NIH, Grant/Award Number: R01 Results: Four hundred seventy patient records were analyzed. Following internal
NS097719 validation, the full Clinical + EEG model achieved an optimism-­corrected c-­index
of 0.65, whereas the c-­index of the model without EEG data was 0.59. The pres-
ence of focal to bilateral tonic-­clonic seizures (FBTCS), high preoperative seizure
frequency, absence of hippocampal sclerosis, and presence of nonlocalizable sei-
zures predicted worse outcome. The presence of FBTCS had the largest impact for
predicting outcome. The analysis of the models’ interactions showed that in pa-
tients with unilateral interictal epileptiform discharges (IEDs), temporal lobe sur-
gery cases had a better outcome. In cases with bilateral IEDs, abnormal magnetic
resonance imaging (MRI) predicted worse outcomes, and in cases without IEDs,
patients with extratemporal epilepsy and abnormal MRI had better outcomes.
Significance: This study highlights the value of scalp EEG, particularly the sig-
nificance of IEDs, in predicting surgical outcome. The nomogram delivers an

© 2021 International League Against Epilepsy

Epilepsia. 2021;00:1–12.  wileyonlinelibrary.com/journal/epi   |  1

 |
2       FITZGERALD et al.

individualized prediction of postoperative outcome, and provides a unique as-

sessment of the relationship between the outcome and preoperative findings.

KEYWORDS
epilepsy surgery, focal epilepsy, scalp EEG, surgery outcome

1  |  I N T RO DU CT ION
Key Points
For patients with pharmacoresistant focal epilepsy, re-
• We developed a nomogram and an online risk
sective brain surgery offers the best chance for seizure
calculator using scalp electroencephalography
freedom. Even though the surgical treatment of epilepsy
(EEG) data to facilitate clinical counseling on
has improved with the advent of new noninvasive evalu-
seizure outcome following resective surgery.
ation techniques like magnetic resonance imaging (MRI),
• The full Clinical + EEG model achieved a con-
positron emission tomography (PET), ictal single-­photon
cordance index of 0.65, whereas the concord-
emission tomography (SPECT), and magnetoencephalog-
ance index of the model without EEG data was
raphy (MEG), as well as invasive techniques like electro-
0.59.
corticography (ECoG) and stereo-­electroencephalography
• This nomogram highlights the importance of
(SEEG), the current chances of achieving seizure free-
interictal epileptiform discharges in the predic-
dom following resective surgery are still highly variable.
tion of resective epilepsy surgery outcome.
Between 40% and 80% of patients achieve complete post-
• The analysis of the model's interactions pro-
operative seizure freedom.1 This variability highlights the
vides a unique assessment of the relationship
need to pursue a deeper understanding of the predictive
between the preoperative laterality of interictal
factors involved in surgical outcome.
epileptiform discharges and surgical outcome.
Previous studies have examined the associations be-
tween preoperative findings and postoperative seizure
outcome and have identified possible outcome determi-
nants.2 Overall, patients attain a higher rate of seizure free- focus on a single lobe of resection, or only focus on a par-
dom when there is a high degree of agreement between ticular feature of the EEG (eg, spikes, ictal spread, onset
multiple diagnostic modalities and when this agreement pattern).6,9–­12
suggests a well-­localized epileptogenic zone. However, In this study, we evaluate the predictive value of scalp
the prediction of surgical outcome performed by clini- EEG in epilepsy surgery outcome and specifically mea-
cians lacks precision because it is based on the overall im- sure its prognostic contribution beyond a patient's obvious
pression of how concordant the preoperative findings are. clinical characteristics. We accomplish this by developing
This type of prediction can sometimes be highly subjective nomograms, models for individualized risk prediction,
and experience dependent, as often reflected by the lack using scalp EEG data to predict seizure freedom at 2 post-
of agreement between clinicians.3 The development of operative years in a large, multicenter, deeply phenotyped,
statistical models that provide objective and quantifiable epilepsy surgery cohort.
predictions of seizure freedom on a case-­by-­case basis is
crucial to improving presurgical counseling and to expand
our understanding of the surgical outcome. 2  |  METHODS
Since its inception, electroencephalography (or EEG)
has been paramount in the evaluation of epilepsy. Scalp 2.1  |  Patient selection
EEG is a valuable “front line” tool to localize and delin-
eate the relative focality of the epileptic zone.4–­6 Except In this retrospective cohort study, we identified 470 pa-
for using scalp EEG to determine potential candidacy tients who underwent epilepsy surgery at Cleveland
for surgery, most of the current literature focusing on Clinic (n = 334), Mayo Clinic (n = 53), and University of
resective intervention comprises studies using invasive Campinas (n = 73) from 2010 to 2017. These study patients
methods.7,8 Furthermore, available studies that address had at least 6  months of postoperative follow-­up, and
the relationship between scalp EEG and the surgical out- no prior brain surgery, no multilobar resections, and no
come are often outdated, rely on relatively small samples, postoperative events of unclear nature. All patients were
FITZGERALD et al.      |  3

at least 10-­years-­old. Younger patients were excluded to to the first postoperative seizure. Patients who presented
avoid the potential confounding effects of age-­related dif- seizures exclusively in the acute phase (first month after
ferences observed in EEG patterns and in neuroplasticity. surgery) were classified as being seizure-­free. If seizures
Clinical data were collected from patients’ medical persisted beyond the first postoperative month, the date of
charts. All patients underwent a preoperative evaluation recurrence during the acute phase was used as the date of
that included, at minimum, MRI and EEG. When appro- the first postoperative seizure.
priate, SPECT, PET, MEG, invasive evaluation, and neuro-
psychological testing were also performed.
2.4  |  Statistical methods

