Paroxysmal slow wave events as a diagnostic biomarker for epilepsy: Lessons from rural Zambia

Abstract

Objective

Epilepsy affects more than 50 million people globally, with low‐ and middle‐income countries (LMICs) bearing the greatest burden due to limited medical resources and stigma. Electroencephalography (EEG) is a cost‐effective diagnostic tool, but its interpretation often requires unavailable expertise in rural areas. There is a pressing need for reliable, quantitative EEG biomarkers to enhance diagnosis, guide imaging, and monitor treatment.

Methods

We investigated paroxysmal slow wave events (PSWEs), transient markers of cortical network slowing, in scalp EEG recordings from epilepsy patients at the Kakumbi Rural Health Center in Zambia (n = 127) and from Bonn Epilepsy Center (n = 132). PSWE characteristics, including occurrence, duration, and spatial distribution, were analyzed. Source localization of PSWEs was performed using standardized low‐resolution brain electromagnetic tomography software.

Results

PSWEs were observed in all patients with epilepsy. Time in PSWE showed negative correlation with patient age (r = −.26, p = .003) and disease onset (r = −.25, p = .005), regardless of age. PSWE characteristics, including temporal and spatial distribution, were associated with disease severity and similar to drug‐resistant patients from Bonn Epilepsy Center. EEGs reported as “abnormal” had greater time in PSWE compared with “normal” EEGs (p = .024). Focal PSWE source localization suggested the presence of an intracranial lesion on computed tomography (area under the curve = .7).

Significance

This study supports previous research on the potential of PSWEs as a quantitative EEG biomarker in epilepsy. Automated analysis of PSWEs can enhance diagnostic accuracy and assist in screening patients for brain imaging, particularly in resource‐constrained settings. This approach offers a practical solution to bridge the diagnostic gap in LMICs that can potentially be used to improve epilepsy management and patient outcomes.

Key points.

• PSWEs are EEG patterns reported to occur more frequently in PWE.

• We studied PSWE features on EEGs from PWE in rural Zambia and at Bonn Epilepsy Center.

• Relative summated PSWE duration was higher in younger patients, with early disease onset, and in PWE with drug‐resistant epilepsy.

• Source localization of PSWEs could help identify lesional epilepsy and provide a promising approach for screening the need for brain imaging where resources are limited.

1. INTRODUCTION

Epilepsy stands as one of the most prevalent chronic neurological disorders, affecting more than 50 million individuals globally, with 80% of people with epilepsy (PWE) residing in low‐ and middle‐income countries (LMICs). ¹ The higher prevalence in LMICs is attributed to elevated instances of brain injury, increased exposure to central nervous system infections (e.g., neurocysticercosis, cerebral malaria), and suboptimal prenatal and perinatal care. ¹ , ² Although antiseizure medications (ASMs) serve as the conventional treatment, aiding most patients in symptom control, up to 75% of individuals in LMICs remain untreated due to various challenges, including scarcity of human professional resources, inconsistent availability of ASMs, and stigma. ¹ , ³ , ⁴

Electroencephalography (EEG) is a cost‐effective, noninvasive diagnostic tool that can assist in epilepsy diagnosis by detecting epileptiform activity and determining type of epilepsy and localization. However, the interpretation of EEG recordings necessitates the expertise of trained neurologists. ⁵ , ⁶ The small number of neurologists in many LMICs underscores the importance of developing computer‐assisted analysis tools, aiming to assist in EEG interpretations and enhance interrater reliability in the diagnosis and management of epilepsy.

Paroxysmal slow wave events (PSWEs) are often visually undetected EEG transients characterized by a median power frequency (MPF) between 1 and 6 Hz, lasting for at least 5 consecutive seconds. ⁷ This EEG feature has been shown to best distinguish between PWE and healthy controls, with PSWEs observed to be more prevalent in PWE compared with controls, suggesting their potential utility as a diagnostic tool for epilepsy. ⁷ Furthermore, high PSWE occurrence in EEGs recorded within 72 h of a first unprovoked seizure has been shown to predict future seizures and epilepsy with high diagnostic accuracy, achieving an area under the curve (AUC) of >.8, regardless of ASM use. ⁸ These findings collectively underscore the potential of PSWEs as a diagnostic biomarker for epilepsy.

