This study was conducted at a Level 3 Epilepsy Center within Children’s Hospital of Chongqing Medical University, China. We initially screened 135 children exhibiting SWI > 50% during nonrapid eye movement sleep stage (NREM) II sleep at our Liangjiang Campus. Based on ILAE SeLECTS criteria (clinical/MRI findings) (Specchio et al., 2022), 77 children were enrolled. All patients had normal cognition before SWAS onset on EEG. Upon SWAS detection, participants immediately underwent the Infant-Junior Middle School Student’s Social Life Ability Scale and Wechsler Intelligence Test. These assessments were repeated over the subsequent 2 years. Children displaying cognitive regression/stagnation—attributed to intense sleep-related epileptic activity—were classified as EE-SWAS (n = 33), while those with sustained normal cognition comprised the SeLECTS with SWAS group (n = 36). Eight patients were excluded due to loss to follow-up. For the modeling cohort, we collected: demographic data (age, sex), seizure onset age, family history of epilepsy/febrile convulsions, antiseizure medication use, seizure frequency at first SWAS detection, age at SWAS onset, and initial SWAS EEG parameters. A flow chart of this process is shown in Figure 1.

Using the model-identified risk factors, we established an external validation cohort (n = 60) following identical inclusion/exclusion criteria, comprising 30 EE-SWAS and 30 SeLECTS with SWAS patients from our Yuzhong Campus. In this cohort, only the age of first seizure, the high-amplitude SW with secondary generalization, and the long spike-and-wave clusters were recorded. See Figure 2 for cohort selection workflow.

This study was approved by the Ethics Committee of Children’s Hospital of Chongqing Medical University. Written informed consent was obtained from all participants and their guardians.

EEGs were recorded using an international standard 10–20 system with a sampling frequency of 1,000 Hz (EEG-1200C, Nihon Kohden). The participants were deprived of sleep the night before the EEG examination. All patients were monitored by video-EEG for at least 4 h, and the EEGs during the waking period and stage NREM I/II/III sleep were recorded. EEG analysis was performed using the average or bipolar reference montage.

We analyzed eight EEG parameters in the cohort: (1) interictal centrotemporal discharge laterality (left/right/bilateral), (2) sleep SWI (spike-and-wave duration in first 10 min NREM II/600 s × 100%) (Uliel-Sibony and Kramer, 2015; Ji et al., 2023), (3) wake discharge frequency (counts/5 min), (4) non-dipole spikes (frontotemporal dipole incidence <80%) (Gregory and Wong, 1992), (5) multifocal discharges (≥2 non-Rolandic foci) (Massa et al., 2001), (6) intermittent focal slow waves, (7) high-amplitude SW with secondary generalization (3 Hz absence-like, 1–5 s) (Massa et al., 2001; Nicolai et al., 2007; Aeby et al., 2021) (Figure 3), and (8) the long spike-and-wave clusters (≥6 s) (Massa et al., 2001; Nicolai et al., 2007; van Klink et al., 2016; Figure 4), with only parameters 7–8 assessed in the validation cohort, all independently reviewed by two board-certified epileptologists (Q. H., P. Y.) with >10 years’ experience.

Statistical analyses were performed using SPSS 26.0 (IBM Corp.). Group comparisons (EE-SWAS vs. SeLECTS with SWAS) employed χ² tests, Fisher’s exact tests, t-tests (normal distribution), and Mann–Whitney U tests (non-normal distribution). Multivariate logistic regression identified risk factors for EE-SWAS progression, preceded by multicollinearity assessment (tolerance >0.1, VIF < 10, condition index <30). Significant predictors (p < 0.05) were incorporated into a nomogram, with model performance evaluated via ROC curve analysis (AUC, sensitivity, specificity). External validation was subsequently conducted on an independent cohort.

A total of 69 children were included in the model cohort. Notably, the EE-SWAS group exhibited significantly earlier seizure onset (5.60 ± 1.67 years) compared to the SeLECTS with SWAS group (7.39 ± 1.71 years; p < 0.001). Febrile convulsion history (p = 0.024) and epilepsy family history (p = 0.008) also showed statistically significant intergroup differences. For comprehensive demographic and clinical characteristics, refer to Table 1.

EE-SWAS group showed significantly higher prevalence of high-amplitude SW with secondary generalization (60.61% vs. 2.78%, p < 0.001) and prolonged spike-and-wave clusters (66.67% vs. 5.56%, p < 0.001), along with greater median SWI (0.78 vs. 0.62) and bilateral interictal discharges (75.76% vs. 38.89%) compared to SeLECTS+SWAS group, with additional significant differences in nondipole spikes and discharge frequency (all p < 0.05) but not in multifocal discharges or slow waves, while pre-SWAS ictal EEG frequency was also higher in EE-SWAS (p = 0.014), all based on first SWAS detection EEG data (see Table 2).

Multivariate logistic analysis of the clinical and EEG data revealed that long spike-and-wave clusters, high-amplitude SW with secondary generalization, and young age of first seizure were risk factors for the progression of SeLECTS with SWAS into EE-SWAS. Detailed data are shown in Table 3.

Individual risk factor scores were derived from the nomogram, with the cumulative score predicting the probability of EE-SWAS progression in SeLECTS patients with SWAS (Figure 5).

The predictive model demonstrated excellent discriminative ability, with a C-index of 0.932 (95% CI: 0.872–0.992, p < 0.001) in the derivation cohort and 0.934 (95% CI: 0.868–1.000, p < 0.001) in the external validation cohort using three key predictors (younger seizure onset age, the long spike-and-wave clusters, and high-amplitude SW with secondary generalization), confirming robust performance in stratifying progression risk from SeLECTS with SWAS to EE-SWAS (Figures 6A,B).

