Prolonged Postictal hypercapnia is associated with Delayed Recovery of Consciousness after Generalized Convulsive Seizure

Abstract

Background

Postictal impairment of breathing and arousal is a significant contributor in many SUDEP cases. Central CO₂ chemoreception, which mediates ventilatory response, is often blunted after generalized convulsive seizures (GCS). It is plausible that the central chemoreceptors that mediate arousal in response to hypercapnia are similarly impaired after GCS, thereby contributing to delayed recovery of consciousness (ROC). Therefore, we hypothesized that greater severity of postictal hypercapnia is associated with longer latency to early signs of ROC following GCS.

Methods

Epilepsy patients were monitored with video-EEG, respiratory effort and airflow, transcutaneous (Tc) CO₂, and pulse oximetry. Primary outcomes were latency to early signs of ROC and duration of postictal immobility (PI). The interictal hypercapnic ventilatory response (HCVR) was measured in a subset of patients. Univariable and multivariable generalized linear models were created.

Results

A total of 149 GCS were recorded from 86 of 351 participants. Evaluable peri-ictal TcCO₂ and video-EEG data were available in 47 GCS in 38 participants (Female = 50%). In multivariable models, duration of postictal hypercapnia and temporal lobe seizures were significantly associated with prolonged latency to early signs of ROC, while preictal TcCO₂ level and postictal generalized EEG suppression (PGES) duration were significantly associated with PI duration. Interictal HCVR slope was negatively associated with postictal hypercapnia.

Interpretation

Interictal HCVR is negatively associated with postictal hypercapnia, and the latter is associated with delayed ROC. In addition, preictal TcCO₂ and PGES duration are associated with PI duration. The significance of these findings should be examined in future studies as potential risk markers for SUDEP.

Introduction

Sudden unexpected death in epilepsy (SUDEP) accounts for a significant proportion of premature deaths in people with epilepsy. [1] SUDEP occurs more often in bed, often unwitnessed, and typically with the decedent in the prone position. [2] The risk is 27 times greater for those who have had one or more generalized convulsive seizures (GCS) in the previous year, and 67 times greater when combined with not sharing a bedroom. [3] Increasing evidence suggests postictal impairment of breathing and arousal plays a critical role in the pathophysiology of many SUDEP cases. [2, 4–6]

Consciousness is a complex cognitive process often broken down into two components—content of consciousness and level of consciousness. The latter primarily refers to arousal, alertness, and attentiveness. [7] After GCS, patients often exhibit depression of consciousness and immobility, likely related to widespread dysfunction of cortical and subcortical neural networks. [8, 9] Unlike arousal from normal sleep, which is well characterized, [10] “arousal” after GCS is not. In addition, sequence of events during recovery of consciousness (ROC) after GCS has not been systematically characterized. However, it may resemble emergence from general anesthesia, where patients first regain muscle tone and exhibit non-purposeful movements, followed by recovery of the level of consciousness, and later the content of consciousness including higher cognitive functions. [11] Because brainstem serotonergic dysfunction is thought to be one mechanism underlying postictal immobility (PI) [12], the emergence of spontaneous non-purposeful movement, marking the end of PI, likely suggests improving subcortical neural network function that may precede ROC. On the other hand, postictal emergence of clinical behaviors like—purposeful movement and/or the ability to engage with the environment requires some degree of attentiveness/alertness and arousal indicating recovery of the level of consciousness. Therefore, these behaviors can be considered early signs of ROC. Together, these early signs of ROC, and potentially the end of PI, can be used to measure postictal "arousal" following GCS. Currently, the relationships between duration of PI, latency to ROC, and periictal ventilation (CO₂) are not clear.

The automatic respiratory rhythm is generated in the lower brainstem from activity of complex neural networks that include the pre-Bötzinger complex. [13, 14] “Chemoreceptor drive,” mediated by central CO₂ chemoreception (CCR), and “wakefulness drive,” mediated by forebrain activity during normal consciousness, represent the two main sources of tonic drive to control the level of ventilation. [15, 16] A population of serotonin (5-HT) neurons in the medullary raphe are essential for CCR mediated respiratory response to elevated CO₂. [17] In contrast, substantial evidence indicates that a group of dorsal raphe nucleus 5-HT neurons play a critical role in CO₂-induced arousal. [18, 19] Nevertheless, other key regions—including the retrotrapezoid nucleus, external lateral parabrachial neurons, orexinergic neurons in the lateral hypothalamus, and other areas—may also contribute. [20–22]

