Paroxysmal fast activity is a biomarker of treatment response in deep brain stimulation for Lennox–Gastaut syndrome

Abstract

Objective

Epilepsy treatment trials typically rely on seizure diaries to determine seizure frequency, but these are time‐consuming and difficult to maintain accurately. Fast, reliable, and objective biomarkers of treatment response are needed, particularly in Lennox–Gastaut syndrome (LGS), where high seizure frequency and comorbid cognitive and behavioral issues are additional obstacles to accurate diary‐keeping. Here, we measured generalized paroxysmal fast activity (GPFA), a key interictal electrographic feature of LGS, and correlated GPFA burden with seizure diaries during a thalamic deep brain stimulation (DBS) treatment trial (Electrical Stimulation of the Thalamus in Epilepsy of Lennox–Gastaut Phenotype [ESTEL]).

Methods

GPFA and electrographic seizure counts from intermittent, 24‐h electroencephalograms (EEGs) were compared to 3‐month diary‐recorded seizure counts in 17 young adults with LGS (mean age ± SD = 24.9 ± 6.6) in the ESTEL study, a randomized clinical trial of DBS lasting 12 months (comprising a 3‐month baseline and 9 months of postimplantation follow‐up).

Results

Baseline median seizures measured by diaries numbered 2.6 (interquartile range [IQR] = 1.4–5) per day, compared to 284 (IQR = 120.5–360) electrographic seizures per day, confirming that diaries capture only a small fraction of seizure burden. Across all patient EEGs, the average number of GPFA discharges per hour of sleep was 138 (IQR =72–258). GPFA duration and frequency, quantified over 2‐h windows of sleep EEG, were significantly associated with diary‐recorded seizure counts over 3‐month intervals (p < .001, η² _p  = .30–.48). For every GPFA discharge, there were 20–25 diary seizures witnessed over 3 months. There was high between‐patient variability in the ratio between diary seizure burden and GPFA burden; however, within individual patients, the ratio was similar over time, such that the percentage change from pre‐DBS baseline in seizure diaries strongly correlated with the percentage change in GPFA.

Significance

When seeking to optimize treatment in patients with LGS, monitoring changes in GPFA may allow rapid titration of treatment parameters, rather than waiting for feedback from seizure diaries.

Key Points.

• Accurate recording of seizure diaries in patients with Lennox–Gastaut syndrome can be difficult

• GPFA is a key electrographic feature of Lennox–Gastaut syndrome and is easily identifiable on sleep EEG

• Changes in GPFA burden were strongly associated with changes in diary‐recorded seizures during a deep brain stimulation trial in patients with Lennox–Gastaut syndrome

• GPFA may be a usable as a biomarker of treatment response in patients with Lennox–Gastaut syndrome

1. INTRODUCTION

Most antiseizure medication and device trials continue to rely on seizure frequency as the primary outcome variable, recorded from patient/carer seizure diaries. However, diaries are well known to underestimate seizure frequency, ¹ contaminating estimates of treatment effect and contributing to the requirement for large sample sizes and lengthy trial durations.

Lennox–Gastaut syndrome (LGS) is a severe, childhood onset, treatment‐resistant epilepsy phenotype. In this condition, measurement of treatment effect is confounded by the inherent inaccuracies of seizure diaries, amplified in the presence of intellectual disability, challenging behaviors, ² and seizures that are variably detectable according to circumstance (e.g., whether the patient is standing, seated, or lying at seizure onset). Nocturnal, subtle, and clustered seizures can be easily missed; there are often rotating carers with varied levels of experience; and management of behaviors/autistic traits can often deflect the focus from seizure documentation. These limitations highlight the need to identify objective and reliable biomarkers of treatment efficacy, as an adjunct or alternative to seizure diaries.

A "biomarker" is an objectively measurable characteristic of a normal or pathological biological process, in this case, the propensity for brain tissue to generate a seizure. ³ Potential biomarkers for epilepsy include imaging and electrophysiologic measurements, changes in gene expression, and metabolites in blood or tissues. ³ However, many of these markers have limited applicability in LGS, due to the wide range of etiological types that can cause epilepsy in this population (e.g., structural, genetic, metabolic, and still unknown causes).

LGS can be conceptualized as a "secondary network epilepsy," ⁴ in which the shared electroclinical manifestations reflect epileptic activity expressed through a common epileptic network, rather than the specific lesional, genetic, or other cause. ⁴ Hence, a biomarker capable of tracking the behavior of this shared brain network may have broad utility across patients with LGS, independent of the specific etiology.