In the current study, our primary aim is to validate the detectability of PSWEs in interictal EEGs recorded from PWE in a rural, low‐resource community. PWE were examined in the Kakumbi Chiefdom in the Mambwe District, Eastern Province of Zambia (for more details, see www.mfuweepilepsy.org). EEG was retrospectively analyzed for PSWE features in time and space. We further explored their potential as a diagnostic and a screening tool for patients' referral to advanced imaging (e.g., computed tomography [CT])—often not easily accessible in LMICs. Through this research, we aim to provide insights that may inform more targeted and effective strategies for the diagnosis and management of epilepsy in resource‐limited settings.

2. MATERIALS AND METHODS

2.1. Study design and site

The study received ethical approval from the University of Zambia Biomedical Research Ethics Committee. At the outset, only a small number of patients with epilepsy (n = 6) were being followed at the local clinic or receiving ASM. To address this gap, a community‐based door‐to‐door survey was conducted between 2018 and 2019 by local health workers affiliated with the Mfuwe Epilepsy Foundation. The survey involved 15 290 participants from the Kakumbi Chiefdom in Zambia's Eastern Province and aimed to identify individuals with known or suspected epilepsy. A structured questionnaire comprising 10 validated screening questions (Table S1) was used. ⁹ To this date, only individuals who reported a prior diagnosis of epilepsy (n = 198) were referred to the clinic, where they underwent a comprehensive, structured clinical interview conducted by a trained neurologist. The interview included patient demographic data, risk factors for epilepsy, details of seizure semiology, and treatment history. A neurological examination was performed on all patients. Participants provided informed consent, and for those unable to do so, a legal guardian consented on their behalf.

2.2. EEG recording and interpretation

EEG was available for 127 of 198 patients with epilepsy. EEG was not done in all patients due to compliance, technical issues (lack of power), or lack of trained manpower (EEG technician). EEG data were acquired utilizing a Micromed SAM 32FO fc1 EEG machine with 19 scalp electrodes, following the international 10–20 system. The average recording duration was 23.2 min, and a sampling rate of 256 Hz was employed. The recordings encompassed periods of both eyes open and eyes closed, as well as sessions involving hyperventilation.

EEG recordings were shared with four trained epileptologists (B.W., A.R., T.W., and F.B.) who classified the EEGs as normal or abnormal, and described the type and localization of epileptiform activity.

2.3. EEG analysis

We analyzed EEG data from all 127 patients who were examined at the Kakumbi Rural Health Center and underwent EEG between August 2019 and December 2024. The first 20 min of EEG data were analyzed using custom MATLAB scripts as previously described. ⁷ Data were preprocessed using the EEGLAB toolbox. ¹⁰ Preprocessing involved applying a bandpass filter (1–45 Hz) and baseline removal. Independent components analysis was done using the EEGLAB plugin ICLabel. Eye, muscle, or heart signals found to have a likelihood of 90% or greater were removed. Following preprocessing, recordings were segmented into 2‐s sliding windows with 1‐s overlap. For each segment, the MPF was calculated. A PSWE was defined when 5 or more consecutive seconds had an MPF between 1 and 6 Hz in more than a single channel (Figure 1). PSWE features, including occurrence (PSWE/min) and the summated relative duration (overall time percentage of the recording) were calculated. PSWE data were compared to data obtained from 132 PWE randomly selected from Bonn Hospital Epilepsy Center (Germany), of whom 62 were drug‐resistant and 70 were drug‐responsive, who were seizure‐free for at least 1 year.

FIGURE 1.

Detection of paroxysmal slow wave events (PSWEs) from a patient with epilepsy. (A) Upper trace shows average (avg) 90‐s recording taken from electrode T4. The lower trace shows the calculated median power frequency (MPF). Note that MPF mostly remains above 6 Hz. (B) Upper trace shows an average 90‐s recording from electrode T6 for the same patient. Note five PSWEs are detected (*MPF >1 and ≤ 6 Hz for >5 s). Inset: A PSWE segment with a slower time scale is shown.