Based on the critical role of 5-HT neurons in CCR, it has been proposed that seizure-related dysfunction of these neurons likely contributes to postictal respiratory and arousal dysfunction, thereby increasing the risk of SUDEP. [23] Supporting this hypothesis, recent studies indicate that some seizures cause prolonged impairment of ventilation and CCR through a 5-HT-dependent mechanism. [6, 24, 25] In humans, CCR can be measured by the hypercapnic ventilatory response (HCVR), which may remain attenuated for hours after GCS. [6] Furthermore, GCS are also frequently associated with varying degrees of postictal hypercapnia, consistent with hypoventilation. [26–28] Given these findings, it is possible that chemoreceptors that mediate arousal in response to hypercapnia are also impaired after GCS, thereby contributing to delayed ROC after GCS. Therefore, we hypothesize that greater severity of postictal hypercapnia is associated with delayed ROC following GCS. We tested this by examining the relationship between severity of postictal hypercapnia and latency to early signs of ROC after GCS in patients with epilepsy in a prospective observational study.

Methods

Study population

This prospective observational study was conducted in the Epilepsy Monitoring Unit (EMU) of the University of Iowa and received Institutional Review Board approval. Participants were people with epilepsy who were admitted for video-EEG monitoring from March 2015 to April 2025. Participants with active cardiopulmonary disease, progressive neurological disorder (other than epilepsy), previous stroke or intracerebral space occupying lesion, and confirmed or possible pregnancy were excluded. Some data from some participants have been previously published which included preliminary findings on peri-ictal TcCO₂, interictal HCVR, heart rate variability, other respiratory changes including inter-breath variability in relation to GCS. [27, 29–32] However, the relationship between periictal TcCO₂ and early signs of ROC and PI evaluated in the current study is novel.

Continuous video-EEG and cardio-respiratory monitoring

Continuous, time-synchronized video-EEG was recorded (Nihon Kohden, Toyko, Japan; sampling rate = 1000 Hz) with scalp electrodes applied following the 10–20 international system, along with additional inferior-anterior temporal electrodes (T1/T2). Electrocardiography (EKG) was recorded using three leads. Continuous respiratory monitoring included capillary oxygen saturation (SpO₂) and transcutaneous (Tc) CO₂ (SenTec Digital Monitoring System,Therwil BL, Switzerland). Prior studies have demonstrated good agreement between ETCO₂ and TcCO₂ during steady-state conditions. [33, 34] Respiratory inductance plethysmography was performed using chest and abdominal belts (Pro-Tech zRIP DuraBelt, Philips Respironics, Murrysville, PA). Airflow assessment was monitored using a nasal pressure transducer (BiNaps, Salter Labs, Arvin, CA) and oronasal thermistor (ThermiSense, Salter Labs, Carlsbad, CA).

The clinical team determined anti-seizure medication withdrawal, activation procedures, duration of video-EEG and peri-ictal interventions (repositioning, oral suctioning and administration of oxygen) based on each patient’s specific clinical needs.

Only GCS with evaluable peri-ictal TcCO₂ data were considered for semiological evaluation and statistical modeling. GCS without video or with inadequate and/or obscured video recordings were excluded from analysis. Video-EEG data was reviewed by two epileptologists (RKS and PL), one blinded and one non-blinded to respiratory data. Seizure-related variables included type of GCS as classified by Alexandre et al. (GCS with symmetric tonic arm extension [type 1], GCS with symmetric clonic without tonic arm extension or flexion [type 2] and GCS with unilaeral or asymmetric tonic arm extension or flexion [type 3]) [35], seizure localization, vigilance state prior to seizure onset, seizure duration, convulsion duration, tonic phase duration, duration of PI as previously defined [36], and latency to early signs of ROC— defined as the time from seizure end to the first occurrence of: (a) purposeful movement, or (b) the ability to engage with the environment, even if the individual is not yet fully oriented or able to speak normally. Postictal generalized EEG suppression (PGES) was assessed by blinded reviewers (JH, SL) using a previously published method. [37] The periictal period was defined as the time interval from 2 minutes before seizure onset to 20 minutes after seizure termination on EEG. Periictal apneas were classified as ictal central apnea (ICA), defined as one or more missed breaths occurring before GCS, and post convulsive central apnea (PCCA), defined as the absence of breathing movement for at least 5 seconds following termination of the convulsion. [30] The preictal baseline TcCO₂ was defined as the average TcCO₂ level measured every 15 seconds over the 1-minute interval from 2 minutes to 1 minute before seizure onset on EEG, consistent with prior studies. [27] [38, 39] Duration of TcCO₂ rise (duration of hypercapnia) was defined as the time elapsed between elevation of TcCO₂ ≥10% above preictal baseline value to return to <10% above preictal baseline level. [27] Delta TcCO₂ was calculated as the difference between peak postictal (the highest TcCO₂ value recorded during the postictal period) and baseline preictal TcCO₂ levels. Duration of hypoxemia was defined as the time elapsed between when SpO₂ dropped below 90% to when it returned to 90% or higher.