2.2  |  EEG variables 2.4.1  |  Candidate predictors

We reviewed scalp video-­EEG reports. In all study sites, Demographic, clinical, and EEG candidate predictors
patients were admitted for a few days until seizures were were selected a priori, based on published literature and
recorded. EEG studies were performed using an extended clinical experience, and included the following: sex,17 age
International 10–­20 system (31 electrodes that include the at surgery,17–­19 age at epilepsy onset,17,19–­21 disease dura-
inferior temporal chain).13,14 tion,1 monthly seizure frequency,1,19,22,23 lobe of surgery
Interictal findings were coded based on laterality as bi- (temporal vs extratemporal),24–­26 MRI findings (normal
lateral interictal epileptiform discharges (IEDs), unilateral vs abnormal),24,27,28 presence or absence of focal to bilat-
IEDs (more than 80% of the discharges on one side15,16), eral tonic-­clonic seizures (FBTCS),17,22,24,29,30 pathological
or no IEDs. Based on the EEG report description, we cause (mesial temporal sclerosis [MTS], malformation
collected data on the typology of the IEDs (sharp waves, of cortical development [MCD], unknown, tumor, and
spikes/polyspikes, both, or no IEDs), on the percent of other),23,24,28 interictal EEG side (unilateral vs bilat-
temporal lobe IEDs, and on the frequency of discharges eral),16,31 presence or absence of nonlocalizable ictal pat-
(rare or frequent). The frequency of spikes was based on terns,7,32 number of ictal patterns, EEG concordance to
the impression reported by the EEG reader and not by a resected area,33 invasive evaluation, interictal typology
precise number. The percent of temporal lobe IEDs was (sharp waves, spike waves, both, and none), IEDs (frequent
based on the location and the estimated frequency of the vs rare), and location of IEDs (percent of temporal lobe
different subtypes of IEDs reported per patient. The loca- IEDs). To account for the expected variable correlations
tion was defined as the region with maximum electroneg- of EEG with outcome across different epilepsy etiologies
ativity. When the maximum electronegativity was located and localization, we also included several pre-­specified
in-­between lobes (eg, F7/F8, P7/P8) we used the spread of interaction terms as candidates for the model: MRI find-
the activity to define the location of the IEDs. For exam- ings and lobe of surgery each by interictal laterality, pres-
ple, if a patient had two types of IEDs—­that is,(1) sharp ence of nonlocalizable seizures, and the number of ictal
wave maximum F8 spreading to F4 (10%); (2) sharp wave patterns.
maximum F8 spreading to T4 (90%)—­we would classify
this patient as having 90% of temporal IEDs. Ictal findings
were classified according to the number of ictal patterns 2.4.2  |  Model-­building
(one or fewer vs two or more) and according to the pres-
ence or absence of nonlocalizable ictal patterns, which in- Descriptive statistics were calculated for the sample and
cluded generalized and nonlocalized ictal onset. are presented using mean with standard deviation, me-
Interictal and ictal EEG concordance was coded as con- dian with interquartile range, or count with proportion,
cordant with resection, discordant with resection (includ- as appropriate. A Cox proportional hazards regression
ing EEG findings that extended beyond the resection), model was fit to include all candidate clinical and EEG
normal EEG (interictal only), or nonlocalizable (ictal variables, including pre-­specified interaction terms.
only). Harrell's step-­down procedure was used to sequentially
eliminate the least informative variables from the model
until the change in the concordance index (c-­index) with
2.3  |  Seizure outcomes variable removal was less than 0.01.34 The c-­index ranges
from 0.50 and 1.0 and measures the model's ability to dis-
The primary outcome in this study was complete seizure criminate between high-­ and low-­risk patients. For a Cox
freedom at last follow-­up, allowing for isolated auras. Time model, the c-­index is an extension of the area under the
to seizure recurrence was measured as time (in months) receiver-­operating characteristic (ROC) curve (AUC), but
 |
4       FITZGERALD et al.

calculation of the c-­index considers the censoring inher- 2.4.4  |  Comparison with the clinical model
ent in time-­to-­event data, unlike the simple AUC more
commonly used for logistic regression. A value of 0.50 To investigate the relative effect of EEG predictors on sei-
indicates that the model is no better than chance, and a zure freedom at 2  years, EEG predictors were removed
value of 1.0 indicates the model correctly classifies 100% of from the model and model validation procedures were
patients. Restricted cubic splines were fit to each continu- repeated for the reduced model containing clinical predic-
ous variable to relax the assumption of linearity between tors only (“Clinical Model”). After confirming adequate
the predictor and the outcome, but nonlinear terms were calibration of the Clinical Model, the c-­index was com-
retained only in the final model if significant. Visual ex- pared to the Clinical + EEG model to determine whether
amination of Schoenfield residuals confirmed the propor- EEG variables increased discriminative ability compared
tional hazards assumption. to clinical predictors alone.

