Seizure prediction is of paramount clinical importance for optimizing interventions and improving patient quality of life. Recent advancements in ECG-based seizure prediction have garnered attention due to their potential for portability and ease of data collection. However, existing methods primarily focus either on heart rate variability (HRV) to capture ECG trends or on convolutional neural networks (CNNs) for local feature extraction, but often fail to effectively combine both global and local information, limiting the prediction performance. To address this challenge, we propose the Progressive Feature Fusion Mamba (PFFMamba) network for ECG-based seizure prediction. PFFMamba enhances prediction accuracy in two key ways. First, it utilizes Mamba’s selective scanning mechanism to focus on important historical data while discarding irrelevant information, enabling the model to effectively capture subtle local variations and the long-term dependencies among them. Second, we introduce a progressive feature fusion structure to overcome Mamba’s limitation in combining parallel features and align with the hierarchical nature of features at different stages. This structure integrates frequency channels in the shallow layers, feature channels in the intermediate layers, and HRV representations in the deep layers, ensuring robust feature interaction throughout the network. By combining these strategies, PFFMamba extracts features that embody both long-term dependencies and multi-perspective representations. We evaluate our approach through seizure prediction experiments and demonstrate that PFFMamba achieves a sensitivity of 96.7%, a false prediction rate of 0.548/h, an AUC of 0.775 on the USTC Epileptic ECG database and a sensitivity of 95.9%, a false prediction rate of 0.641/h, an AUC of 0.724 on the Siena database, surpassing current state-of-the-art methods.

Objective.Deep learning has shown significant potential in electroencephalogram (EEG)-based seizure prediction. However, translating these advances into practical applications faces critical challenges: the prohibitive time costs of reviewing accumulated historical data during frequent model updates, the need for rapid adaptation to diverse model architectures across hardware platforms, and the substantial burden of massive data storage. To this end, we introduce the first dataset distillation study tailored for seizure prediction, which condenses the original dataset into a compact, information-dense synthetic dataset. This tiny-scale proxy allows models to match the performance of those trained on the full-scale dataset.Approach.Methodologically, we formulate the distillation as a min-max optimization task, introducing an EEG-aware alignment objective that jointly preserves latent feature distributions, temporal dynamics in the time-frequency domain, and inter-channel correlation structures. Through signal-compatible amplitude and phase decomposition, we optimize the frequency parameter to maximize the CF discrepancy via a sampling network, while simultaneously tuning the synthetic data to minimize this gap. Additionally, we integrate hard labels to refine soft labels to enhance the generalization of synthetic data.Main results.Extensive validation across 37 public and 35 clinical subjects demonstrate the method's robustness across diverse scenarios and model architectures. With merely 1% of the original data volume, our approach retains roughly 95% of the original performance while reducing training duration by approximately 98%.Significance.This method addresses clinical translation bottlenecks by distilling massive EEG data into compact synthetic sets, offering a promising solution for efficient model maintenance, flexible model switching, and optimized storage.