Generalized paroxysmal fast activity (GPFA), historically described by various names including paroxysmal fast rhythms, ⁵ grand mal discharge,⁶ and repetitive fast discharges, ⁷ is a characteristic interictal epileptiform discharge seen on the electroencephalogram (EEG) in LGS, and occasionally that of patients with other developmental and epileptic encephalopathies or generalized epilepsies. ⁸ It is characterized by 10–30‐Hz bursts of generalized, often frontally predominant activity, and occurs most frequently during non‐rapid eye movement sleep. ⁹ , ¹⁰ The presence of GPFA has been associated with seizure risk ¹¹ and drug resistance. ¹² GPFA is morphologically similar to the low‐voltage fast activity seen at the onset of tonic seizures (Figure 1), suggesting GPFA and tonic seizures are expressed via a similar brain network. ⁴ We previously used combined EEG with functional magnetic resonance imaging to study this network, showing prominent activation of frontoparietal cortex, consistent with the diffuse electrical field of GPFA. ¹³

FIGURE 1.

Electrographic similarities between generalized paroxysmal fast activity (GPFA) and tonic seizures. The panel on the left shows an example of GPFA, with fast (typically 12–30 Hz) repetitive discharges in a generalized distribution on scalp electroencephalogram (bipolar montage, 10 μV/mm), taken from a patient with Lennox–Gastaut syndrome. The panel on the right shows the beginning of an electrographic tonic seizure (bipolar montage, 10 μV/mm), with electrodecremental response, low‐voltage fast activity (typically 15–30 Hz, similar to that seen during GPFA), followed by a sustained burst of higher amplitude fast rhythms with admixed theta/delta activity. The similarity between interictal GPFA and low‐voltage fast activity of tonic seizures is thought to reflect a shared network abnormality ⁴ .

We recently reported efficacy and safety outcomes of the Electrical Stimulation of the Thalamus in Epilepsy of Lennox–Gastaut Phenotype (ESTEL) study, ¹⁴ a randomized controlled trial of deep brain stimulation (DBS) to bilateral centromedian thalamic nucleus in young adults with LGS. To ensure DBS‐related changes in seizure frequency were captured as reliably as possible, ¹⁵ ESTEL included 12 months of carer‐reported seizure diaries as well as four 24‐h EEGs, performed once every 3 months. This dataset provided the unique opportunity to correlate changes in electrographic epileptic activity with seizure diaries, preceding and following DBS implantation and treatment. Similar associations have been explored in other epilepsy treatment trials. ¹⁶

Here, we aimed to determine the associations between interictal GPFA, electrographic seizures, and diary‐recorded seizures in LGS. Specifically, we hypothesized that changes in the burden of GPFA, captured during discrete windows of EEG, would correlate with changes in diary‐recorded seizures measured over longer (3 month) periods.

2. MATERIALS AND METHODS

2.1. Participants

Twenty young adults with LGS underwent DBS of the centromedian nucleus of the thalamus as part of the ESTEL trial (Australian New Zealand Clinical Trials Registry number 12621001233819), completed at Austin Health in Melbourne, Australia. ¹⁴ , ¹⁷ , ¹⁸ One participant was excluded due to a cerebral infection, with detailed outcomes of the remaining 19 participants described previously. ¹⁴ Study exit EEGs were not obtained for two ESTEL participants due to COVID‐19. Thus, for the present study, seizure diaries and 24‐h EEGs from 17 of 19 participants (mean age ± SD = 24.9 ± 6.61 years, 13 females) were analyzed.

Table 1 details demographic, clinical, and electrographic characteristics of the 17 participants. Parents or a responsible guardian provided written informed consent according to the Declaration of Helsinki before any study‐specific procedures commenced. The trial protocol received institutional approval from Austin Health Human Research Ethics Committee prior to trial commencement (approval number HREC/16/Austin/139). We conducted this study following the Standards for Reporting of Diagnostic Accuracy guidelines. ¹⁹

TABLE 1.

Baseline clinical and electrographic features of 17 ESTEL participants analyzed

Participant number/sex	Age at implantation, years	Duration of epilepsy, years	Etiology	Baseline characteristics
				Seizures/day, diary	EEG‐clinical seizures/day	EEG‐seizures/day	GPFA discharges/min, EEG
1/F	24	23.5	Unknown	7.41	31.83	335.18	11.86
2/F	24	12	Unknown	.46	4.71	9.41	3.03
3/F	18	16	Unknown	.75	16.91	51.66	7.32
4/M	20	13	Unknown	1.97	120.00	532.71	18.88
5/F	30	25	Structural	2.32	45.72	426.76	4.9
6/F	28	14	Genetic	1.40	61.64	283.56	2.12
7/F	37	34.5	Genetic/structural	1.26	23.09	73.46	4.73
8/M	37	34	Structural	5.88	32.21	295.30	5
9/M	24	20	Unknown	4.99	18.71	416.21	4.52
10/F	35	34	Structural	1.90	33.47	281.12	1.72
11/F	28	27.7	Unknown	3.05	13.85	145.85	8.91
12/F	17	12	Genetic/structural	16.15	37.45	359.95	1.26
13/F	23	22.25	Genetic	3.98	113.97	851.92	5.03
14/F	20	15	Structural	.80	241.73	323.26	5.45
15/M	20	19.5	Unknown	3.80	27.81	120.53	2.1
16/M	18	16.5	Genetic	2.64	19.28	115.69	.75
17/F	20	19.5	Genetic/structural	62.18	96.77	203.23	2.17

Abbreviations: EEG, electroencephalographic; ESTEL, Electrical Stimulation of the Thalamus in Epilepsy of Lennox–Gastaut Phenotype; F, female; GPFA, generalized paroxysmal fast activity; M, male.