2.4.3  |  Model validation 2.4.5  |  Online risk calculator

A model will always perform better in the data set on which We generated an online risk calculator presenting the
it was built than when it is applied to a new set of data. estimated individualized chance for any given patient
Therefore, accurate assessment of discrimination and cali- for seizure-­freedom 2  years after surgery, using the
bration should be corrected for the optimism that arises final Cox model, R software, and previously published
when evaluating a model using model-­building data. Thus methods.38
we subjected our final model to internal validation to cor-
rect the c-­index for optimism using bootstrap resampling.
Bootstrap resampling has been found repeatedly to be a 2.4.6  |  Invasive evaluation
more stable, more efficient, and less biased method for
model validation than the more traditional split-­sample Considering the difficulty in interpreting the value of
approaches.35–­37 In detail, 500 bootstrapped data sets—­the scalp EEG without considering the influence of invasive
same size as the original—­were created through resam- data, we tested the variable “Invasive evaluation” as a po-
pling the original data with replacement. The 500 boot- tential stratification variable. The full data set was divided
strapped data sets are meant to simulate new samples from based on invasive status and the model was applied sepa-
the population, as each will vary in how similar it is to the rately to each group.
original, training data set, and the c-­index will be lower in
each bootstrapped data set when compared to the training
data. The Cox model is then applied to each bootstrapped 2.4.7  |  Statistical tools
data set, and the c-­index is calculated for each. The differ-
ences in c-­index between each bootstrapped sample and Analyses were conducted on complete cases using SAS
the original data were averaged to provide an estimate of Studio v.3.5 (SAS Institute) and R Studio v.3.3.0 (“rms”
optimism, which is subtracted from the raw c-­index to package).
yield the optimism-­corrected c-­index.35
Although external validation on a separate sample is
ideal to validate prediction models, our external sample 2.5  |  Classification of evidence
size precluded this work. Thus we elected to combine data
from all sites and perform an internal validation using This study provides Class IV evidence that scalp EEG find-
bootstrap resampling. Bootstrap-­based internal validation ings could enhance the individualized surgical outcome
methods have been shown to be reasonable approxima- prediction.
tions of how a model would likely perform in an external
sample.36
An optimism-­corrected calibration curve was created 2.6  |  Standard protocol approvals,
to compare predicted probabilities generated by the model registrations, and patient consent
to actual seizure freedom. A 45-­degree calibration curve
indicates the model is perfectly calibrated. A nomogram The Cleveland Clinic and Mayo Clinic Institutional
was created to provide a visual representation of each Review Boards, and the University of Campinas Ethics
model. Committee approved this study.
FITZGERALD et al.      |  5

2.7  |  Data availability statement

T A B L E 1   Characteristics of the sample

n (%) unless
Data not provided in the article will be available to any otherwise
qualified investigator upon request. specified
Site
Cleveland Clinic 344 (73.2)
3  |  R E S U LTS Mayo 53 (11.3)
UNICAMP 73 (15.5)
3.1  |  Demographics and clinical data Age at surgery (years), mean (SD) 37.5 (14.0)
Female 238 (50.6)
The sample included 470 patients from three centers
and was 51% female with an average age at surgery of Age at epilepsy onset (years), median (IQR) 15 (6–­28)

37.5 years. Forty-­eight percent of patients experienced sei- Time from epilepsy onset to surgery (years), 16 (7–­28)
zure recurrence following surgery, with a median time to median (IQR)

the first seizure of 4.1 months (interquartile range [IQR] Preoperative monthly seizure frequency 5 (2–­14)
1.0–­12.2). Among patients who did not experience seizure Side of surgery
recurrence, the median follow-­up was 42.1 months (IQR Left 253 (53.8)
21.0–­69.4). Additional demographic and clinical descrip- Right 217 (46.2)
tions of the sample can be found in Table 1. The median Temporal surgery 376 (80.0)
follow-­up time in the total sample was 44  months (IQR Pathological cause
22.0–­67.0).
MTS* 157 (33.4)
Unknown 121 (25.7)

3.2  |  Predictors of seizure freedom MCD 88 (18.7)

Tumor 36 (7.7)
3.2.1  |  Multivariable analysis and nomogram Other 59 (12.6)
development Missing 9 (1.9)
Normal MRI 91 (19.4)
The final model included the following variables: pres- Nonlocalizable seizures 104 (22.1)
ence or absence of FBTCS, preoperative monthly seizure Interictal EEG side
frequency, normal or abnormal MRI, pathological cause
>80% Unilateral 316 (67.2)
(recoded during model-­building to MTS vs non-­MTS),
Bilateral 110 (23.4)
temporal vs extratemporal surgery, presence or absence of
nonlocalizable seizures, interictal EEG laterality, and two No epileptiform discharges 44 (9.4)
interactions (between surgery site and interictal EEG lat- Ictal EEG concordance with resection
erality and between normal/abnormal MRI and interictal Concordant 262 (55.7)
EEG laterality). Full model coefficients can be found in Discordant 182 (38.7)
Table 2. Nonlocalized 18 (3.8)
Following internal validation, the full Clinical + EEG Missing 8 (1.7)
model achieved an optimism-­corrected c-­index of 0.65, in-
Interictal EEG concordance with resection
dicating that when presented with two patients at a given
Concordant 221 (47.0)
follow-­up time, one who had experienced seizure recur-
Discordant 205 (43.6)
rence and one who had not, the model would correctly
assign the patient with recurrence a lower probability of Normal EEG 44 (9.4)
seizure freedom 65% of the time. The optimism-­corrected Two or more ictal patterns 133 (28.3)
calibration curve for the model is presented in Figure 1. Interictal epileptiform typology
The model displays good calibration, as evidenced by the Sharp waves 255 (54.3)
calibration curve (blue) not substantially deviating from Spike waves 31 (6.6)
the ideal 45-­degree line (gray). The pattern of slight de- Both spike and sharp waves 129 (27.5)
viation observed is likely due to regression to the mean:
None 44 (9.4)
patients assigned a probability of seizure freedom below
Missing 11 (2.3)
0.4 in reality have a slightly higher probability of seizure
freedom, whereas patients assigned a probability of above (Continues)
 |
6       FITZGERALD et al.

T A B L E 1   (Continued)