BACKGROUND:Severe cutaneous adverse reactions (SCARs) are rare but potentially fatal immune-mediated toxicities associated with antiseizure medications (ASMs). Phenotype-resolved post-marketing safety profiles for newer-generation ASMs have not been comprehensively characterized across pharmacovigilance systems. METHODS:A cross-database disproportionality analysis was conducted using FAERS and JADER (January 2004-September 2025). Twenty-two newer-generation ASMs coded as primary-suspect drugs were included; outcomes were the four SCAR phenotypes (Stevens-Johnson syndrome, SJS; toxic epidermal necrolysis, TEN; drug reaction with eosinophilia and systemic symptoms, DRESS; acute generalized exanthematous pustulosis, AGEP) defined using MedDRA preferred terms. Signals were evaluated using reporting odds ratios (RORs) and Bayesian information component metrics. Time-to-onset (TTO) was characterized by Weibull modeling with confidence-interval-based failure-pattern classification and Kaplan-Meier analysis. Pre-specified sensitivity analyses excluded reports with concomitant valproic acid or other SCAR-inducing co-medications, applied multivariate logistic regression, and restricted FAERS reports to healthcare professionals. RESULTS:In total, 10,073 SCAR reports were included (SJS, 3776; TEN, 1762; DRESS, 4298; AGEP, 237). Lamotrigine and zonisamide showed the strongest, most reproducible associations across SJS/TEN/DRESS (FAERS RORs for lamotrigine, 35.60/22.35/27.30; zonisamide, 28.08/25.05/40.78), with concordant directions in JADER. Sensitivity analyses attenuated but did not abolish the principal lamotrigine, zonisamide, levetiracetam, and eslicarbazepine signals, whereas the gabapentin-TEN signal in JADER lost statistical significance after exclusion of high-risk co-medications. Median TTO was 22 days overall; SJS/TEN occurred earlier, while DRESS showed delayed and more dispersed onset (median, 28 days), with eslicarbazepine-DRESS uniquely exhibiting a wear-out pattern. TEN carried the highest reported fatality proportion (FAERS, 17.7%; JADER, 17.5%). Key signals remained robust under healthcare-professional restriction, with eslicarbazepine-DRESS further enhanced. CONCLUSIONS:Across FAERS and JADER, reporting associations between newer-generation ASMs and SCARs were highly concentrated in a limited set of agents and exhibited phenotype-specific latency patterns. After accounting for polytherapy confounding, levetiracetam and eslicarbazepine emerged as the most consistent under-recognized signals, warranting heightened vigilance during initiation-particularly when co-administered with aromatic ASMs. Drug-phenotype combinations with both high disproportionality and a high reported fatality proportion (notably lamotrigine-TEN and zonisamide-TEN) warrant intensified early monitoring for SJS/TEN and sustained vigilance into maintenance therapy for DRESS, although population-level absolute risk cannot be inferred from spontaneous-report data.

Seizure prediction through scalp electroencephalogram (EEG) holds considerable practical potential. The primary challenge faced by existing algorithms lies in the individual heterogeneity, which hinders the generalizability of models to new patients. Additionally, inconsistencies in channel settings across various epilepsy centers further limit the applicability of models to diverse datasets. To address these challenges, we incorporate large language models (LLMs) into EEG analysis and propose a novel seizure prediction method based on LLMs (SPLLM), significantly enhancing both model generalizability and applicability. Specifically, this approach reprograms LLMs by transforming EEG signals into textual representations compatible with LLMs via a single-channel pre-training strategy. The method integrates cross-domain knowledge from both text and EEG data through a cross-attention mechanism, utilizing autoregressive pretrained LLMs to capture the temporal dependencies inherent in EEG signals. Moreover, the cross-domain generalization ability of LLMs alleviates patient heterogeneity, while the single-channel pre-training strategy enables the model to adapt to diverse channel settings. On two public datasets and one private dataset, SPLLM increases the average AUC by 8.2%, and the average balanced accuracy by 8.4% compared to existing methods. Experimental results demonstrate that the proposed method not only enhances cross-patient prediction accuracy but also adapts to data from different datasets, offering a scalable solution for the clinical application of seizure prediction.

Federated learning (FL) has advanced in predicting epileptic seizures using electroencephalogram (EEG) data. However, the inherent variability among patients poses challenges for existing prediction methods, which struggle to maintain robust performance in heterogeneous client environments. To address this issue, we propose a novel framework called personalized federated learning with mixture of experts for EEG-based seizure prediction (PFL-MoE-ESP). Under the personalized federated learning paradigm, model parameters are decoupled into globally shared components (the gating network and shared expert) and locally private components (the routed experts). We split the parameters and adopt a personalized server aggregation strategy. This approach supports collaborative learning of general knowledge from multiple patients’ data and enhances model personalization for individual patients. We introduce a dynamic gating mixture of experts (DG-MoE) model as the core network of PFL-MoE-ESP. Unlike traditional mixture of experts (MoE) with a static gating mechanism, our model incorporates a dynamic gating mechanism driven by multi-head self-attention and a heterogeneous pool of expert networks. Through this adaptive and fine-grained routing mechanism, our model effectively captures the distinct spatiotemporal dynamics unique to each patient, thereby not only enhancing the model’s robustness to individual adaptability but also boosting prediction accuracy. Extensive evaluations on two public datasets, CHB-MIT and Kaggle, were conducted using a last five-fold cross-validation strategy to ensure reliable results. The results demonstrate that our PFL-MoE-ESP framework achieves superior performance, significantly outperforming existing baseline methods across key metrics including area under the curve (AUC), sensitivity, and false-positive rate (FPR).