2.2. Trial design and study variables

Detailed information about the ESTEL trial design is provided in our previous work, ¹⁴ with a schematic of its study design in Figure 2. In brief, participants' carers completed standardized, monthly (28 day) seizure diaries in which they were asked to record “countable seizures” each day. Countable seizures included convulsions, tonic, atonic, myoclonic, and focal seizures. Atypical absence seizures were not included due to difficulty in reliably distinguishing these from background behavior, particularly in the presence of cognitive impairment.

FIGURE 2.

Study design. Study timeline depicts each 3‐month phase of the trial and key timepoints relative to centromedian thalamic nucleus deep brain stimulation (DBS) implantation (Week 0). Randomization occurred at Week 12. The blue line depicts the time period in which participants received stimulation. The head symbols depict the time within the study that a 24‐h electroencephalogram (EEG) was performed in Electrical Stimulation of the Thalamus in Epilepsy of Lennox–Gastaut Phenotype (ESTEL), with four in total. The current study compares seizure counts from ESTEL seizure diaries with EEG abnormalities on EEGs from each study phase (i.e., preimplantation, postimplantation [without stimulation], blinded, and unblinded phases). The main comparison is between diary‐recorded seizure frequency (over 3 months) and generalized paroxysmal fast activity (GPFA) burden (over 2 h of sleep EEG) in each study phase.

Average number of seizures per day was calculated over each of four 3‐month study phases: (1) "baseline" = 3‐month baseline observation phase prior to DBS implantation, (2) "prestimulation" = 3‐month recovery period following implantation without any stimulation delivered, (3) "blinded" = 3‐month period where half of participants were randomized to have their device turned on (active arm stimulation, early treatment group) or remain off (control arm, delayed stimulation group), and (4) "unblinded" = final 3 months of study where all participants received active stimulation. Nine of 17 (53%) participants were in the early treatment group (i.e., these participants had received 3 months of stimulation by end of blinded phase/Week 24). At study exit/Week 36, the remaining eight of 17 (47%) participants in the control group had received 3 months of treatment, whereas those in the early treatment group had received 6 months of treatment.

Four 24‐h ambulatory EEGs (17 participants × 4 = total 68 EEGs) were acquired, once per end of each study phase: baseline, Week 12, Week 24, and Week 36. Each EEG was analyzed twice, blinded to study phase and treatment group by author L.J.D., a board‐certified clinical epileptologist with advanced EEG training. To test interrater reliability (see methods below), a second assessor with EEG fellowship training (author C.S.) reviewed a random selection of 50% of all baseline EEG recordings. EEGs were acquired using the international 10–20 system with at least 24 electrode channels, acquired with a sampling rate of 256 Hz. EEG was reviewed using ProFusion 5 software (Compumedics) by visual analysis, using a low‐frequency filter of .5 Hz, high‐frequency filter of 70 Hz, and 50‐Hz notch filter.

Burden of interictal GPFA was assessed over a continuous 2‐h time window of sleep EEG, typically between 12:00 and 2:00 a.m. This period was chosen due to the higher likelihood of interictal abnormalities occurring in sleep in patients with LGS. ²⁰ Where participants had >10 arousals or were awake >5 min during this period, the 2‐h window of analysis commenced 30 min later. GPFA was defined as paroxysmal EEG events of (1) 10–30‐Hz frequency, (2) >250‐ms duration, and (3) voltage amplitudes greater than background. ⁹ GPFA events were differentiated from normal sleep spindles, faster normal sleep K‐complexes, and muscle/movement/electrode artifact by visual inspection aided by electrical field and morphology. The onset and duration of GPFA was marked on bipolar or transverse montages. Where there was <1 s between sequential bursts of GPFA, bursts were marked as a single continuous event (Supplementary Figure S1). For each 2‐h window, we recorded the total number of GPFA events and their total cumulative duration (in seconds).

We additionally measured the number of electrographic seizures over the entire 24‐h EEG recording (“EEG‐seizures”). Electrographic seizures for each participant were defined as those similar to their seizures documented on prior video‐EEG monitoring (i.e., “electroclinical seizures”), ²¹ or by the following criteria: (1) a sustained run of generalized, fast, or epileptiform activity with evolution, causing a change in the background rhythm; and (2) a duration of ≥5 s (Supplementary Figure S1). As such, counts of EEG‐seizures incorporate a broader range of epileptic discharges, including trains of generalized fast activity, focal electrographic seizures, and evolving trains of rhythmic sharp slow waves, provided these were captured with clinical correlate prior to study enrollment. We did not include runs of slow spike‐waves, as atypical absences were not included on seizure diary analysis. We note that our chosen seizure duration differs from recent definitions of electrographic seizures, which recommend a cutoff of >10 s ²¹ ; however, in our experience, clinically subtle tonic seizures in LGS commonly occur with shorter (5–10 s) bursts of fast activity. ² Within this group of EEG‐seizures, we further defined “EEG‐clinical seizures,” the subgroup of electrographic events that had a clear clinical correlate on prior video‐EEG or with evidence of muscle involvement defined by electromyographic artifact.