Peri-ictal intervention was defined as the occurrence of one or more of the following nursing actions during the seizure or within three minutes of seizure termination: repositioning, oral suctioning, and/or administration of supplemental oxygen. Interventions implemented within one minute of seizure termination were classified as early interventions.

Clinical variables

Clinical variables were collected using a standardized questionnaire and review of the electronic medical record. Variables considered for analysis included: age, sex, duration of epilepsy, putative epileptogenic zone (based on video-EEG and imaging when available), history of obstructive sleep apnea (OSA), current and past anti-seizure medicines (ASMs), and active use of selective serotonin or serotonin-norepinephrine reuptake inhibitors (SSRI/SNRIs).

A subset of participants also underwent measurement of HCVR during an interictal awake state using a hyperoxic rebreathing technique (Ultima PFX Respiratory Gas Analyzer, MGC Diagnostics, Saint Pul, MN) as described previously.[27] End-tidal (ET) CO₂ was continuously monitored during this test but it was not available during the periictal period. HCVR slope (change in minute ventilation/change in ETCO₂) was derived as previously described. [29]

Statistical methods

Continuous variables were summarized using median and inter-quartile range (IQR), while categorical variables were summarized using counts and percentages. Univariable models were constructed to assess the effect of independent variables on each of the two outcomes: latency to early signs of ROC and duration of PI. Since both outcomes were continuous and right-skewed, gamma models were used. To ensure data integrity, only one GCS per patient was included in the analysis. When multiple GCS were recorded from a participant, the seizure with a complete set of variables was selected for analysis; if multiple eligible seizures were available, the first recorded seizure was used.

For each outcome, multivariable model selection was conducted using an all-subsets regression approach aimed at identifying the variable set with minimum Akaike Information Criterion (AIC). Variables with considerable missing data were excluded to maximize the statistical power of these multivariable models. All possible models incorporating three or fewer variables from the candidate variable pool were considered, and the model with the lowest AIC was selected as the best model for each outcome. Due to the relatively small sample size, only main effects were evaluated and not interaction terms.

Estimate of parameters was expressed as mean ratio (MR) for generalized linear model with log link function. Scaling factors for covariates were selected to facilitate interpretation of the statistical results (e.g., for PGES, a 1-unit change corresponded to 10 seconds; for epilepsy duration, a 1-unit change corresponded to 5 years). Spearman’s correlation was used to assess relationships between the two primary outcomes. An exploratory analysis was performed to evaluate the relationship between interictal HCVR slope and peri-ictal TcCO₂ changes using Spearman’s correlation without adjustment for multiple comparisons. We also performed an additional exploratory analysis to evaluate the relationship between baseline ETCO₂ during HCVR and preictal TcCO₂ for seizures that occurred during wakefulness using Spearman’s correlation.

All analyses were conducted using R version 4.3.2. A p-value <0.05 was considered statistically significant.

Results

Patient population, clinical variables and seizure characteristics

Of 351 patients enrolled, 86 experienced one or more GCS, resulting in a total of 149 GCS. Forty-seven GCS in 38 patients (50% female) had evaluable TcCO₂ monitoring and video-EEG data suitable for detailed analysis of ictal and postictal behavior. Six patients had two GCS, and one patient had three; however, only one seizure per patient was included for building univariable and multivariable models to avoid within-patient bias. Interictal HCVR measurement was available in 31 patients.

Of the 38 GCS, 35 (~92%) were focal to bilateral tonic-clonic seizures (24/35; ~69% were temporal lobe in origin), and the remaining 3/38 (~8%) were generalized tonic clonic seizures related to genetic generalized epilepsy.

Assessment of ICA was not possible in 5 GCS because of significant artifacts in the respiratory channels. ICA was identified in 27 seizures (~71%), of which 21(~78%) occurred in patients with temporal lobe epilepsy. PCCA was present in 4 GCS (10.53%). Preictal TcCO₂ levels were available for 37 GCS. Peak postictal TcCO₂ level was available in 31 GCS, duration of hypercapnia was available for 33 GCS, and the change in TcCO₂ induced by the seizure (delta TcCO₂) was available for 31. All seizures except three had postictal TcCO₂ data available and were associated with varying degrees of postictal TcCO₂ elevation. Examples of peri-ictal TcCO₂ changes and latency to early signs of ROC in two participants are shown in Figure 1. Hypoxemia duration was available for 26 GCS while SpO₂ nadir data was available for 23. Peri-ictal interventions were performed in 37 of 38 seizures. In 35 of 37 cases (92%) this occurred within one minute of seizure termination; therefore, intervention was excluded from statistical modeling. Descriptive statistics for the variables are summarized in Table 1.

Figure 1.