Background:Stroke remains the leading cause of long-term disability worldwide. Approximately 60% of individuals with chronic ischemic stroke experience persistent upper limb impairment that limits daily activities. The Repair Study aims to evaluate the safety and efficacy of vagus nerve stimulation (VNS) paired with rehabilitation in patients with chronic ischemic stroke in developing countries, including those with severe upper limb dysfunction, thereby generating evidence to support broader global application. Methods:It is a multicenter, triple-blinded, randomized controlled trial conducted across 13 centers in China. Up to 99 participants with upper limb motor impairment, 9 months to 10 years post-stroke, will be enrolled. All participants will undergo VNS implantation (Model G115R/G115, PINS Medical, Beijing, China) and be randomized 2:1 by a central randomization system to active stimulation (0.8 mA) or sham stimulation (0 mA) paired with standardized upper limb rehabilitation. The blinded phase includes 6 weeks of clinical therapy (three sessions/week, 90-120 min/session, ≥300 stimulation-movement repetitions) followed by 6 weeks of home-based therapy (30 min/day). Post-unblinding, the active VNS group continues home-based therapy, while the sham group receives 6 weeks of clinic-based therapy. Discussion:The primary outcome is the between-group difference in Fugl-Meyer Assessment for Upper Extremity scores at the end of 6 weeks of clinical therapy. Secondary outcomes include additional motor, functional, and quality-of-life measures. Safety will be assessed through adverse event monitoring. The Repair Study is a multicenter randomized controlled trial targeting chronic ischemic stroke populations in developing countries. It supplements the existing clinical evidence by enrolling patients with more servere motor dysfunction and being conducted in a developing country. Trial Registration:ClinicalTrials.gov: NCT06722677.

Intracranial electroencephalography (iEEG) plays a crucial role in the preoperative evaluation of drug-resistant epilepsy. However, visual inspection of iEEG recordings is time-consuming and labor-intensive, highlighting the need for automated analysis methods to assist clinicians. While deep-learning approaches have shown promise in iEEG classification, existing methods are constrained by single-network architectures, restricting the diversity of extracted information and failing to fully exploit complementary information from both temporal and spectral domains. To this end, we propose a dual-branch temporal-spectral fusion model (DTSF), which employs a dual-branch CNN-GRU architecture to extract temporal and spectral features within domains, and a transformer fusion module to explore the global relationships between domains. Specifically, the temporal and spectral branches extract features from the raw signal and its STFT representation, respectively, where CNN captures local patterns in the features and GRU models their long-range dependencies. Features from both branches are fused using a Transformer-based block with a self-attention mechanism, which dynamically adjusts the importance of temporal and spectral features based on their contextual relationships, capturing the global dependencies between both domains to generate a unified temporal-spectral representation for final classification. Furthermore, Mixup data augmentation is utilized in both branches, facilitating diverse feature learning and improving model generalization. We evaluate the proposed method on two publicly available datasets, FNUSA and MAYO. Experiments demonstrate that DTSF outperforms several state-of-the-art models across multiple evaluation metrics, including accuracy and Kappa coefficient. This study presents an effective method for leveraging temporal and spectral features in iEEG classification.