2.4. Low‐resolution electromagnetic tomography

The three‐dimensional cortical distributions of PSWE current density were calculated using standardized low‐resolution brain electromagnetic tomography (sLORETA) software ¹¹ and as published for localization of epileptiform activity. ¹² , ¹³ , ¹⁴ For each participant, PSWE segment recordings were concatenated, and the resulting data were transformed into sLORETA files by applying the sLORETA transformation matrix to the cross‐spectral data. sLORETA images were generated for PSWE segments across all frequency bands. Scalp current density at each voxel was standardized within a realistic head model based on the MNI152‐2009c template. The cortical model comprised 6239 voxels, each with a spatial resolution of 5 mm, constrained to the cortical gray matter.

2.5. Drug levels

Based on international recommendations and availability, the proposed treatment for patients with PSWEs included carbamazepine, phenobarbital, valproic acid, or levetiracetam (the latter was kindly supplied by the ROW Foundation). ¹⁵ ASM levels were measured in a subgroup of randomly selected individuals (n = 50) who were prescribed ASM at least 1 month prior to the visit. Measurements were done at Metropolis Lab (Lusaka, Zambia).

2.6. CT scans

Brain CT scans were taken using GE Healthcare Revolution ACT. Scans were available for 42 patients who had EEG recordings. Brain CTs were conducted before and after the injection of an Ultravist (50 mL) agent. CTs were interpreted by a certified radiologist (G.B.‐A.).

2.7. Statistical analysis

Statistical analysis was conducted using Prism 9.5.0 (GraphPad). Nonnormal distribution necessitated the use of nonparametric tests. Spearman correlations were used to compare the variables age, age at onset, and duration of disease with time in PSWE. For group comparisons, a Mann–Whitney or a Kruskal–Wallis test with Dunn multiple comparisons test was used.

3. RESULTS

3.1. Participant and PSWE characteristics

Demographics of participants with EEG were similar to the entire cohort (N = 198) and are shown in Table 1. For the entire cohort (48.9% females), risk factors for epilepsy were documented for 128 participants (64.6%); the most commonly reported were family history (26.8%), febrile seizures (15.2%), pork consumption (14.6%), human immunodeficiency virus (7.1%), history of cerebral malaria (5.0%), developmental delay (3.0%), and traumatic brain injury (2.5%; Table 1). Of 67 school‐aged children, 43 (72.9%) reported attending school; of 92 adults (>18 years), only 13 (15.3%) reported being employed in the past year.

TABLE 1.

Participant demographics and seizure frequency of PWE.

Characteristic	Participants with EEG recordings, n = 127	All participants, n = 198
Age, years (IQR)	21 (22)	23 (22)
Sex, female, n (%)	63 (49)	97 (48.9)
Age at onset, years, (IQR)	9 (17.8) a	9.5 (17.8)
Duration of disease, years (IQR)	6 (10.2) a	8 (11.0)
Seizure frequency, n (%)	121	132
Daily	12 (10)	19 (14)
Weekly	19 (16)	21 (17)
Monthly	73 (60)	73 (55)
Yearly	17 (14)	19 (15)
Risk factor, n (%) b
Unknown	62 (48.8)	70 (35.4)
Family history	31 (24.4)	53 (26.8)
Febrile seizures	20 (15.7)	30 (15.2)
Pork consumption	15 (11.8)	29 (14.6)
HIV	10 (7.9)	14 (7.1)
Cerebral malaria	5 (3.9)	10 (5.0)
Developmental delay c	4 (3.1)	6 (3.0)
Traumatic brain injury	2 (1.6)	5 (2.5)
Stroke	1 (.8)	3 (1.5)

Note: The median and the IQR are reported for age, age at onset, and duration of disease.

Abbreviations: EEG, electroencephalographic; HIV, human immunodeficiency virus; IQR, interquartile range; PWE, people with epilepsy.

^a Based on 126 respondents.

^b Some individuals reported more than one risk factor.

^c Including reports on abnormal pregnancy and complications during delivery.