Examples from two participants showing peri-ictal TcCO₂ level and latency to early signs of recovery of consciousness (ROC) after generalized convulsive seizure. Graph starts ~two minutes prior to seizure onset (black vertical bar). Seizure end is depicted by purple vertical bars. Nursing interventions were performed before seizure end in both patients. Latency to early signs of ROC (red arrow) was delayed in the participant with a higher preictal baseline, higher peak, and longer duration of TcCO₂ (black curve) compared to the second patient (blue curve).

Table 1.

Patient, clinical, and seizure characteristics

Characteristics	Median (1st, 3rd quartile) or n (%)
Number of patients	38 (100%)
Age, yrs	33 (26, 40)
Sex, F	19 (50%)
Epilepsy duration, yrs	11 (4, 18)
Epilepsy localization
Temporal	24 (63%)
Temporal plus	3 (8%)
Extra-temporal	8 (21%)
Generalized	3 (8%)
Obstructive sleep apnea, Y	3 (8%)
SSRI/SNRI use, Y	7 (18%)
HCVR slope, L/min/mm Hg	1.82 (1.26, 2.47)
GCS type
1: Symmetric tonic arm extension	2 (5%)
2: Symmetric clonic without tonic arm extension or flexion	5 (13%)
3: Asymmetric tonic arm extension or flexion	31 (82%)
Seizure arising from sleep, Y	14 (37%)
Seizure duration, sec	98 (78, 160)
Duration of convulsion, sec	61 (53, 75)
Duration of tonic phase, sec	7 (5, 9)
PGES presence, Y	33 (87%)
PGES duration, sec	17 (4, 42)
ICA (Y)	27 (71%)
PCCA (Y)	4 (11%)
Preictal TcCO2,mm Hg	40 (36, 44)
Peak postictal TcCO2, mm Hg	56 (49, 66)
Duration of postictal hypercapnia, sec	363 (268, 583)
Periictal delta TcCO2, mm Hg	17 (11, 20)
Duration of periictal hypoxemia, sec	115 (76, 181)
Periictal SpO2 nadir, %	58 (42, 72)
Duration of PI, sec	46 (29, 66)
Latency to early signs of ROC, sec	68 (47, 291)
Periictal intervention, Y	37 (97%)
Early intervention, Y	35 (92%)

Abbreviations: Yrs- years, F- female, Y-yes or present, SSRI/SNRI- selective serotonin reuptake inhibitors/serotonin-norepinephrine reuptake inhibitors, HCVR- hypercapnic ventilatory response, L- liter, Min- minute, GCS- generalized convulsive seizure, Sec- seconds, PGES- postictal generalized EEG suppression, ICA- Ictal central apnea, PCCA- post convulsive central apnea, TcCO₂- transcutaneous carbon dioxide, mm Hg- millimeters of mercury, SpO₂- oxygen saturation, PI- postictal immobility, ROC- recovery of consciousness

Sequence of events from end of convulsion to early signs of ROC

Immediately after convulsion termination, all participants exhibited loss of consciousness accompanied by PI coinciding with spontaneous return of breathing in all except 4/38 (~11 %) participants who exhibited brief post-convulsive central apnea before resuming breathing. During this period, peri-ictal hypoxemia and hypercapnia of varying severity was observed. Oral suctioning was performed within 1 minute of seizure termination in 20/38 (52.6%) participants; however, only 3/20 (15%) demonstrated a cough reflex. As recovery progressed, spontaneous cough prior to ROC was observed in 4/35 (11.4%) participants.

At the end of convulsions or shortly thereafter, eyes were completely or partially closed in all participants. Roving eye movements were observed in 2/38 (5.2%), while 4/36 (11.1%) participants exhibited rare or infrequent eye opening or blinking without roving movements before early signs of ROC. Postictal purposeless, unintentional movements consistent with agitation were observed in 14/38 (37%) participants. In most participants (25/38; 66.8%), PI resolved with a median of 38 secs (IQR = 253 secs) before early signs of ROC; in the rest, they coincided.

Association of latency to early signs of ROC and duration of PI

In univariable analyses, duration of hypercapnia showed a trend towards a statistically significant association with latency to early signs of ROC (MR 1.10, 95% CI 1.00 to 1.23, p = 0.05) while there was no significant association with duration of PI (MR 1.07, 95% CI 0.92 to 1.25, p = 0.3). On the other hand, preictal TcCO₂ level was significantly associated with duration of PI (MR 1.11, 95% CI 1.05 to 1.17, p = 0.001) but not with latency to early signs of ROC (MR 1.04, 95% CI 0.97 to 1.11, p = 0.3). Likewise, duration of PGES was not associated with latency to early signs of ROC (MR 1.03, 95% CI 0.97 to 1.11, p = 0.4) while there was a significant association with duration of PI (MR 1.08, 95% CI 1.01 to 1.19, p = 0.04). Furthermore, temporal lobe seizures were significantly associated with duration of PI (MR 2.36, 95% CI 1.06 to 4.98, p = 0.03) but not latency to early signs of ROC (MR 1.98, 95% CI 0.76 to 4.82, p = 0.15). Additionally, compared to type 2 GCS (reference), type 3 GCS showed a trend towards statistically significant association with duration of PI (MR 3.69, 95% CI 0.82, 11.2, p = 0.05) but not with latency to early signs of ROC (MR 1.43, 95% CI 0.38, 3.90, p = 0.5). Point estimates with corresponding 95% CI for these and other variables are presented in Table 2.