Accurate motor trajectory estimation from physiological signals is essential for developing advanced motor rehabilitation and bionic devices. Fusion of electroencephalography (EEG) and surface electromyography (sEMG) leverages complementary information, yet existing methods primarily target discrete intent classification. Current studies often utilize simultaneously collected EEG and sEMG, assuming temporal alignment between these signals and thereby overlooking the inherent latency between the two modalities. This oversight induces semantic misalignment and insufficient consistency representation, ultimately degrading performance in continuous motion trajectory decoding. To overcome these limitations, this paper proposes AtpFusion, an EEG-sEMG asynchronous time-frequency progressive fusion model for enhanced 3-dimensional (3D) hand trajectory decoding. Key contributions: 1) asynchronous time-frequency inputs, constructed using a physiologically-inspired long-short time window segmentation strategy for semantic alignment, comprising long-window frequency-domain EEG (amplitude/phase) and short-window time-domain sEMG signals; and 2) a progressive hierarchical fusion architecture with intra-modal and inter-modal branches, designed for effective hierarchical feature refinement and integration for regression. AtpFusion is evaluated on the public WAY-EEG-GAL dataset, performing, to our knowledge, the first EEG-sEMG-based continuous hand trajectory estimation on this benchmark. The proposed model yields state-of-the-art accuracy with a Pearson Correlation Coefficient (PCC) of 0.9278 and a Root Mean Square Error (RMSE) of 0.0916, significantly outperforming existing approaches. This work presents a novel asynchronous EEG-sEMG fusion framework, offering a high-performance solution for practical multimodal bionic interfaces.

Predictive coding theory proposes a hierarchical framework for auditory novelty processing, with multiple brain regions involved. However, how these regions are organized spatiotemporally within the auditory hierarchy remains unclear. To address this issue, we deployed a passive pure tone oddball paradigm in a group of patients with stereoelectroencephalography recording for epilepsy treatment. Event-related high frequency activity was elicited by novel auditory stimuli across multiple brain regions. Based on latency, these regions could be segregated into a two-level hierarchy, with the temporal, insular, and parietal regions at the lower level and the frontal and hippocampal regions at the higher level. The lower, but not the higher, level showed a significant response to standard stimuli. Granger causality analysis further indicated that bidirectional connectivity between levels was asymmetrically modulated. The bottom-up connections were constant, whereas top-down connections were selectively modulated by novel stimuli. These results are consistent with a mechanism in which the lower level transmits auditory signals to the higher level, where comparison is made and potential prediction updating is triggered upon novelty detection. Such prediction updating would be mediated by dynamically modulated top-down pathways. Together, our findings are compatible with a two-level spatiotemporal hierarchical model for auditory novelty processing.

Data augmentation (DA) can significantly strengthen the electroencephalogram (EEG)-based seizure prediction methods. However, existing DA approaches are just the linear transformations of original data and cannot explore the feature space to increase diversity effectively. Therefore, we propose a novel diffusion-based DA method called DiffEEG. DiffEEG can fully explore data distribution and generate samples with high diversity, offering extra information to classifiers. It involves two processes: the diffusion process and the denoised process. In the diffusion process, the model incrementally adds noise with different scales to EEG input and converts it into random noise. In this way, the representation of data can be learned. In the denoised process, the model utilizes learned knowledge to sample synthetic data from random noise input by gradually removing noise. The randomness of input noise and the precise representation enable the synthetic samples to possess diversity while ensuring the consistency of feature space. We compared DiffEEG with original, down-sampling, sliding windows and recombination methods, and integrated them into five representative classifiers. The experiments demonstrate the effectiveness and generality of our method. With the contribution of DiffEEG, the multiscale CNN achieves state-of-the-art performance, with an average sensitivity, FPR, AUC of 95.4%, 0.051/h, 0.932 on the CHB-MIT database and 93.6%, 0.121/h, 0.822 on the Kaggle database.