2.3. Statistical analyses

Data processing and statistical analyses were performed using R Studio (v2021.09.1+372). Exploratory data visualization was undertaken to determine data distributions. Data are summarized as median values with interquartile ranges (IQRs). In all tests, the threshold for significance was set at p < .05.

2.4. Interrater agreement of EEG markups

To assess reliability of EEG markups, we tested interrater agreement (IRA) between two manual markups of all 17 patients' baseline EEG recordings. The markup of Assessor 1 (L.J.D.) of 2 h of sleep EEG was designated the gold standard. Assessor 2 (C.S.) marked GPFA events on clean versions of all 17 baseline recordings (i.e., without annotations from L.J.D. or knowledge of participant identity). Blinding procedures were described previously. ¹⁴ The two assessors' markups were compared by exporting timestamps for all discharge onsets and durations, and then calculating (1) the proportion of matching GPFA discharge onsets, across participant EEGs (where a “match” was defined as a discharge onset marked by Assessor 2 that was within 200 ms of an onset marked by Assessor 1); and (2) agreement between the total durations of GPFA (in seconds) marked, across participant EEGs. IRA was calculated using the intraclass correlation coefficient (ICC) and related 95% confidence intervals (CIs). IRA scores ranged from 0–1, with 0 = poor, 0–.20 = slight, .21–.40 = fair, .41–.60 = moderate, .61–.80 = substantial, and .81–1.0 = almost perfect. ²²

2.5. Relationship between EEG features and seizure diaries

We performed linear mixed effects (LME; from the lme4 package ²³ ) analyses to test for associations between diary‐recorded seizure frequency and each of the three EEG features (total GPFA duration, number of GPFA events, and number of electrographic seizures), across the four study timepoints (i.e., baseline, Week 12, Week 24, Week 36).

All analyses included a random variable for patient, with diary‐recorded seizure frequency as the dependent variable, EEG feature as a fixed effect, and by‐patient random slopes for the effect of EEG feature (Supplementary Table S1). In four analyses (Models I, III, V, and VII in Supplementary Table S1), we additionally specified study timepoint as a random effect and included by‐timepoint random slopes for the effect of EEG feature. In the remaining four analyses (Models II, IV, VI and VIII), we instead specified treatment group (i.e., either receiving stimulation or not receiving stimulation) as an additional random effect and included by‐treatment random slopes for the effect of EEG feature. Significance was calculated using the lmerTest package, ²⁴ which applies Satterthwaite's method to estimate degrees of freedom and generate p‐values for mixed models. Prior to analysis, diary‐recorded seizure frequency and all EEG features were square‐root transformed to normalize their distributions. Effect size was assessed using partial eta‐squared coefficients (η² _p ), where values between .01 and .06 indicate little effect, values .07–.14 indicate moderate effect, and values >.14 indicate great effect of EEG features. ²⁵ Pearson correlation assessed the relationship between total GPFA duration and the number of GPFA discharges per EEG study across all timepoints.

To explore changes before/following DBS, we expressed the following outcome variables (measured postimplantation at study Weeks 12, 24, and 36) as percentage change from their baseline (i.e., preimplantation) values: GPFA number/duration, electrographic seizures, and diary‐recorded seizures. Pearson correlations were used to explore associations between percentage change in diary‐recorded seizures and GPFA burden or electrographic seizures.

3. RESULTS

The mean ± SD EEG recording duration was 22.5 ± 1.8 h. Across all participants' 24‐h EEG recordings (all study timepoints combined), there were a total of 16 588 electrographic seizures (i.e., EEG‐seizures), of which 2808 had a clinical correlate (i.e., EEG‐clinical seizures). Across all participants' 2‐h sleep windows, there were 26 351 GPFA discharges with a cumulative duration of 536.5 min (i.e., 6.6% of the record analyzed; Table 2). The median number of EEG‐seizures per 24‐h recording was 276 (IQR = 91–346), with a median duration of 26.7 s (IQR = 17.4–49.3). The median number of EEG‐clinical seizures per 24‐h recording was 32 (IQR = 20–60), with a median duration of 36.7 s (IQR = 13.5–83). The median number of GPFA events per minute was 2.3 (IQR = 1.2–4.3), with a median per‐discharge duration of 1.2 s (IQR = .9–1.6; Table 2). GPFA often occurred in association with subsequent epileptiform spike and slow wave discharges.

TABLE 2.