Prior studies identified various risk factors for EE-SWAS progression: early seizure onset and increased frequency (Bebek et al., 2015; Desprairies et al., 2018), dysarthria/somatosensory auras (Porat Rein et al., 2021), multiple seizure types (Caraballo et al., 2019), and new seizure emergence (Desprairies et al., 2018). While these studies reported diverse factors, we confirmed younger seizure onset age as paramount-consistent with Fejerman/Posar (Fejerman, 2009; Posar and Visconti, 2024) and You et al.’s finding that SeLECTS onset <3 years predicts severe progression (You et al., 2006). Our data showed significantly earlier onset in EE-SWAS (5.60 ± 1.67 years) versus SeLECTS+SWAS (7.39 ± 1.71 years). The mechanism linking early seizure onset to atypical neurodevelopment (particularly cognitive regression/stagnation) remains unclear. Neuroimaging evidence suggests subtle structural remodeling occurs in both epileptogenic zones and distal regions of SeLECTS brains (Garcia-Ramos et al., 2015; Pardoe et al., 2013), with progressive changes observed in cognitively impaired patients (Garcia-Ramos et al., 2015; Pardoe et al., 2013; Ciumas et al., 2017). This implies epileptogenesis may disrupt normal developmental trajectories through cumulative interference with neuronal network maturation (Maltoni et al., 2016; Bourel-Ponchel et al., 2019). Earlier seizure onset likely extends this disruptive exposure window, amplifying neurodevelopmental risks.

While age of onset, febrile seizure history, epilepsy family history, and seizure frequency showed univariate significance, only age retained predictive value in multivariate analysis. The familial patterns suggest genetic predisposition–GRIN2A mutations (affecting synaptic GluN2A) (Gong et al., 2021) and CNKSR2 variants (Higa et al., 2021) are established EE-SWAS contributors, with additional genes (KCNA2, SLC9A6, HIVEP2, RARS2) implicated in developmental/epileptic encephalopathies with SWAS (Gong et al., 2021; Gong et al., 2020). Limited genetic data in our cohort (3 exome-sequenced cases: 1 negative, 1 SCN1A/SCN8A heterozygous, 1 PTEN heterozygous) preclude definitive conclusions.

Beyond seizure onset age, SeLECTS interictal EEGs demonstrate atypical patterns—multifocal discharges, high-amplitude SW with secondary generalization, focal slow waves, and prolonged spike-and-wave clusters (Massa et al., 2001; van Klink et al., 2016). Crucially, these patterns (not discharge frequency) better predict neurocognitive outcomes (Bourel-Ponchel et al., 2019; Metz-Lutz and Filippini, 2006), consistent with Gencpinar et al.’s finding that SWI (>85% vs. 50–85%) lacked clinical correlation (Gencpinar et al., 2016). Our data confirm that high-amplitude SW with secondary generalization and prolonged clusters—not SWI—independently predict EE-SWAS progression.

First described in 1976 (Bernardina and Beghini, 1976), high-amplitude SW with secondary generalization which also described as 3 Hz generalized spike-and-waves (lasting 1–5 s during NREM I/II sleep; Gelisse et al., 1999) and prolonged spike-and-wave clusters (≥6 s across sleep stages; Massa et al., 2001; Berroya et al., 2005) are hallmark EEG patterns in SeLECTS/EE-SWAS. While their cognitive impairment mechanism remains unclear, synaptic homeostasis disruption during sleep is hypothesized—evidenced by absent sleep slow-wave activity (SWA) changes in active EE-SWAS, with post-remission SWA normalization (Rubboli et al., 2019; Rubboli et al., 2023). This suggests spike-induced interference with synaptic pruning, potentially explaining EE-SWAS neuropsychological deficits and guiding future research.

While the mechanisms linking our three identified risk factors (early seizure onset, high-amplitude SW with secondary generalization, and prolonged spike-and-wave clusters) to cognitive regression or stagnation remain unclear, this first externally validated prediction model for EE-SWAS progression in SeLECTS patients provides clinically actionable insights. Statistically robust calculations confirm these factors enable early risk stratification, allowing targeted intensive monitoring and preemptive therapy for high-risk cases to mitigate neurocognitive deterioration (van den Munckhof et al., 2015; van den Munckhof et al., 2018).

Managing cognitive impairment in EE-SWAS remains challenging, with no established optimal treatment (Sánchez Fernández et al., 2014). While conventional antiseizure drugs show limited efficacy [49% improvement rate (van den Munckhof et al., 2020)], benzodiazepines (68%) and glucocorticoids (81%) demonstrate better outcomes. However, their use lacks standardized protocols (Sánchez Fernández et al., 2012; Veggiotti et al., 2012), and side effects (sedation, metabolic disturbances, infection risks) warrant caution. Our prediction model addresses this dilemma: high-risk cases may justify aggressive therapy, whereas low-risk patients—despite evident SWAS—should avoid unnecessary steroid/benzodiazepine exposure, as supported by Bebek/Posar’s findings (Bebek et al., 2015; Posar and Visconti, 2024). Crucially, treatment goals should prioritize cognitive impact over purely achieving EEG normalization.

This single-center cohort study has inherent limitations, including potential selection bias and restricted sample size affecting model calibration. While multicenter validation would enhance generalizability, our model establishes foundational predictors for cognitive regression or stagnation in SeLECTS with SWAS. These factors warrant targeted investigation through hypothesis-driven studies and longitudinal validation in larger cohorts.