BACKGROUND:Hemorrhagic stroke has a high incidence, often leaving patients with significant complications such as limb mobility disorders after treatment. Traditional treatment methods for stroke patients mainly include limb function exercises and hyperbaric oxygen therapy, which have shown effective results. In recent years, there have been reports utilizing vagus nerve stimulation (VNS) to treat limb paralysis in ischemic stroke patients, achieving encouraging outcomes. However, there are rare related reports on hemorrhagic stroke. CASE PRESENTATION:This report presents a case of a patient who developed left upper limb hemiplegia and recurrent seizures after a hemorrhagic stroke. The patient showed a poor response to standard anti-epileptic treatment and was diagnosed with stroke-related epilepsy. To manage the recurrent seizures, VNS was performed. After the device was activated, the patient reported a significant reduction in abnormal muscle tone and increased mobility impairment in the affected upper limb. Parameters were adjusted, and intermittent stroke electrical stimulation was combined with upper limb rehabilitation exercises. After three months of active treatment, the patient's seizures were well controlled, and there was significant improvement in upper limb function. CONCLUSIONS:VNS has potential in the rehabilitative treatment of stroke patients with upper limb dysfunction. It is hoped that more patients will benefit from this advanced treatment method in the future, regaining their health and vitality. Additionally, future research needs to further explore the mechanisms and methods of brain remodeling to provide theoretical support and more effective treatment options for stroke patient rehabilitation.

Electroencephalography (EEG) signals are often contaminated with various physiological artifacts, seriously affecting the quality of subsequent analysis. Therefore, removing artifacts is an essential step in practice. As of now, deep learning-based EEG denoising methods have exhibited unique advantages over traditional methods. However, they still suffer from the following limitations. The existing structure designs have not fully taken into account the temporal characteristics of artifacts. Meanwhile, the existing training strategies usually ignore the holistic consistency between denoised EEG signals and authentic clean ones. To address these issues, we propose a GAN guided parallel CNN and transformer network, named GCTNet. The generator contains parallel CNN blocks and transformer blocks to respectively capture local and global temporal dependencies. Then, a discriminator is employed to detect and correct the holistic inconsistencies between clean and denoised EEG signals. We evaluate the proposed network on both semi-simulated and real data. Extensive experimental results demonstrate that GCTNet significantly outperforms state-of-the-art networks in various artifact removal tasks, as evidenced by its superior objective evaluation metrics. For example, in the task of removing electromyography artifacts, GCTNet achieves 11.15% reduction in RRMSE and 9.81% improvement in SNR over other methods, highlighting the potential of the proposed method as a promising solution for EEG signals in practical applications.

The human hippocampus has been claimed to play an important role in long-term memory, or episodic memory. It has long been argued whether the hippocampus does contribute to short-term memory. Here, we demonstrate human hippocampus has been involved in auditory short-term memory (ASTM) trace formation. We used a classic oddball paradigm and intracranial recordings across various human brain areas. High frequency activities evoked by deviant stimuli in subjects indicate the generation of ASTM trace. Auditory response latencies recorded from the hippocampus, insula, temporal lobe, parietal lobe, and frontal lobe showed the early processing of ASTM trace at a pre-attention stage. Moreover, Granger causality analyses showed ASTM trace is processed in hierarchical cortical areas and interactions among this memory process have been clarified. Specifically, bottom-up auditory signals processed by the insula, and auditory regions of the temporal lobe and the frontal lobe at an early stage, then auditory top-down information transmitted from the frontal lobe to the hippocampus, the inferior temporal lobe, as well as insula at a relatively late pre-attention stage of the auditory information processing. These results provide evidence that the human hippocampus contributes to the short-term memory of auditory perception at a pre-attention stage and suggests a challenge to entrenched beliefs in the classification/definition of memory systems. ### Competing Interest Statement The authors have declared no competing interest. University of Science and Technology of China, https://ror.org/04c4dkn09, WK3460000008 National Natural Science Foundation of China, GG2070000587

Automated classification techniques for intracranial electroencephalography (iEEG) using deep learning have gained increasing attention. Due to the privacy protection of patient data, source-free domain adaptation (SFDA) transfers the pre-trained source model rather than the source patient data to the unlabeled target patient domain to address the domain shift issue arising from individual differences. However, SFDA encounters performance limitations due to the absence of supervised information from target patients. To break through the performance bottleneck of SFDA in iEEG classification, we propose a novel active source-free domain adaptation (ASFDA) approach based on neighborhood uncertainty and diversity (NUD), which facilitates the target model adaptation by identifying a small yet informative subset of target patient data for expert annotation through active learning. NUD proposes three key strategies to select the most informative samples for annotation. Specifically, NUD employs neighborhood uncertainty estimation (NUE) to assess the uncertainty of a sample, identifying samples with high NUE values as informative samples. Subsequently, NUD utilizes neighborhood diversity preservation (NDP) to mitigate the selection of similar samples by NUE, which ensures the diversity of informative samples. Additionally, NUD employs multiple round annotation (MRA) to ensure the representativeness of informative samples, which aids in the model’s learning of the target sample distribution. Extensive experiments on two public iEEG datasets and representative models demonstrate that NUD achieves over 90% classification accuracy, outperforming several state-of-the-art SFDA and ASFDA methods. It is a promising privacy-preserving domain adaptation method for iEEG classification with robust adaptability and low annotation burden.