Characteristics of identified EEG abnormalities within and across all study timepoints

EEG abnormality	EEG timepoint				Combined timepoints, n = 68
	Baseline, n = 17	Week 12, n = 17 a	Week 24, n = 17 b	Week 36, n = 17 c
GPFA, median (IQR)
Events per minute, n	2.6 (1.7–5)	2.9 (.8–3.8)	2.3 (1.4–4.7)	2.1 (1.1–3.5)	2.3 (1.2–4.3)
Event duration, s	1.2 (.9–1.8)	1.4 (1–1.6)	1.1 (.8–1.6)	1.2 (.8–1.4)	1.2 (.9–1.6)
Burden, s/2 h	471 (268–724)	415 (140–591)	390 (222–539)	288 (156–459)	370 (173–560)
EEG‐clinical seizures, median (IQR)
n/h	1.3 (.8–2.6)	1.4 (.5–2.2)	1.0 (.3–2.0)	.5 (.3–1.3)	1.1 (.4–2.2)
Duration, s	28.3 (14.7–69.1)	27.2 (16.5–47.4)	33.2 (16.7–73.9)	26.9 (13.9–57.5)	36.7 (13.5–83)
EEG‐seizures, median (IQR)
n/h	11.8 (5–15)	12.1 (5.2–15.1)	8 (5.2–14)	7.4 (3.6–11.6)	9.4 (4.6–14)
Duration, s	26.1 (17.4–49.3)	22.9 (14.5–43)	29.2 (17.6–53.6)	28 (14.9–39.6)	26.7 (17.4–49.3)

Abbreviations: EEG, electroencephalographic; GPFA, generalized paroxysmal fast activity; IQR, interquartile range.

^a EEG performed 3 months after deep brain stimulation surgery and before treatment/stimulation was delivered.

^b EEG performed at end of blinded phase. Data shown includes nine participants who had received 3 months of treatment/stimulation.

^c EEG performed at study exit, where nine participants had received 6 months of treatment/stimulation and eight participants had received 3 months of treatment/stimulation.

3.1. Interrater agreement

Among the 17 baseline 24‐h EEG recordings, the ICC for the number of GPFA events was .96 (95% CI = .89–.99), suggesting "almost perfect" agreement. A similar level of agreement was observed for the total duration of GPFA events (ICC = .98, 95% CI = .94–.99).

3.2. Relationship between EEG features and seizure diaries

3.2.1. Generalized paroxysmal fast activity

There was high variability in GPFA burden (number of seconds of GPFA) and diary‐recorded seizure counts between participants, as well as high variability in the ratio between GPFA discharges and diary seizures from one participant to the next. This was reflected by a weak and nonsignificant Pearson correlation between GPFA discharges and diary seizures when combining all patients and study timepoints together (Figure 3A).

FIGURE 3.

Correlation between electroencephalographic (EEG) measures and diary‐recorded seizure frequency in 17 Electrical Stimulation of the Thalamus in Epilepsy of Lennox–Gastaut Phenotype (ESTEL) participants. In each plot, either generalized paroxysmal fast activity (GPFA) burden (total seconds over a 2‐h window of sleep) or electrographic seizures (total number over 24 h of ambulatory EEG) is plotted on the y‐axis, and diary‐recorded seizure frequency is plotted on the x‐axis, in 17 participants. Each participant had four EEGs performed throughout the ESTEL study, resulting in 68 data points. (A) Association between GPFA burden and daily seizure frequency, ignoring within‐patient structure, shows poor correlation (Pearson r = .15, p = .24). (B) In contrast, using linear mixed effects (LME) modeling with random slopes specified for patients, there is a strong correlation between GPFA burden and diary‐recorded seizure frequency (p < .001). Here, values are grouped by participant (i.e., four data points per participant), with each assigned a different color and slope of best fit. The legend at the bottom of the middle panel assigns each participant a color (i.e., P1 = Participant 1, P2 = Participant 2, etc.), with the participant numbers corresponding to those shown in Table 1. The plot highlights the variability in measured variables, with some participants having a high number of diary‐recorded seizures but lower GPFA burden (e.g., the participant shown in dark blue, lower right), and others having high GPFA burden but fewer seizures (e.g., the participant shown in light blue, upper left). (C) The same datapoints as those shown in A/B are depicted, but now separated by timepoint when the EEG was performed. Each ellipse captures 95% of observations for that timepoint. Baseline observations in red show wide variability (outermost ellipse), but by the end of the ESTEL trial, when all participants had received stimulation (innermost ellipse; purple), there appears to be reduction in both GPFA and diary‐recorded seizures, suggesting treatment effect. (D) Comparison of EEG‐seizures with diary‐recorded seizures, showing poor overall correlation (Pearson r = .18, p = .13). (E) Use of LME modeling showed that diary‐recorded seizures were weakly but nonsignificantly associated with EEG‐seizure counts (p = .12), with more variable intraparticipant relationships than seen for GPFA burden in B. (F) Data grouped by study timepoint, as per C, for EEG‐seizure versus diary‐recorded seizures. PFA, paroxysmal fast activity.