Table 2.

Univariable regression model for latency to early signs of recovery of consciousness (ROC) and duration of postictal immobility (PI)

	Latency to early signs of ROC				Duration of PI
Characteristic	N	Mean ratio	95% CI1	p-value	Mean ratio	95% CI1	p-value
Age (year)	38	1.02	1.0, 1.05	0.3	1.02	0.99, 1.07	0.3
Sex (Female)	38	0.98	0.45, 2.13	>0.9	1.85	0.75, 4.55	0.9
Epilepsy duration (years)#	38	0.99	0.94, 1.03	0.5	0.99	0.95, 1.04	0.7
SSRI/SNRI use (Yes)	38	1.09	0.44, 3.32	0.9	1.43	0.49, 5.63	0.6
Duration of PGES (seconds)##	38	1.03	0.97, 1.11	0.4	1.08	1.01, 1.19	0.04*
GCS type
1: Symmetric tonic arm extension	2	0.41	0.07, 4.65	0.4	2.0	0.26, 33	0.5
2: Symmetric clonic without tonic arm extension or flexion (reference)	5	-	-	-	-	-	-
3: Asymmetric tonic arm extension or flexion	31	1.43	0.38, 3.90	0.5	3.69	0.82, 11.2	0.05
Temporal lobe seizures (Yes)	38	1.98	0.76, 4.82	0.15	2.36	1.06, 4.98	0.03*
Seizure during sleep (Yes)	38	0.48	0.23, 1.04	0.06	0.70	0.30, 1.79	0.4
Duration of seizure (seconds)	38	1.00	0.99, 1.01	0.7	1.00	0.99, 1.01	0.4
Duration of convulsion (seconds)	38	1.02	0.99, 1.04	0.2	1.01	0.98, 1.04	0.6
Duration of tonic phase (seconds)	38	1.02	0.96, 1.09	0.7	0.95	0.88, 1.05	0.3
Duration of postictal hypercapnia (minute)	33	1.10	1.00, 1.23	0.05	1.07	0.92, 1.25	0.3
Preictal TcCO2 (mm Hg)	37	1.04	0.97, 1.11	0.3	1.11	1.05, 1.17	0.001*
Peak TcCO2 (mm Hg)	31	1.03	0.99, 1.07	0.2	1.07	1.04, 1.10	<0.001*
Delta TcCO2 (mm Hg)	31	1.04	0.98, .10	0.2	1.09	1.03, 1.15	0.005*
Duration of hypoxemia (minutes)	36	1.00	0.99, 1.01	0.8	1.00	1.00, 1.01	0.3
Severity of hypoxemia (SpO2 nadir, %)	23	1.02	1.00, 1.04	0.09	0.98	0.96, 1.00	0.12

Abbreviations: SSRI/SNRI- selective serotonin reuptake inhibitors/serotonin-norepinephrine reuptake inhibitors, HCVR- hypercapnic ventilatory response, L- liter, Min- minute, GCS- generalized convulsive seizures, PGES- postictal generalized EEG suppression, PCCA- post convulsive central apnea, TcCO₂- transcutaneous carbon dioxide, mm Hg- millimeters of mercury, SpO₂- oxygen saturation

^#1 unit = 5 years,

^##1 unit = 10 seconds,

^*p <0.05

In multivariable analyses, the best fitting multivariable model showed both duration of postictal hypercapnia (MR 1.14, 95% CI 1.05–1.25, p = 0.005) and temporal lobe seizures (MR 3.28, 95% CI 1.63–6.41, p < 0.001) were significantly associated with latency to early signs of ROC. These relationships are illustrated in Figure 2 (A & B). In contrast, sex was not statistically significant (Table 3). In this model, each additional minute of postictal hypercapnia is associated with a 14% increase in latency to early signs of ROC, after adjusting for temporal lobe seizure and sex. Accordingly, an increase of approximately 5.3 minutes of postictal hypercapnia is associated with doubling of latency to early signs of ROC.

Figure 2.