A brain-computer interface (BCI) that decodes speech directly from neural activity provides a rapid and natural means of communication for individuals with speech impairments or aphasia. Recent advances in deep learning have led to several studies demonstrating promising outcomes using electrocorticography (ECoG) placed on cortical surfaces. In contrast, stereo electroencephalography (SEEG) captures neural signals from multiple brain regions, including the cortex and subcortex. These signals, encompassing rich information from deeper brain structures, have significant potential to enhance the characterization of speech generation processes and improve decoding performance. However, effective SEEG-based decoding schemes remain limited. Existing deep learning methods often struggle with overfitting due to insufficient data. In addition, the reconstructed speech tends to be blurry and detail-deficient, indicating an urgent need for more refined SEEG modeling. To address these issues, a convolutional encoder-decoder with scale-recursive reconstructor (ConvED-SR) is proposed for SEEG speech decoding. ConvED-SR first extracts multiscale speech-related features from SEEG signal using a convolutional encoder-decoder architecture. This creates a compact and efficient latent feature space with reduced parameters, thus mitigating overfitting. Furthermore, these multiscale features, effectively characterizing the intricate relationships between deeper neural signals and speech, are used to generate a refined Mel-spectrogram by a scale-recursive reconstructor. The reconstructor initially models low-frequency information, gradually interacts with high-frequency information, and ultimately refines a coarse Mel-spectrogram into a detailed final one. Finally, a HiFi-GAN vocoder converts the spectrogram into speech. Comprehensive experimental results on the SingleWordProductionDutch-iBIDS dataset demonstrate that ConvED-SR achieves superior performance, providing a promising solution for SEEG-based speech decoding.

Although deep neural networks have shown promise in the patient-specific context closely related to training and testing data distributions, they face challenges when applied to real-world applications’ dynamic and diverse environments. The inherent individual differences and the dynamic nature of EEG signals can lead to substantial domain shifts, hindering the models’ ability to generalize from training data (source domain) to new, unseen patient data (target domain). To tackle this issue, we investigate the domain generalization (DG) setting, which aims to learn a model that can generalize to an arbitrary unknown target domain by learning from multiple source domains. In this paper, we propose a novel Fourier-based data augmentation method called UFA, which approaches DG from an uncertainty perspective. UFA works by nonlinearly perturbing the feature statistics of the frequency spectrum obtained by the Fourier transform, thus increasing the diversity of the data distribution in the source domain. In particular, UFA first decomposes the spatial features of the network modules into phase and amplitude spectrums using the Fourier transform (FT). It then randomly samples the feature statistics of frequency spectrums from a specific distribution and uses inverse normalization to generate new frequency spectrums. Finally, we use the inverse Fourier transform to synthesize new spatial feature variants, simulating potential domain shifts in the unknown target domain. Our UFA can be seamlessly integrated as a plug-and-play module within network architectures, positioned immediately after each submodule. The UFA operates in training and is inactive in testing. Extensive experiments on two publicly available benchmarks show that our method achieves significant performance improvements over existing Fourier-based data augmentation methods.