However, LME models, which accommodate a different relationship (“slope of the line”) between GPFA and diary seizures for each patient (Figure 3B), revealed a significant association. Across the four study timepoints, 3‐month average diary seizures/day were significantly associated with both total GPFA duration (p < .001, η² _p  = .41–.48) and total number of GFPA events (p < .001, η² _p  = .30–.39) measured over the 2‐h window of sleep EEG (Table 3). The association between diary‐recorded seizures and these EEG features was similar across different model specifications, when either study timepoint or treatment was included as a random effect (Table 3). As expected, a strongly positive correlation was seen between the two measures of GPFA burden (i.e., total duration and event count; Pearson r = .88, p < 2.2 × 10⁻¹⁶, 95% CI = .81–.92; Supplementary Figure S2).

TABLE 3.

Results of linear mixed effects models: Relationship of EEG abnormalities to diary‐recorded seizures

Variable	Estimate	SE	df	t	F	p	η p 2 (90% CI)
Model I: Total GPFA duration as fixed effect and timepoints a plus participant as random effect
Intercept	.86	.22	28.08	3.95	–	<.001***	–
Total GPFA duration	.03	.005	55.27	6.20	38.39	<.001***	.41 (.25–.54)
Model II: Total GPFA duration as fixed effect and treatment b plus participant as random effect
Intercept	.76	.22	25.45	3.42	–	<.001***	–
Total GPFA duration	.04	.005	54.60	7.12	50.64	<.001***	.48 (.32–.60)
Model III: Number of GPFA discharges as fixed effect and timepoints plus participant as random effect
Intercept	.73	.25	36.82	2.92	–	.006**	–
GPFA discharges, n	.04	.009	60.43	5.12	26.96	<.001***	.30 (.15–.44)
Model IV: Number of GPFA discharges as fixed effect and treatment plus participant as random effect
Intercept	.59	.26	28.98	2.32	–	.03*	–
GPFA discharges, n	.05	.008	58.77	6.15	37.85	<.001***	.39 (.23–.52)
Model V: EEG‐clinical seizures as fixed effect and timepoint plus participant as random effect
Intercept	1.46	.23	29.13	6.37	–	<.001***	–
EEG‐clinical seizures c	.01	.02	61.65	.65	.42	.52	.006 (.00–.08)
Model VI: EEG‐clinical seizures as fixed effect and treatment plus participant as random effect
Intercept	1.35	.24	17.44	5.55	–	<.001***	–
EEG‐clinical seizures	.03	.02	58.21	1.18	1.40	.24	.02 (.00–.12)
Model VII: All EEG seizures as fixed effect and timepoint plus participant as random effect
Intercept	1.17	.29	42.09	3.99	–	<.001***	–
EEG‐seizures c	.02	.02	64.01	1.58	2.51	.12	.04 (.00–.14)
Model VIII: All EEG seizures as fixed effect and treatment plus participant as random effect
Intercept	1.07	.32	27.15	3.31	–	.003**	–
EEG‐seizures	.03	.02	47.55	1.67	2.79	.10	.06 (.00–.19)

Abbreviations: CI, confidence interval; EEG, electroencephalographic; GPFA, generalized paroxysmal fast activity.

^a Timepoint refers to timing of EEG (i.e., baseline, Week 12, Week 24, and Week 36).

^b Treatment refers to whether the participant was receiving stimulation at the timepoint the EEG was recorded.

^c Refers to electrographic seizures.

^* p < .05,

^** p < .01,

^*** p < .001.

Model estimates are shown in Table 3. Although there was high variability between patients, the models estimated that, on average, for every one GPFA event on 2 h of sleep EEG, seizure diaries captured 20–25 clinical seizures over 3 months (estimate ± SE: .04 ± .009, t = 5.12 [Model III] and estimate ± SE: .05 ± .008, t = 6.15 [Model IV]; Table 3). Another way of viewing this result is that in a patient with LGS, for every second of GPFA seen on a 2‐h sleep EEG, there is likely approximately 2–2.8 witnessed seizures over 1 week. Conversely, for every one diary‐recorded seizure in 3 months, there is likely 30–40 ms of GPFA on 2 h of sleep EEG (estimate ± SE: .03 ± .005, t = 6.20 [Model I]; estimate ± SE: .04 ± .005, t = 7.12 [Model II]).

3.2.2. Electrographic seizures

We did not find a significant association between electrographic seizures recorded over 24 h (EEG‐seizures or EEG‐clinical seizures) and diary‐recorded seizures (Figure 3D–F and Table 3). Even when accounting for a patient‐specific ratio between EEG‐seizures and diary seizures with LME, no significant association was found. An exploratory LME analysis was undertaken post hoc to investigate the relationship between diary‐recorded seizures and electrographic seizures including only those with a longer duration (>10 s) than originally defined (≥5 s). In contrast to the result with electrographic seizures ≥5 s, the analysis using only >10 s electrographic seizures revealed a significant association between EEG‐seizures and diary‐recorded seizures, with two different models using treatment (F _{1, 53.39} = 5.48, p = .003) or timepoint (F _{1, 65.84} = 4.67, p = .003) as random effect. We outline possible explanations for this apparent discrepancy in the Discussion section below.