n (%) unless
otherwise
specified
IED
Frequent 245 (52.1)
Rare 137 (29.2)
Missing 88 (18.7)
Percent of temporal lobe interictal epileptiform discharges
0% 100 (21.3)
100% 281 (59.8)
Between 0% and 100% 74 (15.7)
Missing 15 (3.2)
Invasive evaluation
Yes 150 (31.9)
No 320 (68.1)
F I G U R E 1   Calibration curve for the model predicting
*Includes two patients coded mesial temporal sclerosis + other.
probability of seizure freedom at 2 years. The x-­axis displays the
Abbreviations: IQR, interquartile range; SD, standard deviation.
predicted probabilities generated by the statistic model and the
y-­axis shows the proportion of the patients who were seizure-­free
at each predicted probability. The 45-­degree line (gray) indicates
T A B L E 2   Coefficients for each variable perfect calibration, whereas the blue line represents model-­
predicted probabilities compared with observed seizure freedom.
Variable Coefficient
The curve presented indicates that the model is well calibrated.
Presence of focal to bilateral tonic-­clonic 0.7551 The very minor deviation from the ideal line likely represents the
seizures common phenomenon of regression to the mean: patients with
Preoperative monthly seizure frequency 0.001 predicted probabilities of less than 0.5 likely have a slightly higher
Pathological cause = MTS −0.3278 actual probability, while patients with predicted probabilities of
more than 0.5 likely have a slightly lower actual probability
Presence of nonlocalizable seizures 0.4283
Interictal EEG side
>80% unilateral 0 (reference)
a c-­index of 0.59, a drop in the c-­index of 0.06 from the
Bilateral −0.1314 Clinical + EEG model.
No epileptiform discharges −1.2614
Normal MRI
If >80% unilateral interictal EEG −0.0218 3.2.2  |  Nomogram
If bilateral EEG −.3957
If no epileptiform discharges 0.4615 We created a nomogram to predict complete seizure free-
dom at 2 years (Figure 2). The manual application of the
Temporal resection
nomogram to predict complete seizure freedom requires
If >80% unilateral interictal EEG −0.4442
drawing a vertical line from the appropriate value of each
If bilateral EEG 0.2128
outcome predictor up to the “Points” line on top (eg,
If no epileptiform discharges 0.6642 Normal MRI  +  extratemporal  +  no IEDs  =  40 points).
Note: Coefficients for Normal MRI and Temporal Resection represent This is repeated for each individual patient character-
interactions with Interictal EEG side. istic. All the points are then added to yield a total. The
Abbreviations: MTS, mesial temporal sclerosis; Ref, reference. user will then find the total points in the ‘Total Points
Line” (second to last line on figure), and draw a vertical
0.7 in reality have a slightly lower probability of seizure line down to the estimated 2-­year probability of seizure
freedom. freedom.

Reduced clinical model Online risk calculator

After removal of all EEG variables from the model and The calculator runs all calculations in the background and
following internal validation, the Clinical Model achieved generates an individualized seizure-­freedom probability
FITZGERALD et al.      |  7

F I G U R E 2   Nomogram to predict the probability of complete seizure freedom after 2 years. The manual application of the nomogram to
predict complete seizure freedom requires drawing a vertical line from the appropriate value of each outcome predictor up to the “Points”
line on top. This is repeated for each individual patient characteristic. All the points are then added to yield a total. The end-­user would
then find the total points in the ‘Total Points Line” (second to the last line on figure), and draw a vertical line down to the estimated 2-­year
probability of seizure freedom

(The model is available at: https://riskc​alc.org/Clini​calAn​ with a relatively low spatial resolution and inability to
dEEGP​redic​tiono​fSeiz​ureFr​eedom​At2Ye​ars/). Figure assess subcortical structures, thereby limiting its role for
3 shows an example demonstrating the use and the output independently defining the exact location of the epilepto-
from the unified online risk calculator. genic zone. However, it has excellent temporal resolution
and is a pillar of epilepsy surgery evaluations. In essence,
scalp EEG is a ubiquitous test that is integral to getting the
3.2.3  |  Invasive evaluation patient to the point of surgery: optimizing its utility for
individualized outcome prediction after surgery becomes
There was not a difference in predictive performance of a low-­hanging fruit.
the model based on invasive status. Statistical models are not yet part of the current clin-
ical epilepsy practice; however, these algorithms have a
promising role in the era of precision care and personal-
4  |  DI S C USSION ized medicine.39–­41 The addition of a calculated risk score
to the traditionally subjective and variable prediction
Our results provide an updated assessment of extended of seizure outcome would enhance and modernize the
10–­20  scalp EEG (31 electrodes with a subtemporal preoperative epilepsy surgery counseling. Our study ad-
chain)13 in surgical epilepsy and its ability to predict post- vances that goal by providing individualized clinical and
operative seizure freedom. Scalp EEG has its limitations, scalp EEG analysis based on a large multicenter cohort.
 |
8       FITZGERALD et al.

F I G U R E 3   Example of online risk calculator to predict seizure outcome. The online risk calculator requires users to enter a
numerical value for “Preoperative Monthly Seizure Frequency” and select options from drop-­down menus for the remaining variables.
After clicking “Run Calculator,” a percent chance value of predicted probability of seizure freedom at 2 years will be displayed under the
“Probability” heading. For example, a patient presenting without generalized tonic-­clonic, 30 seizures a month preoperatively, mesial
temporal sclerosis, abnormal magnetic resonance imaging (MRI), sometimes non-­localizable seizures, unilateral interictal epileptiform
discharges, and a temporal resection will have a 75.1% chance of achieving seizure freedom following surgery

Our statistical methods represent current best practice for are seen in patients with MTS as compared to those with
prediction modeling and have been used successfully in other epilepsy causes is also in agreement with previous
other fields. Nomograms provide a visual assessment of studies.24 It is important to highlight that even though
the model's results, and this is the first nomogram includ- some variables statistically added some predictive value
ing scalp EEG to predict outcomes of epilepsy surgery. to the model, the individual contribution of each variable
was not necessarily substantial.
A clinical history of FBTCS has a major impact on
4.1  |  Interpretation of the the model. Unsurprisingly, FBTCS were associated with
predictive variables worse surgical outcomes, which is a well-­established
finding.29 Because FBTCS can indicate a more wide-
The final Clinical  +  EEG model achieved an optimism-­ spread epileptogenic zone,42,43 and at times can extend
corrected c-­index of 0.65 with good calibration (Figure 1), into eloquent cortex, full removal of the epileptogenic
and performed better than the Clinical model alone (c-­ zone is exceedingly difficult, resulting in lower rates of
index 0.59). Albeit modest, this improvement in predictive seizure freedom.
performance is still clinically helpful since this EEG infor- Nonlocalizable seizure pattern on EEG was the only
mation is obtained on every surgical patient and does not ictal pattern included in the final model, which predicted
therefore require any additional resources beyond routine worse surgical outcome. This finding agrees with the liter-
clinical practice. ature on patients who underwent both scalp and invasive
The clinical variables included in the final model are EEG studies.7 A few explanatory hypotheses were raised:
consistent with previously described predictors of surgical The presence of nonlocalizable seizures might indicate
outcomes. Our findings are in accordance with the liter- a more widespread region of epileptogenicity and might
ature showing that the presence of FBTCS,24 high preop- suggest a faster pattern of propagation. Prior studies sug-
erative seizure frequency,1 and nonlocalizable seizures7 gest an association between rapid spread and worse surgi-
predicts worse outcome. Our finding that better outcomes cal outcome.7,44
FITZGERALD et al.      |  9