Background: To evaluate the efficacy and safety of the first cohort of people in China treated with a responsive neurostimulation system (EpilcureTM, GenLight MedTech, Hangzhou, China) for focal drug-resistant epilepsy in this study.Methods: This multicenter, before-and-after self-controlled study was conducted across 8 centers from March 2022 to June 2023, involving patients with drug-resistant epilepsy who were undergoing responsive neurostimulation (RNS). The study was based on an ongoing multi-center, single-blind, randomized controlled study. Efficacy was assessed through metrics including median seizure count, seizure frequency reduction (SFR), and response rate. Multivariable linear regression analysis was conducted to explore the relationships of basic clinical factors and intracranial electrophysiological characteristics with SFR. The postoperative quality of life, cognitive function, depression, and anxiety were evaluated as well.Results: The follow-up period for the 19 participants was 10.7 ± 3.4 months. Seizure counts decreased significantly 6 months after device activation, with median SFR of 48% at the 6th month (M6) and 58% at M12 (P <0.05). The average response rate after 13 months of treatment was 42%, with 21% (n = 4) of the participants achieving seizure freedom. Patients who have previously undergone resective surgery appear to achieve better therapeutic outcomes at M11, M12 and M13 (β <0,P <0.05). No statistically significant differences were observed in patients’ scores of quality of life, cognition, depression and anxiety following stimulation when compared to baseline measurements. No serious adverse events related to the devices were observed.Conclusions: The preliminary findings suggest that EpilcureTM exhibits promising therapeutic potential in reducing the frequency of epileptic seizures. However, to further validate its efficacy, larger-scale randomized controlled trials are required.Registration: Chinese Clinical Trial Registry (No. ChiCTR2200055247)

BACKGROUND AND OBJECTIVES:Resective surgery is an effective treatment for tuberous sclerosis complex (TSC)-related drug-resistant epilepsy (DRE). However, high-level evidence validating the effectiveness and safety of this technique is lacking. This study aimed to provide prospective multicenter data comparing resective surgery with medicine treatments. METHODS:This prospective cohort study included patients with TSC-related DRE who were suitable for resective surgery from 2019 to 2022 were identified from 21 centers in China. After collection, patients who underwent resective surgery were entered into the surgery group, whereas those who refused surgery and continued medicine treatment were included in the medicine group. Follow-ups for ≥2 years tracked seizure control, IQ, quality of life (QOL), antiseizure medications (ASMs), and adverse events. Propensity score matching (PSM) was performed to balance baseline demographics and clinical characteristics. Seizure-freedom probabilities were estimated using Kaplan-Meier analysis, and the mixed effects Cox proportional hazards model assessed the risk of seizure recurrence alongside other covariates. RESULTS:Each group comprised 100 patients with a total median age of 6.70 years (interquartile range, 3.73-12.80), and 35% were female patients. PSM-adjusted seizure-freedom rates in the surgery group were 77%, 67%, 66%, and 67% at 1-, 2-, 3-, and 4-year follow-ups, respectively, compared with 13%, 7%, 6%, and 4% in the medicine group. The surgery group had a higher probability of achieving seizure freedom. Resective surgery was identified as the strongest independent predictor reducing post-collection seizure recurrence (hazard ratio, 0.10; 95% CI, 0.05-0.17; p < 0.0001). Significant improvements in IQ and QOL were found at the 2-year follow-up, along with reductions in ASMs across all follow-ups after surgery (p < 0.01). Outstanding tubers were the independent factor for postoperative low seizure recurrence. Serious complications were not found, and no significant differences in adverse events were observed. DISCUSSION:Resective surgery demonstrates significantly higher effectiveness than medicine in seizure control and improvements of IQ and QOL in patients with TSC-related DRE, achieving a 1-year seizure-freedom rate of 77%. Furthermore, the safety of resective surgery is comparable with that of medicine. Outstanding tubers on MRI is an independent predictor of postoperative seizure freedom. Selection bias and possible confounding by site are limitations of this study. TRIAL REGISTRATION INFORMATION:Registered with ClinicalTrials.gov as TRE-RES. Registration number: NCT04198181. Date of the registration submitted to the registry: November 10, 2019. Date of first patient enrollment: December 12, 2019. CLASSIFICATION OF EVIDENCE:This study provides Class III evidence that resective surgery provides better seizure control, and improved IQ and quality of life, than medication alone in patients with TSC.