3.2.3. Treatment effects

Changes in GPFA and diary‐recorded seizures after DBS treatment, across the four study timepoints, are shown in Figure 3C. At baseline, spread of data was highest, with successive reductions at each timepoint leading to the smallest spread at study exit when all participants had received 3–6 months of DBS treatment. Comparison of data spread between electrographic and diary‐recorded seizures demonstrated less variability between Week 12 and 24 data, but overall, both electrographic and diary‐recorded seizures showed a visual tendency to reduce from baseline to study exit (Figure 3F). Note that we did not seek to determine the statistical significance of these changes here, efficacy being the focus of our previous work. ¹⁴

To account for the high variability in the ratio of EEG events to diary seizures from one patient to the next, we expressed Week 12, 24, and 36 data as percentage changes from baseline values. This revealed a strongly positive correlation between GPFA duration and diary‐recorded seizures (Pearson r = .72, p = 5 × 10⁻¹²; 95% CI = .16–.57; Figure 4A). Viewed another way, this suggested that although the ratio between GPFA duration and diary seizures varied from one patient to the next, every 10% reduction in GPFA burden was associated with a similar percentage reduction in diary seizures. Regarding the percentage change in electrographic seizures, we found a similar relationship with percentage change in diary seizures for EEG‐clinical seizures (Pearson r = .38, p = .001; 95% CI = .16–.57; Figure 4B) but not for EEG‐seizures, where a weak and nonsignificant association was observed (Pearson r = .07, p = .58; 95% CI = −.17 to .3; Figure 4C).

FIGURE 4.

Percentage changes in generalized paroxysmal fast activity (GPFA) burden track changes in seizure frequency. (A) There was a strong correlation between percentage changes in GPFA from baseline, measured from a 2‐h window of electroencephalogram (EEG), and percentage changes in diary‐recorded seizures over 3 months (Pearson r = .72, p = 5 × 10⁻¹², 95% confidence interval [CI] = .16–.57). Changes in GPFA burden and diary‐recorded seizure frequency are normalized to baseline values, for Week 12, 24, and 36 data, with a "zero" value representing baseline levels. Negative values on the x‐ and y‐axes correspond to reductions in GPFA and diary‐recorded seizures, respectively (i.e., values within the box marked with a dotted line). Values outside of the box represent an increase in GPFA and diary‐recorded seizures from baseline. The line of best fit is shown in blue, with the 95% CI shaded in gray. (B) Similarly, there was a positive correlation between percentage change from baseline in EEG‐Clinical seizures (y‐axis) and diary‐recorded seizures (Pearson r = .38, p = .001, 95% CI = .16–.57). (C) Conversely, changes in EEG‐seizure counts from baseline (y‐axis) correlated poorly with percentage changes of diary‐recorded seizures from baseline (Pearson r = .07, p = .58, 95% CI = −.17 to .3).

4. DISCUSSION

Using data from the ESTEL trial of DBS, we show that changes in GPFA duration and count, measured from intermittent 2‐h recordings of sleep EEG, track changes in the average number of seizures per day recorded over 3‐month intervals of seizure diaries. Although the ratio of GPFA to diary seizures varied widely from patient to patient, it appeared stable within the same patient over time, meaning that percentage reductions from baseline in GPFA were strongly associated with similar percentage reductions in diary seizures. This suggests that GPFA burden may be a fast and objective method to measure treatment response in patients with LGS.

Patients with LGS are usually unable to reliably self‐report seizures, leaving this task to carers who are busy comanaging seizures, their related injuries, and cognitive/behavioral comorbidities. ²⁶ Additional challenges to keeping accurate seizure diaries include frequent nocturnal seizures, which are sometimes subtle and easily missed if patients are sleeping independently. Conversely, seizure mimics, such as stereotypies in patients with intellectual disability, may be included in diary counts. A surrogate marker of seizure susceptibility could reduce the dependence on diary‐keeping, providing an objective marker of efficacy in clinical trials. Using an EEG biomarker to track treatment response may widen eligibility criteria for clinical trials, allowing inclusion of participants who are unable to maintain a seizure diary, for example, in the commonly encountered scenario where participants live in a group residential setting with multiple and/or rotating carers.

Although encouraging, we emphasize that our study only provides evidence for the potential utility of measuring GPFA in the context of DBS. Further studies including larger numbers of patients are required to determine whether the same association between GPFA and seizures persists during standard clinical care or in the setting of other treatment trial types (e.g., antiseizure medications). Furthermore, although GPFA is occasionally found in idiopathic generalized epilepsies ²⁷ , ²⁸ , ²⁹ and other syndromes associated with developmental and epileptic encephalopathy, ³⁰ , ³¹ we cannot comment on whether GPFA burden would be a helpful biomarker is those contexts. Moreover, although in the current study we used a manual approach to estimate GPFA burden, clinical feasibility may be enhanced by developing automated detection approaches.