PSWEs were detected in all participants. The median summated PSWE duration (as a percentage of recording duration or “time in PSWE”) was 88.4% (range = 5.7–100). In 33 participants (25.9%), time in PSWE encompassed >99% of the recording time (Figure 2A). The median PSWE occurrence (min⁻¹) was 1/min (range = .04–2.99; Figure 2B). The relationship between PSWE occurrence and time in PSWE followed a nonlinear parabolic fashion. A low PSWE occurrence could mean two different scenarios depending on time spent in PSWE (Figure 2C). If the PSWE duration is short, a low occurrence would indicate minimal time spent in a PSWE. However, if the duration is long, then a low occurrence means that most of the time was spent in a PSWE (Figure 2D). Consequently, time in PSWE was selected as the most reliable indicator of high PSWE burden.

FIGURE 2.

Paroxysmal slow wave event (PSWE) features in patients with epilepsy (n = 127). (A) Relative frequency distribution of the median frequency. (B) Frequency distribution of PSWE occurrence (min⁻¹). (C) Relative sum of PSWE duration (time in PSWE). (D) Scatterplot showing the relationship between PSWE occurrence and time in PSWE.

3.2. Time in PSWE is greater in younger participants and those with earlier age at onset

Time in PSWE, but not occurrence (PSWEs/min), was negatively correlated with age (r = −.26, p = .003 and r = −.06, p = .51, respectively; Figure 3A). Notably, this correlation was absent when analysis was limited to participants >9 years old (n = 96, median age = 26, interquartile range = 19.3), indicating that the effect of age was predominantly in young children. Similarly, a negative correlation was also found between time in PSWE and epilepsy age at onset (r = −.25, p = .005; Figure 3B) only when young participants (<9 years old) were included. No correlation was found between duration of disease and PSWE features.

FIGURE 3.

Time in paroxysmal slow wave event (PSWE) is prolonged in children and with more severe disease. (A) Weak negative correlation between the percentage of time in PSWE and age in all Zambia participants (Spearman correlation: R = −.26, p = .003). Correlation becomes insignificant in participants >9 years old. (B) Negative correlation between time in PSWE and age at onset in all participants (r = −.25, p = .005). (C) Z‐score for time in PSWE data for 52 individuals with recurrent electroencephalograms (EEGs). No differences were found. (D) Time in PSWE was higher in the Zambian cohort compared with the Bonn cohort (p < .001). (E) Drug‐resistant participants and the Zambian cohort had significantly higher time in PSWE compared with Bonn seizure‐free people with epilepsy (PWE; p = .006, p < .001). (F) PWE with abnormal findings on EEG had greater time in PSWE compared to those with normal EEG interpretation (p = .02). *p < .05, **p < .01, ***p < .001.

3.3. PSWE features are not related to ASM treatment

To further evaluate whether PSWE features are influenced by ASM treatment and course of disease, we compared PSWE features in a subset of individuals (n = 52) who underwent a second EEG between 1 month and 3 years (mean = 1.3 years) after their initial diagnosis and treatment initiation. No significant differences were observed in any PSWE characteristics (PSWEs/min, median duration, frequency, or time in PSWE) between the first and second EEGs (paired t‐test, p = .8; Figure 3C; see Discussion). We next compared PSWE data from participants in Zambia to a cohort of PWE from Bonn, Germany (n = 132). To age‐match the cohorts, young (<5 years old, n = 10) participants from the Zambia cohort were excluded. Broad range of time in PSWE was found in both cohorts, with a higher time in PSWE in the Zambian sample (n = 117, p < .001; Figure 3C). We then compared the Bonn drug‐responsive group (seizure‐free for >1 year, n = 70) with the Bonn drug‐resistant (n = 62) group and Zambia cohort. Time in PSWE was similar between the Zambia cohort and the Bonn drug‐resistant group, and both were higher than in the Bonn drug‐responsive group (p < .001 and p = .006, respectively). No differences in PSWE occurrence were found between the groups.

To further test for the effect of medications on PSWE features, serum drug levels were measured in a subgroup of 50 participants from Zambia who were prescribed with either carbamazepine or phenobarbital. Therapeutic range levels were measured (in one of the two drugs) in only 14 participants (28%); in the remaining 36 participants (72%), the levels of the drugs were either lower than the therapeutic range or below the detectable range. No differences in PSWE features were found between the groups. Overall, these results support the notion that PSWEs are more likely to be related to disease severity and not to ASM treatment effect (see Discussion).