4.2  |  Interpretation of model's 4.2.3  |  No interictal epileptiform discharges

interactions
For patients with no IEDs on EEG, a normal MRI imparts
The nomogram also provides an interesting analysis of the a lower probability of seizure freedom than an abnormal
interactions between the location of the IEDs, MRI find- MRI, as would be expected.45,46 However, in this group of
ings, and type of surgery. patients, a temporal lobe surgery imparts a lower prob-
ability of seizure freedom than an extratemporal surgery,
the opposite of what is reported in the literature on surgi-
4.2.1  |  Unilateral interictal cal outcome in general.24,25 We could hypothesize that in
epileptiform discharges extratemporal epilepsies, the absence of IEDs could point
to a more localized epileptogenic zone or more restricted
In patients with unilateral interictal EEG, MRI findings recruitment of the epilepsy network, thereby explaining
did not influence risk prediction. For these same patients, our finding. However, more studies are needed to confirm
resection of the temporal lobe imparts a higher probabil- this hypothesis.
ity of seizure freedom when compared to an extratem- Even though, in theory, we would expect patients with
poral resection. These findings are in accordance with unilateral IEDs to have better outcomes compared to pa-
published studies showing that patients who underwent tients with no IEDs, there are scant data in the literature
temporal lobe resections had either better24 or similar25 to support this hypothesis. A study comparing “oligospik-
outcomes compared to extratemporal cases. ers” vs patients with frequent IEDs found similar surgi-
cal outcomes.47 Another study evaluating patients with
nonlesional intractable TLE with rare or absent IEDs even
4.2.2  |  Bilateral interictal suggested that the rarity of spikes could reflect a less se-
epileptiform discharges vere disease with excellent surgical outcomes.48 In other
words, the real meaning of the absence of IEDs is still
In this patient subgroup, a normal MRI was correlated unclear49 and needs to be further investigated, as empha-
with a higher probability of seizure freedom than an ab- sized by our results.
normal MRI. This unexpected and seemingly contradic-
tory finding could be explained by the fact that patients
with normal MRI and bilateral IED might require a more 4.2.4  |  Interpreting interactions from the
extensive, and therefore more precise, preoperative eval- direction of the surgery site
uation to define the epileptogenic zone. The increased
attention given to these patients may result in a more ac- As discussed previously, extratemporal resections with no
curate demarcation of the epileptogenic zone. In our study IEDs correlated with higher postoperative seizure free-
cohort, 23 of 28 patients (82.1%) with bilateral IEDs and dom compared to extratemporal cases with unilateral or
normal MRI had an invasive EEG evaluation as opposed to bilateral IED. Temporal resections correlate with higher
23 of 85 (27.1%) of those with bilateral IEDs and an abnor- postoperative seizure freedom in the setting of unilateral
mal MRI (p < .0001). The presence of a lesion may provide IEDs but lower seizure freedom for bilateral EEG and no
clinicians with a false sense of security, thereby biasing IEDs. This finding corresponds to the existing literature,
them toward assuming that the epileptogenic zone com- which reports better outcomes in temporal lobe epilepsy
prises solely the lesion when, in reality, the true epilepto- patients with strictly unilateral IED.16
genic zone extends beyond it. This conclusion would lead
clinical teams to suggest more restrictive resections that
do not adequately remove the epileptogenic zone, result- 4.3  |  Nomogram
ing in worse surgical outcomes. The sample size limita-
tion could also be interfering in these unexpected results. In summary, the relationship between scalp EEG findings
To evaluate the model's interactions, we end up creating and the surgical outcome might be more complicated than
subgroups with smaller sample sizes. For example, even initially thought. The current concept of an abnormal
though we had a total of 113 patients with bilateral IEDs, MRI and unilateral IEDs always being associated with
in the end, the subgroup “normal MRI, with extratempo- better surgical outcomes could be an example of how the
ral resection and bilateral IED” included only 9 patients. oversimplification of a complex relationship could con-
Taking into consideration the limitations described here, tribute to the lack of precision seen in the current surgical
this is still an interesting finding that should be further outcome prediction. To compare our findings to the exist-
investigated in future studies. ing literature, we discussed some of the nomogram results
 |
10       FITZGERALD et al.