A) Scatter plot (N = 33) with regression curve and 95% confidence interval demonstrating the association between latency to early signs of recovery of consciousness (ROC) and duration of hypercapnia. (B) Box and whisker plot with mean and 95% confidence intervals of latency to early signs of ROC by seizure type. C and D (N = 37). Scatter plots with regression curve and 95% confidence interval of the relationship between duration of postictal immobility (PI) and preictal transcutaneous CO₂ (C), and the association between duration of PI and duration of postictal generalized EEG suppression (D).

Table 3.

Multivariable Model- Latency to early signs of recovery of consciousness (ROC)

Covariates (N = 33)	Mean ratio	95% CI1	p-value
Duration of postictal hypercapnia (minutes)	1.14	1.05, 1.25	0.005*
Temporal lobe seizure (Yes)	3.28	1.63, 6.41	<0.001*
Sex (Female)	0.55	0.27, 1.08	0.08

Abbreviations: CI- confidence interval, TcCO₂- transcutaneous carbon dioxide

^*p<0.05

In the best fitting multivariable model for duration of PI, both preictal TcCO₂ level (MR 1.11, 95% CI 1.06–1.16, p < 0.001) and duration of PGES (MR 1.07, 95% CI 1.03–1.13, p < 0.001) were significantly associated with duration of PI while epilepsy duration was not (Table 4 and Figure 2, C & D). This model showed that each 1mm Hg increase in preictal TcCO₂ level is associated with an 11% increase in PI duration. Accordingly, an increase of approximately 6.6 mm Hg in preictal TcCO₂ is associated with a two-fold increase in duration of PI. Likewise, each additional 10-second increase in PGES is associated with a 7% increase in PI duration, such that an increase of approximately 100 seconds in PGES corresponds to a two-fold increase in PI duration.

Table 4.

Multivariable Model- Duration of postictal immobility (PI)

Covariates (N = 37)	Mean ratio	95% CI1	p-value
Preictal TcCO2 (mm Hg)	1.11	1.06, 1.16	<0.001*
Duration of PGES (seconds)##	1.07	1.03, 1.13	<0.001*
Duration of Epilepsy (years)#	0.97	0.95, 1.00	0.07

Abbreviations: CI- confidence interval, TcCO₂- transcutaneous carbon dioxide, mm Hg- millimeters of mercury, PGES- postictal generalized EEG suppression

^#1 unit = 5 years,

^##1 unit = 10 seconds,

^*p<0.05

Because the best-fit models for each outcome included at least one covariate that was not significant in univariable analyses, we reviewed the top five models for each outcome to assess the stability of the different covariates and whether they were included in all or most of the top five possible models. As expected, for each outcome, two statistically significant variables were consistently selected in all five of the five best-performing models while the non-significant predictor was only selected for the chosen optimal model and not in any other top five models (Supplemental eTable 1 and eTable 2).

The two outcome variables, latency to early signs of ROC and duration of PI, were significantly correlated with each other (Spearman’s rho = 0.38, p= 0.02).

Interictal HCVR slope and postictal hypercapnia

We previously reported that an attenuated interictal HCVR slope was associated with more severe postictal hypercapnia following GCS [27]. In this study, we re-examined this relationship in a larger sample. We found significant negative relationships between interictal HCVR slope with peri-ictal TcCO₂ (preictal, peak postictal and delta [peak minus preictal]) levels and a trend towards a statistically significant negative correlation between interictal HCVR and with duration of postictal hypercapnia (Supplemental eTable 3 and eFigure 1). These correlations suggest a link between impaired baseline CCR and delay in recovery of respiratory dysfunction after GCS. Because only a subset of participants underwent HCVR measurement, to maximize statistical power we were unable to include it as a covariate in the primary analysis to examine the relationship between HCVR slope and latency to early signs of ROC.

An additional exploratory analysis showed an excellent correlation between baseline ETCO₂ during the HCVR test and preictal TcCO₂ of GCS that occurred during wakefulness (Spearman’s rho = 0.61, p = 0.006, Supplemental eFigure 2).

Discussion

Postictal impairments of respiratory and arousal mechanisms, either independently or in combination, are believed to play a crucial role in the pathophysiology of SUDEP. [2, 4, 5, 23] However, the relationship between these phenomena remains poorly understood. We have previously shown that seizures can impair respiratory CCR. [6] Here, we show for the first time that prolonged postictal hypercapnia is independently associated with delayed ROC following GCS. Although we cannot establish direct cause-and-effect, CO₂ is a potent stimulus for arousal. [20] Therefore, the observed delay in ROC despite elevated CO₂ levels may suggest impaired CO₂-induced arousal, which could represent a mechanism contributing to prolonged postictal impaired consciousness. Furthermore, we also found that the preictal CO₂, a proxy for preictal CCR or HCVR slope, and PGES duration were associated with duration of PI after GCS.