Patients with LGS have a very high burden of epileptic activity. GPFA is particularly common (~2 discharges per minute of sleep EEG), consistent with a fundamentally unstable brain state. Although there was high variability between patients, LME modeling showed that, on average, one GPFA discharge (median duration = 1.2 s) corresponded to 20–25 diary‐recorded seizures over 3 months (equating to ~2–3 seizures over 1 week). We measured both the total number of GPFA events and their duration due to uncertainty about which method would be superior in capturing the gestalt of GPFA burden. Both performed similarly in correlating with diary seizures over 3‐month intervals, consistent with the high correlation between the number of GPFA events and their durations across the four study timepoints. This suggests that either could be a suitable biomarker when tracking treatment response in LGS.

Curiously, we did not find a statistical association between diary‐recorded seizures and EEG‐seizures or EEG‐clinical seizures, despite an overall reduction in both electrographic and diary‐recorded seizures after DBS treatment. ¹⁴ One possible explanation is that EEG‐seizures incorporated all seizure types, including trains of slow spike‐waves similar to those captured on prior video‐EEG monitoring during "atypical absence." The burden of "slow spike‐wave" electrographic seizures may show little association with seizure diaries in the ESTEL trial, which specifically excluded "atypical absence" due to difficulty in their reliable detection and differentiation from patients' background cognitive impairment. ¹⁴ Hence, seizure diaries in ESTEL largely captured tonic and atonic seizures, which may have a stronger association with GPFA burden due to the EEG similarities between these ictal and interictal events (Figure 1), suggesting they are driven by common brain mechanisms (and thus may respond similarly to treatment). This interpretation is supported by the significant correlation observed between percentage reductions in EEG‐clinical seizures and diary‐recorded seizures, but not for all EEG‐seizures. However, EEG‐seizures > 10 s correlated with seizure diary counts, indicating that longer seizure durations may be more appropriate to track diary‐recorded seizures. Finally, electrographic seizures were measured across the entire 24‐h recording, unlike GPFA burden, which was only measured in light sleep. Hence, another possible explanation is that GPFA burden in sleep is more strongly associated with seizure susceptibility than electrographic seizures in the awake state. Further studies may shed light on this apparent paradox.

Our previous functional neuroimaging studies have shown that GPFA discharges of LGS are expressed via specific cortical and subcortical networks, ⁴ similar to the brain areas active during tonic seizures. ³² We have shown that GPFA activation patterns are similar between pediatric and adult patients, and across underlying causes of the syndrome (structural, genetic, and unknown etiologies). ¹³ The strength of the association between GPFA burden and clinically documented seizures we report here provides further evidence that GPFA is central to the epileptic process of LGS, and as such may be a broadly usable biomarker of treatment response.

5. CONCLUSIONS

We demonstrate that changes in GPFA duration and count, measured from intermittent 2‐h recordings of sleep EEG, track changes in the average number of seizures per day recorded over 3‐month intervals of seizure diaries, in patients with LGS undergoing a DBS treatment trial (ESTEL). Given that percentage reductions from baseline GPFA burden were strongly associated with similar percentage reductions in diary seizures, this suggests that GPFA burden may be a fast and objective method to measure treatment response in LGS.

AUTHOR CONTRIBUTIONS

L.J.D., A.E.L.W., and J.S.A. contributed to the conception and design of the study. All authors contributed to acquisition and analysis of data. L.J.D., A.E.L.W., and J.S.A. contributed to the drafting of text or preparing of figures.

FUNDING INFORMATION

Our work was supported with funding from the National Health and Medical Research Council (project grant #1108881). L.J.D. is supported by an Australian Government Research Training Program Scholarship. A.E.L.W. was supported by a postdoctoral fellowship from the Lennox–Gastaut Syndrome Foundation and an Early Career Researcher Grant from the University of Melbourne.

CONFLICT OF INTEREST

L.J.D., A.E.L.W., C.S., and A.R. report no conflicts of interest relevant to this study. J.S.A. has received honoraria from Medtronic. K.J.B. and W.T. are cofounders and hold shares and options in DBS Technologies. K.J.B. and W.T. are also named inventors on related patents, which are assigned to DBS Technologies. W.T. has received honoraria from Medtronic and Boston Scientific. Medtronic and Boston Scientific are manufacturers of DBS equipment. 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

Dalic LJ, Warren AEL, Spiegel C, Thevathasan W, Roten A, Bulluss KJ, et al. Paroxysmal fast activity is a biomarker of treatment response in deep brain stimulation for Lennox–Gastaut syndrome. Epilepsia. 2022;63:3134–3147. 10.1111/epi.17414

DATA AVAILABILITY STATEMENT

The ESTEL trial data are not publicly available due to organizational ethics constraints.