individually. However, it is important to emphasize that of over-­interpreting the model's interactions. In addition,
the final prediction is based on the combination and inter- because models are calculated based on real data, the pre-
action of all variables and we cannot use these individual diction of unusual combinations of predictive variables
variables as independent predictors. might lack precision. This is a limitation that needs to be
The inclusion of the variable interictal side in the final considered when individually interpreting the data and
model was unexpected and highlights the potential pre- the multiple interactions.
dictive value of IEDs, a variable frequently considered less Another limitation is that we did not quantify the fre-
important in the surgical decision-­making process. One quency of IEDs. We acknowledge that the frequency and
possible explanation would be that interictal discharges percentage of IEDs are largely subjective measurements
outside of the suspected epileptic zone could work as and working with more objective measures will be a fu-
biomarkers of a larger epileptogenic region. Interictal ture line of research.
discharges may also point toward areas that will possibly Despite its limitations, this work attempts to generate
develop as new seizure foci in patients who experience an individualized, evidence-­based surgical outcome pre-
seizure recurrence after surgery. These findings also cor- diction deviating from traditional individual physicians’
relate with the literature on interictal discharges observed experience-­dependent prediction. Further accumulation
in MEG.50 Although IEDs have often been considered a of the data is needed to enhance our ability to predict sur-
secondary factor in surgical planning, our findings argue gical outcome.
for an increased importance to be placed on interictal ac-
tivity in determining the surgical outcome.
In the current study an extended 10–­20 (31-­electrode 4.5  |  Discussion: Summary
scalp montage) was used, which includes an inferior tem-
poral row of scalp electrodes. Although the extended 10–­ Although individual electroencephalographic features
20 montage is superior to the standard 10–­20 (19-­electrode are important in determining the postoperative outcome
scalp montage) for identifying temporal lobe IEDs,14 our after epilepsy surgery, we describe here the role of scalp
study does suggest the potential value of high spatial reso- EEG using a large multicenter cohort. Our findings sup-
lution scalp EEG for extratemporal lobe epilepsy.51 port several widely held beliefs of the predictive value of
The addition of scalp EEG data improved the model’s certain preoperative variables, but we also point toward
performance by 0.06. This increase was not as high as ini- variables whose influence may not be entirely under-
tially expected, highlighting the multifactorial and com- stood. The nomogram created emphasizes the impor-
plex nature of surgical outcome prediction as it cannot be tance of IEDs in the prediction of surgery outcome and
reduced to any single patient characteristic. confirms the predictive value of other clinical variables,
with clinical history of FBTCS having the largest impact in
the model. We highlight the complex interaction between
4.4  |  Study limitations variables when predicting surgical outcome, justifying the
need for advanced statistical techniques to enhance our
Even though pathology and MRI results are frequently ability to predict surgical outcome further.
similar, it is not uncommon to have a normal MRI with
“abnormal” pathology results and vice versa. To build ACKNOWLEDGMENTS
the predictive model we included both pathology (MCD, Study Funding: Lara Jehi, National Institutes of Health
MTS, unknown, tumor, other) and MRI findings (normal (NIH) grant R01 NS097719.
and abnormal). One limitation of this model is that pa-
thology results are usually not available prior to surgery; CONFLICT OF INTEREST
however, we kept this variable given that pathology re- Dr. Worrell has licensed intellectual property to NeuroOne
sults can usually be predicted based on MRI findings and Inc., and licensed intellectual property to Cadence
later confirmed if needed. Neuroscience Inc. Dr. Brinkmann reports that he has li-
For the correct interpretation of our results, the distinc- censed intellectual property to Cadence Neuroscience Inc.
tion between explanatory and predictive statistical models Dr. Cendes reports personal fees from UCB Pharma, grants
is necessary. Explanatory modeling aims to understand re- from Sao Paulo Research Foundation (FAPESP), and grants
lationships, whereas predictive models aim to accurately from CNPq (Conselho Nacional de Desenvolvimento
predict new observations.52 Because of complex relation- Científico e Tecnológico), outside the submitted work; and
ships across variables, in a predictive model, the data are member of the editorial boards of the following journals:
evaluated as a whole and not individually. When we create (1) Neurology (2) Epilepsy Research (3) Epilepsia (Associate
subgroups that contain smaller sample sizes, there is a risk Editor), (4) Frontiers in Neurology (Specialty Chief Editor
FITZGERALD et al.      |  11