The definition of “consciousness” is often nebulous. However, there is broad agreement that consciousness comprises both different “levels of consciousness” and the “contents of consciousness.” [8, 40] We initially considered assessing “arousal” as our primary outcome variable, as arousal is generally regarded as one component of the level of consciousness and is relatively well defined in the sleep literature. However, defining “arousal” in the postictal period is considerably more challenging because the postictal state differs substantially from normal sleep and lacks well-established assessment standards. We therefore selected two observable clinical behaviors—purposeful movement and/or the ability to engage with the environment through verbal or motor responses—as indicators of early ROC.

Seizures are associated with varying degree of postictal depression of both content and level of consciousness. [8] The sequence of events during ROC after GCS has not been well studied, therefore, we attempted to characterize it. Although latency to early signs of ROC varies among participants, we observed a relatively consistent sequence of events: spontaneous recovery of breathing along with a period of PI followed by gradual recovery of brainstem functions such as the cough reflex and eye movements, before the appearance of early signs of ROC. This sequence is analogous to emergence from anesthesia-induced coma. [11, 41] However, complete ROC after GCS is generally a gradual process which may take hours to days. [42, 43] Given that SUDEP often occurs in the immediate postictal period, we carefully chose our primary outcome variables. Because consciousness depends on the integrated function of both cortical and subcortical networks, early signs of ROC likely reflect recovery across these networks. In contrast, resolution of PI may primarily reflect recovery of subcortical network function. This may explain the significant correlation (Spearman’s rho = 0.38, p = 0.02) between the two primary outcomes.

During GCS, intense muscle activity increases metabolic demand resulting in excess CO₂ and lactate production, while ventilation is simultaneously impaired, often resulting in severe acidosis. [44] CO₂ sensitive areas of the brain are responsible for increasing ventilation and induce arousal from sleep in response to elevated CO_2, to maintain CO₂ homeostasis. [18–22] The arousal response leads to an increase in upper airway tone, thus relieving obstruction in obstructive sleep apnea; it also moves the head when the face is obstructed (e.g., by a pillow), and increases “wakefulness drive” for breathing. [15, 16] This arousal response may be critical during the immediate postictal period following GCS, when consciousness is severely depressed, thereby abolishing “wakefulness drive” and predisposing to upper airway collapse or obstruction.

We found that each additional minute of postictal hypercapnia was associated with a 14% increase in latency to early signs of ROC in multiplicative scale after adjusting for temporal lobe localization of seizure onset and sex. This relationship was detectable in our analyses despite the nearly uniform administration of early peri-ictal interventions. Thus, this association may be even stronger in an unattended setting. We also found that temporal lobe seizures were significantly associated with increased latency to ROC after accounting for duration of hypercapnia in multivariable model. Temporal lobe seizures even those that do not progress to GCS, can affect consciousness probably due to inhibition of cortical networks [8]. This may partly explain why those participants who progressed to GCS tended to exhibit a prolonged impaired consciousness.

CCR can be quantified by the HCVR test. [27] We previously showed that a low interictal HCVR slope is associated with severe postictal hypoventilation after GCS. [27] Our current study, with a larger sample, reaffirms this association. Furthermore, recent evidence also supports the notion that seizures can blunt CCR as measured by the HCVR thus further contributing to postictal hypoventilation. [6] This is potentially important because an increasing body of evidence suggests a critical role of 5-HT neurons in dorsal raphe nucleus in CO₂-induced arousal, in addition to other key brain regions that may contribute. [6, 18, 20–22]

There are at least two possible mechanisms for postictal attenuation of CCR. The first is a continued seizure-induced suppression of cortical activity, leading to loss of both cortical drive to breathe and inhibition of cortical networks involved in arousal. This would cause a reduction in both reduced ventilation and the arousal response to hypercapnia. Second, seizures may suppress critical areas of the brainstem including chemosensitive regions of the brain [6, 24, 45] and other subregions including the ascending reticular activating system. The latter mediates cortical arousal through diverse neural networks including serotonergic transmission from the midbrain, cholinergic input from the pedunculopontine and laterodorsal tegmental nuclei of the caudal midbrain and rostral pons, histaminergic neurons in the tuberomammillary nucleus, and noradrenergic transmission from the locus coeruleus. [19, 46] Therefore, attenuated chemoreception may not necessarily reflect inhibition of a specific neuronal population but rather a consequence of severe brainstem or cortical dysfunction resulting from GCS itself. However, focal seizures, without diffuse inhibition of cortical or brainstem function, can also attenuate CCR [6] suggesting that it alone may be enough to cause postictal hypercapnia and impaired CO₂-induced arousal. Notably, a neuropathological study found that patients who died of SUDEP had significant alteration to neuronal populations, including serotonin neurons, in the medulla. [47] These alterations may represent structural correlate of underlying altered chemoreceptor function, which is more vulnerable to suppression by GCS, creating circumstances favorable for SUDEP.