3.4. PSWE features and EEG interpretation

To further test the link between PSWE characteristics and EEG interpretation, a detailed evaluation by at least one epileptologist was available for 104 EEGs (75 studies were read by at least two interpreters). Twenty‐seven EEGs were reported “normal” (26%); for 69 EEGs (66%), epileptiform activity was reported. Although PSWEs were detected in all patients, on only 39 EEGs (38%) was “slowing” observed by an epileptologist (see Discussion). Notably, EEGs reported “abnormal” had greater time in PSWE compared with EEGs that were read as “normal” (p = .02). There were no differences in other PSWE features across EEGs with or without epileptiform activity.

3.5. PSWE localization and lesion on CT

In 42 participants, brain CT was available. Intracranial lesions were found in 12 PWE. Six had findings suspicious for neurocysticercosis (n = 2 with cysts and n = 4 with intraparenchymal calcification). A space‐occupying lesion was found in three participants (n = 1 suspected arteriovenous malformation and n = 2 unidentified intraparenchymal lesions). In two cases, a hypodense lesion presumed to result from previous stroke was observed, and hydrocephalus was observed in one patient. No difference was found in PSWE features (time in PSWE, PSWEs/min, or median frequency) between normal and lesional CTs (p > .5). We next used sLORETA to localize the suspected source of PSWEs (see Materials and Methods). sLORETA localized PSWEs within the same lobe in seven of nine patients with a single lesion (Figure 4). To assess the focality of the source of PSWEs, we measured brain volume with a current source >37% (1/e) of the maximal current source. Notably, PSWE localization in patients with a nonlesional CT was more diffuse compared to patients with a lesional CT. Receiver operating characteristic analysis suggested brain volume with high current density source for PSWEs has a potential to predict lesion on CT with a moderate discriminatory power (AUC = .70, 95% confidence interval = .62–.87; Figure 4).

FIGURE 4.

Representative cases demonstrating spatial paroxysmal slow wave event (PSWE) analysis and corresponding computed tomography (CT) scans. Column 1: Two‐dimensional (2D) topographical maps of time in PSWE (%) for each electrode. Column 2: Axial section showing the maximal standardized low‐resolution brain electromagnetic tomography (sLORETA) current density of PSWEs. Column 3: Three dimensional (3D) brain view sLORETA estimated localization of current density PSWE source. Column 4: Corresponding CT scans. (A) A 25‐year‐old male with a 15‐year history of epilepsy. Seizures (~1/months) were focal to bilateral tonic–clonic. Two of four epileptologists (50%) reported right frontal electroencephalographic (EEG) slowing. On CT, several cystic lesions with rim calcifications were observed in the right frontal region, with significant perifocal edema. Cyst walls were slightly enhanced on contrast scans. (B) A 42‐year‐old female with a 33‐year history of epilepsy. CT revealed an asymmetric ventricular system with prominence on the right side, enlarged external and internal cerebrospinal fluid spaces on the right due to volume loss, and moderate cortical atrophy on the right. (C) A 41‐year‐old male with generalized seizures since childhood (~1/months). CT revealed midline shift to the right with a hyperdense, nodular, lobulated‐appearing lesion in the left temporal to frontal region, consistent with an arteriovenous malformation following contrast administration. (D) A 38‐year‐old male with an 8‐year history of generalized seizures. CT showed hydrocephalus. (E) A 31‐year‐old male with generalized seizures and a normal CT scan. (F, G) Time in PSWE or maximal current source did not show differences between CT with or without lesions. (H, I) Decay of current source along voxels showing a slower decay in EEGs from nonlesional patients with greater brain volume (a more distributed source) in patients with nonlesional CT. (J) Receiver operating characteristic analysis showing sensitivity/specificity of brain volume with >1/e maximum current source for the identification of lesion on CT (area under the curve = .70, 95% confidence interval = .62–.87). L, left; ns, nonsignificant; R, right; sLORETA, standardized low‐resolution brain electromagnetic tomography.