11. Schmeiser B, Zentner J, Steinhoff BJ, Brandt A, Schulze-­

in Epilepsy). All other authors have nothing to disclose. Bonhage A, Kogias E, et al. The role of presurgical EEG param-
“We confirm that we have read the Journal's position on eters and of reoperation for seizure outcome in temporal lobe
issues involved in ethical publication and affirm that this epilepsy. Seizure. 2017;51:174–­9.
report is consistent with those guidelines.” 12. Zakaria T, Noe K, So E, Cascino GD, Wetjen N, Van Gompel
JJ, et al. Scalp and intracranial EEG in medically intracta-
ORCID ble extratemporal epilepsy with normal MRI. ISRN Neurol.
2012;2012:1–­9.
Zachary Fitzgerald  https://orcid.
13. Klem GH, Lüders HO, Jasper HH, Elger C. The ten-­twenty elec-
org/0000-0002-6702-1063 trode system of the International Federation. The International
Marcia Morita-­Sherman  https://orcid. Federation of Clinical Neurophysiology. Electroencephalogr
org/0000-0002-8531-3916 Clin Neurophysiol Suppl. 1999;52:3–­6.
Boney Joseph  https://orcid.org/0000-0002-1576-9344 14. Bach Justesen A, Eskelund Johansen AB, Martinussen NI,
Marina K. M. Alvim  https://orcid. Wasserman D, Terney D, Meritam P, et al. Added clinical
org/0000-0002-6780-0761 value of the inferior temporal EEG electrode chain. Clin
Neurophysiol. 2018;129(1):291–­5. https://doi.org/10.1016/j.
Ingmar Blumcke  https://orcid.
clinph.2017.09.113
org/0000-0001-8676-0788 15. Holmes MD, Dodrill CB, Wilensky AJ, Ojemann LM, Ojemann
Benjamin H. Brinkmann  https://orcid. GA, et al. Unilateral focal preponderance of interictal epilepti-
org/0000-0002-2392-8608 form discharges as a predictor of seizure origin. Arch Neurol.
Fernando Cendes  https://orcid. 1996;53:228–­32.
org/0000-0001-9336-9568 16. Chung MY, Walczak TS, Lewis DV, Dawson DV, Radtke R.
Lara Jehi  https://orcid.org/0000-0002-8041-6377 Temporal lobectomy and independent bitemporal interictal
activity: what degree of lateralization is sufficient? Epilepsia.
1991;32(2):195–­201.
REFERENCES 17. Aull-­Watschinger S, Pataraia E, Czech T, Baumgartner
1. Jehi L, Yardi R, Chagin K, Tassi L, Russo GL, Worrell G, et al. C. Outcome predictors for surgical treatment of tempo-
Development and validation of nomograms to provide individ- ral lobe epilepsy with hippocampal sclerosis. Epilepsia.
ualised predictions of seizure outcomes after epilepsy surgery: 2008;49(8):1308–­16.
a retrospective analysis. Lancet Neurol. 2015;14(3):283–­90. 18. Simasathien T, Vadera S, Najm I, Gupta A, Bingaman W, Jehi L.
2. Liu X, Xu J, Mao K, Wang M, Ren P, Lei D, et al. Risk factors Improved outcomes with earlier surgery for intractable frontal
analyses for seizure recurrence in different periods after re- lobe epilepsy. Ann Neurol. 2013;73(5):646–­54.
fractory epilepsy surgery: a prospective single-­center study. 19. Edelvik A, Rydenhag B, Olsson I, Flink R, Kumlien E, Kallen
World Neurosurg. 2018;112:e454–­64. https://doi.org/10.1016/j. K, et al. Long-­term outcomes of epilepsy surgery in Sweden:
wneu.2018.01.060 a national prospective and longitudinal study. Neurology.
3. Gracia CG, Chagin K, Kattan MW, Ji X, Kattan MG, Crotty L, 2013;81(14):1244–­51.
et al. Predicting seizure freedom after epilepsy surgery, a chal- 20. Jutila L, Immonen A, Mervaala E, Partanen J, Partanen K,
lenge in clinical practice. Epilepsy Behav. 2019;95:124–­30. Puranen M, Kälviäinen R, et al. Long term outcome of tempo-
https://doi.org/10.1016/j.yebeh.2019.03.047 ral lobe epilepsy surgery: analyses of 140 consecutive patients. J
4. Chen H, Koubeissi MZ. Electroencephalography in Epilepsy Neurol Neurosurg Psychiatry. 2002;73(5):486–­94.
Evaluation. Contin Lifelong Learn Neurol. 2019;25(2):431–­53. 21. Paglioli E, Palmini A, Paglioli E, da Costa JC, Portuguez M,
5. Mesraoua B, Deleu D, Al Hail H, Melikyan G, Boon P, Martinez JV, et al. Survival analysis of the surgical outcome of
Haider HA, et al. Electroencephalography in epilepsy: look temporal lobe epilepsy due to hippocampal sclerosis. Epilepsia.
for what could be beyond the visual inspection. Neurol Sci. 2004;45(11):1383–­91.
2019;40(11):2287–­91. 22. Jeha LE, Najm IM, Bingaman WE, Khandwala F, Widdess-­
6. Sadler M, Desbiens R. Scalp EEG in temporal lobe epilepsy sur- Walsh P, Morris HH, et al. Predictors of outcome after temporal
gery. Can J Neurol Sci. 2000;27(SUPPL. 1):22–­8. lobectomy for the treatment of intractable epilepsy. Neurology.
7. Tanaka H, Gotman J, Khoo HM, Olivier A, Hall J, Dubeau 2006;66(12):1938–­40.
F, et al. Neurophysiological seizure-­onset predictors of epi- 23. Boesebeck F, Janszky J, Kellinghaus C, May T, Ebner A.
lepsy surgery outcome: a multivariable analysis. J Neurosurg. Presurgical seizure frequency and tumoral etiology predict
2019;133(6):1863–­72. the outcome after extratemporal epilepsy surgery. J Neurol.
8. Bulacio JC, Jehi L, Wong C, Gonzalez-­Martinez J, Kotagal P, 2007;254(8):996–­9.
Nair D, et al. Long-­term seizure outcome after resective surgery 24. Spencer SS, Berg AT, Vickrey BG, Sperling MR, Bazil CW,
in patients evaluated with intracranial electrodes. Epilepsia. Shinnar S, et al. Predicting long-­term seizure outcome after
2012;53(10):1722–­30. resective epilepsy surgery: the multicenter study. Neurology.
9. Liu X, Wu S, Daif A, Sun T, Chauhan V, Issa NP, et al. Clinical 2005;65(6):912–­8.
implications of scalp ictal EEG pattern in patients with tempo- 25. De Tisi J, Bell GS, Peacock JL, McEvoy AW, Harkness WFJ,
ral lobe epilepsy. Clin Neurophysiol. 2019;130(9):1604–­10. Sander JW, et al. The long-­term outcome of adult epilepsy sur-
10. Raghavendra S, Nooraine J, Mirsattari SM. Role of electroen- gery, patterns of seizure remission, and relapse: a cohort study.
cephalography in presurgical evaluation of temporal lobe epi- Lancet. 2011;378(9800):1388–­95. https://doi.org/10.1016/S0140​
lepsy. Epilepsy Res Treat. 2012;2012:1–­18. -­6736(11)60890​-­8
 |
12       FITZGERALD et al.