Because CCR is responsible for tight control of arterial partial pressure of CO₂ (PaCO₂), baseline PaCO₂ may serve as a useful proxy for HCVR when direct measurement is infeasible (e.g., during sleep or immediately prior to a seizure), a view supported by our prior findings. [27] Furthermore, baseline ventilation and the CCR are both attenuated during quiet sleep, resulting in elevated PaCO₂ compared with wakefulness. [48] In our current cohort, ~40% of GCS occurred during sleep, leading us to use preictal TcCO₂ level as a surrogate for preictal HCVR in statistical modeling. We found that each mm Hg increase in preictal TcCO₂ is associated with an 11% increase in the duration of PI after adjusting for epilepsy and PGES duration, significance of which is unclear to us. Our findings of a relationship between durations of PI and PGES is in line with prior reports.[36, 49, 50] Similarly, we observed a significant association between postictal hypercapnia (higher peak level) and PI duration in univariable analysis, a relationship that has also been previously reported. [51] Although peak TcCO₂ could not be included in the multivariable model due to substantial missing data, these findings nevertheless validate our analysis. While the pathophysiology of PI is unclear, serotonergic neurons may be involved. In animal studies, SSRI treatment prior to seizure induction shortened PI duration [12], and stimulation of dorsal raphe serotonin neurons reduced both the duration of PGES [12] and mortality. [52] In our study, the use of SSRI/SNRI was not significantly associated with both primary outcomes, possibly due to heterogeneity of drug type and dose used along with the limited statistical power of the study.

Few studies have examined factors associated with prolonged impairment of consciousness after GCS. [43, 53, 54] Two of these studies [43, 53] were conducted in the emergency room without EEG monitoring to confirm a true postictal state, did not evaluate breathing, and were not focused on early signs of ROC, which is more relevant to SUDEP. One recent study conducted in the EMU with respiratory monitoring found no association between duration of impaired awareness and duration of PGES or peri-ictal hypoxemia, consistent with our findings. [54]

It is possible that severity of GCS may directly affect both outcome variables. To control seizure severity, we carefully selected seizure related variables that are likely to reflect severity (e.g. type of GCS and durations of— seizure, tonic phase, convulsion etc.). Type of GCS showed a trend towards statistical significance in univariable model for PI duration, but it was not significant in the multivariable model. None of the remaining variables exhibited any meaningful relationship with either of our primary outcomes.

Our study has several limitations. Although we recorded 149 GCS, only one-third had evaluable peri-ictal TcCO₂ data, underscoring the practical challenges of continuous and high-quality monitoring in this setting. Furthermore, some key variables had considerable missing data. This limited our ability to build robust multivariable models with a larger number of candidate variables including some variables that were significant in univariable models. Attrition and incomplete datasets are common challenges in EMU-based studies and have been previously documented in similar investigations. [31, 55, 56] Additionally, peri-ictal interventions and clinical assessments were performed at the discretion of clinical staff, rather than following a standardized testing protocol with repeated assessments at prespecified time intervals. As a result, we were unable to systematically analyze the effects of these interventions on time to early signs of ROC and duration of PI. Indeed, it is possible some patients could have exhibited stimulus-induced ROC even earlier than assessed by video review.

Findings from this study only apply to those who suffer from GCS. Further research is needed to determine the generalizability of our findings to those with other types of seizures. Lastly, our findings demonstrate associations but do not establish causality. To confirm a mechanistic cause-and-effect relationship, future studies could expose participants to different levels of CO₂ or placebo gas during the postictal period following GCS and measure latency to ROC using a standardized protocol.

Conclusions

We found that attenuated interictal HCVR is associated with severity of postictal hypercapnia after GCS. We also showed that prolonged hypercapnia is associated independently with increased latency to early signs of ROC, suggesting a possible disruption of CO₂-induced arousal that may contribute to prolonged postictal coma. In addition, preictal hypercapnia and duration of PGES are associated with longer PI duration. An impaired baseline HCVR, delayed ROC and longer PI duration should be further studied as potential biomarkers for SUDEP which may help identify patients who would most benefit from targeted interventions.

Highlights.

• Interictal central CO₂ chemosensitivity as measured by the hypercapnic ventilatory response is negatively associated with the severity of postictal hypercapnia following GCS.

• Duration of postictal hypercapnia is associated with longer latency to early signs of recovery of consciousness (ROC) following GCS.

• Preictal hypercapnia and longer PGES duration are associated with a longer duration of postictal immobility (PI) following GCS

• An impaired HCVR, delayed ROC and prolonged duration of PI should be further studied as potential biomarkers for SUDEP.

Data Availability Statement

De-identified data are available upon reasonable request.