4. DISCUSSION

Despite the heavy burden of epilepsy in sub‐Saharan Africa, there remains relatively limited published data on EEG features among African PWE. ¹⁶ In a recent systematic review, EEGs were reported for only 6.6% of 38 035 PWE in LMICs, likely due to the paucity of neurophysiology services and trained epileptologists, especially in rural regions. ¹ , ¹⁷ , ¹⁸ Previous EEG analyses using data obtained from high‐income settings have shown that PSWEs are more common in PWE compared to healthy controls. ⁷ Although it is important to emphasize that PSWE analysis does not replace manual EEG interpretation by an epileptologist, or the identification and localization of epileptiform features, we highlight the potential of quantitative, automated EEG analysis—including PSWE detection—as a biomarker for epilepsy. This approach is particularly suited for use in resource‐limited settings. Such automated methods may complement cloud‐based EEG interpretation, which, although available in some places, can be challenging to implement consistently in certain scenarios.

In the present study, PSWEs were detected in all PWE from a cohort in rural Zambia, with their relative summated duration (time in PSWE) associated with younger age (<9 years). Although this may reflect a tendency toward “slower” EEGs in early childhood, it is noteworthy that a similar correlation was found between PSWEs and the age at disease onset irrespective of the individual's current age (Figure 3B). These correlations suggest that epilepsy onset at a younger age results in a higher PSWE burden and distinct characteristics compared to epilepsy acquired later in life. Alternatively, because most patients in our cohort experienced seizures without treatment for many years, PSWEs may reflect alterations in cortical networks that are more pronounced due to a higher accumulated number of seizures. This explanation is supported by (1) our observation that greater PSWE duration was found in patients with drug‐resistant epilepsy during presurgical assessment, compared with seizure‐free patients (Bonn; Figure 3D); (2) imaging studies showing structural abnormalities on magnetic resonance imaging (MRI) scans of patients with chronic epilepsy ¹⁹ ; and (3) functional resting‐state MRI studies demonstrating that high seizure frequency is associated with altered brain connectivity. ²⁰

Notably, no differences were observed between two consecutive EEGs conducted on the same patient, nor between ASM‐treated drug‐resistant patients (Bonn) and ASM‐untreated patients (Zambia), supporting the hypothesis that PSWEs reflect disease severity or disease‐related cortical dysfunction rather than a nonspecific effect of ASMs.

Whether the distribution of PSWEs across scalp electrodes is related to the epileptogenic or ictogenic zone, or to regions with cortical dysfunction not necessarily linked to seizures, is not yet known. Importantly, although PSWEs were more prominent in PWE with abnormal EEGs (compared with “normal” EEGs), they were detected in all PWE. In contrast, on only 38% of the EEGs was a slowing visibly noticeable by the reading epileptologist. These findings confirm that the slowing identified by the algorithm (median frequency of 1–6 Hz for >5 s) can be subtle and visibly undetected by trained interpreters. Future studies should evaluate the spatial distribution of PSWEs in patients where the seizure onset zone is well defined (e.g., using intracranial EEG recordings). Regardless, our results demonstrate the potential for integrating automated PSWE analysis into routine EEG assessments, offering a scalable and cost‐effective solution for improving epilepsy diagnosis in resource‐limited settings.

The implementation of LORETA source localization on segmented PSWEs is proposed, for the first time, as a promising approach for the prediction and localization of lesional epilepsy. Several linear distributed inverse solutions for EEG source localization (e.g., exact LORETA, standardized LORETA, or local autoregressive average) have been used previously for localizing the epileptogenic zone with varying efficacy. ²¹ , ²² , ²³ Most of these studies have focused on localizing interictal epileptiform activity in patients undergoing presurgical investigation. Here, we present an alternative (or complimentary) approach for localizing focal PSWEs as a screening tool for prediction of lesional epilepsy. Source distribution of PSWEs suggested a focal origin (i.e., smaller brain volume) in patients with lesions identified on CT compared to a more distributed source in nonlesional cases (Figure 4H–J). These results underscore the potential utility of combining PSWE detection and source localization for prioritizing patients for advanced brain imaging. By enabling better identification of patients requiring advanced imaging and targeted treatment, this approach has the potential to bridge significant diagnostic and treatment gaps in LMICs.