26. Kumar A, Valentín A, Humayon D, Longbottom AL, Jimenez-­ 42. Blumenfeld H, Varghese GI, Purcaro MJ, Motelow JE, Enev
Jimenez D, Mullatti N, et al. Preoperative estimation of seizure M, McNally KA, et al. Cortical and subcortical networks in
control after resective surgery for the treatment of epilepsy. human secondarily generalized tonicclonic seizures. Brain.
Seizure. 2013;22(10):818–­26. 2009;132(4):999–­1012.
27. Englot DJ, Wang DD, Rolston JD, Shih TT, Chang EF. Rates and 43. Caciagli L, Allen LA, He X, Trimmel K, Vos SB, Centeno M,
predictors of long-­term seizure freedom after frontal lobe epi- et al. Thalamus and focal to bilateral seizures: a multiscale cog-
lepsy surgery: a systematic review and meta-­analysis. Clinical nitive imaging study. Neurology. 2020;95(17):e2427–­41.
article. J Neurosurg. 2012;116(5):1042–­8. 44. Holtkamp M, Sharan A, Sperling MR. Intracranial EEG in pre-
28. Englot DJ, Chang EF. Rates and predictors of seizure free- dicting surgical outcome in frontal lobe epilepsy. Epilepsia.
dom in resective epilepsy surgery: an update. Neurosurg Rev. 2012;53(10):1739–­45.
2014;37(3):389–­405. 45. Chapman K, Wyllie E, Najm I, Ruggieri P, Bingaman W, Lüders
29. Bell GS, De Tisi J, Gonzalez-­Fraile JC, Peacock JL, McEvoy AW, J, et al. Seizure outcome after epilepsy surgery in patients with
Harkness WFJ, et al. Factors affecting seizure outcome after normal preoperative MRI. J Neurol Neurosurg Psychiatry.
epilepsy surgery: an observational series. J Neurol Neurosurg 2005;76(5):710–­3.
Psychiatry. 2017;88(11):933–­40. 46. Fong JS, Jehi L, Najm I, Prayson RA, Busch R, Bingaman
30. McIntosh AM, Kalnins RM, Mitchell LA, et al. Temporal lobec- W. Seizure outcome and its predictors after temporal lobe
tomy: Long-­term seizure outcome, late recurrence and risks for epilepsy surgery in patients with normal MRI. Epilepsia.
seizure recurrence. Brain. 2004;127(9):2018–­30. 2011;52(8):1393–­401.
31. Schulz R, Lüders HO, Hoppe M, Tuxhorn I, May T, Ebner A. 47. Di Gennaro G, Quarato PP, Sebastiano F, Esposito V, Onorati
Interictal EEG and ictal scalp EEG propagation are highly pre- P, Mascia A, et al. Postoperative EEG and seizure out-
dictive of surgical outcome in mesial temporal lobe epilepsy. come in temporal lobe epilepsy surgery. Clin Neurophysiol.
Epilepsia. 2000;41(5):564–­70. 2004;115(5):1212–­9.
32. Jeha LE, Najm I, Bingaman W, Dinner D, Widdess-­Walsh P,
48. Basiri R, Shariatzadeh A, Wiebe S, Aghakhani Y. Focal epilepsy
Luders H, et al. Surgical outcome and prognostic factors of
without interictal spikes on scalp EEG: a common finding of
frontal lobe epilepsy surgery. Brain. 2007;130(2):574–­84.
uncertain significance. Epilepsy Res. 2019;150:1–­6.
33. Radhakrishnan K, So EL, Silbert PL, Jack CR, Cascino GD,
49. Rosati A, Aghakhani Y, Bernasconi A, Olivier A, Andermann
Sharbrough FW, et al. Predictors of outcome of anterior tem-
F, Gotman J, et al. Intractable temporal lobe epilepsy with
poral lobectomy for intractable epilepsy: a multivariate study.
rare spikes is less severe than with frequent spikes. Neurology.
Neurology. 1998;51(2):465–­71.
2003;60(8):1290–­5.
34. Harrell FE, Lee KL, Mark DB. Multivariable prognostic mod-
50. Murakami H, Wang ZI, Marashly A, Krishnan B, Prayson RA,
els: Issues in developing models, evaluating assumptions
Kakisaka Y, et al. Correlating magnetoencephalography to
and adequacy, and measuring and reducing errors. Stat Med.
stereo-­electroencephalography in patients undergoing epilepsy
1996;15(4):361–­87.
surgery. Brain. 2016;139(11):2935–­47.
35. Steyerberg EW, Harrell FE, Borsboom GJJM, Eijkemans MJC,
51. Feyissa AM, Britton JW, Van Gompel J, Lagerlund TL, So
Vergouwe Y, Habbema JDF. Internal validation of predictive
E, Wong-­Kisiel LC, et al. High density scalp EEG in fron-
models: efficiency of some procedures for logistic regression
tal lobe epilepsy. Epilepsy Res. 2017;129:157–­61. https://doi.
analysis. J Clin Epidemiol. 2001;54(8):774–­81.
org/10.1016/j.eplep​syres.2016.12.016
36. Steyerberg EW, Harrell FE. Prediction models need appro-
52. Doherty C, Nowacki AS, McAndrews MP, McDonald CR,
priate internal, internal–­external, and external validation. J
Clin Epidemiol. 2016;69:245–­7. https://doi.org/10.1016/j.jclin​ Reyes A, Kim MS, et al. Response: predicting mood decline
epi.2015.04.005 following temporal lobe epilepsy surgery in adults. Epilepsia.
37. Molinaro AM, Simon R, Pfeiffer RM. Prediction error estima- 2021;62(2):450–­9. https://doi.org/10.1111/epi.16800
tion: a comparison of resampling methods. Bioinformatics.
2005;21(15):3301–­7.
38. Ji X, Kattan MW. Tutorial: development of an online risk calcu- How to cite this article: Fitzgerald Z, Morita-­
lator platform. Ann Transl Med. 2018;6(3):46. Sherman M, Hogue O, Joseph B, Alvim MKM,
39. Cortese G. How to use statistical models and methods for clini- Yasuda CL, et al. Improving the prediction of
cal prediction. Ann Transl Med. 2020;8(4):76. epilepsy surgery outcomes using basic scalp EEG
40. Jehi L. Algorithms in clinical epilepsy practice: can they really findings. Epilepsia. 2021;00:1–­12. https://doi.
help us predict epilepsy outcomes? Epilepsia. 2021;62(S2):1–­7.
org/10.1111/epi.17024
41. Josephson CB, Wiebe S. Precision medicine: academic dream-
ing or clinical reality? Epilepsia. 2021;62(S2):1–­12