Despite differences in disease context, causes, and treatment access, the consistency in PSWE patterns across cohorts from Zambia and Germany reinforces their robustness as a diagnostic measure. It rules out significant unrelated effects of ASMs and suggests that PSWE burden may be related to disease severity, subclinical epileptiform activity, and the likelihood of higher seizure frequency. To what extent PSWE features might serve as a pharmacodynamic biomarker—dynamically changing in response to ASM therapy—or reflect irreversible disease‐related changes in brain activity remains to be tested.

In conclusion, this study advances our understanding of PSWEs as a quantitative tool for epilepsy diagnosis and management. In the context of epilepsy diagnosis in LMICs, EEG can be more than a diagnostic tool; in the absence of a neurologist, semiautomatic quantitative analysis of EEG tests may help stratify patients for a faster consult or more advanced imaging (e.g., CT). Such analysis may also provide insight into the underlying causes of the disease (e.g., structural abnormalities) and its severity, offering a potential avenue for improving care in resource‐constrained settings. Future studies should focus on validating these findings in larger cohorts and refining automated PSWE analysis to maximize its clinical impact.

5. LIMITATIONS

Several limitations should be acknowledged. First, the study cohort was primarily from two clinics in the Kakumbi Chiefdom in the Eastern Province of Zambia, which may limit the generalizability of findings to other LMICs or urban settings where environmental and genetic factors might differ. Second, the reliance on a specific preprocessing algorithm may introduce method‐specific biases that affect PSWE detectability and interpretation. Specifically, the use of ICLabel to automatically reject components that were likely to be physiological artifacts may have affected our results. However, in the setting of the present study, it was important for us to limit human intervention and maximize automatic analysis that does not require expert input. Although we previously published PSWE analysis from a cohort of healthy controls, the lack of recordings from healthy individuals from the same demographic limits the ability to fully contextualize the observed PSWE characteristics. Although the comparison with the Bonn cohort led us to conclude that the effect of ASMs on PSWE features is nonexistent or negligible, future studies with a larger cohort, with patients treated with different ASMs, will be important to conduct. The relatively small sample size for advanced imaging (CT scans in only 42 participants) further restricts robust conclusions regarding the relationship between PSWE localization and structural abnormalities. Implementing this approach in patients with invasive recording should validate how PSWE source localization is related to the ictal onset zone. Future research should address these limitations through larger, multicenter, and longitudinal studies that will allow a better understanding of PSWE dynamics in relation to disease progression and response to treatment.

AUTHOR CONTRIBUTIONS

Andrew Malunga, Alon Friedman, and Stanley Zimba conceptualized and supervised the study. Andrew Malunga, Alon Friedman, Stanley Zimba, Mutale Masamu, Merav Shamir, and George Mshanga coordinated data collection in Zambia. Sina Lash, Alaa Abu‐Ahmad, Laith Alhadeed, Hamza Imtiaz, Anna Minarik, and Yonatan Serlin performed EEG data preprocessing and analysis. Alon Friedman, Laith Alhadeed, Felix Benninger, Gal Ben‐Arie, Nicholas Fearns, Stefan Kunst, Attila Racz, Theodor Ruber, and Benjamin Whatley contributed to data interpretation and methodological development. Felix Benninger, Gal Ben‐Arie, Yonatan Serlin, Oliver Neal, Benjamin Whatley, Theodor Ruber, Tamara Welte, and Stanley Zimba provided clinical input and critical revisions. Sina Lash, Andrew Malunga, Alaa Abu‐Ahmad, Alon Friedman, and Laith Alhadeed wrote the original draft. Alon Friedman and Stanley Zimba supervised the project and contributed to funding acquisition and final manuscript editing. All authors reviewed and approved the final version of the manuscript.

FUNDING INFORMATION

This work has been supported by the Dennis Chair in Epilepsy Research at Dalhousie University.

CONFLICT OF INTEREST STATEMENT

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

Table S1.

Malunga A, Lash S, Abu‐Ahmad A, Alhadeed L, Benninger F, Ben‐Arie G, et al. Paroxysmal slow wave events as a diagnostic biomarker for epilepsy: Lessons from rural Zambia. Epilepsia. 2025;66:4869–4880. 10.1111/epi.18598

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.