Circumstantial,but Compelling: Multimodal Phase 1 Data and the Seizure Onset Zone

Abstract

Predictive Power of Phase 1 Studies for the Identification of Seizure Onset Zone in Phase 2 in a Pediatric Epilepsy Cohort

Katlowitz K, Allam A, Laxpati N, Lee S, McGinnis JP, Katyayan A, Ali I, Houck KM, Nayak A, Sen S, Diaz-Medina G, Mohanty D, Clarke D, Coorg R, Seto ES, Riviello JJ, Anderson AE, Weiner HL, Curry DJ. Epilepsia. 2025 Dec;66(12):4667-4679. doi: 10.1111/epi.18596. Epub 2025 Aug 18. PMID: 40824682

Objective: Surgical outcomes in the management of drug-resistant epilepsy (DRE) rely heavily on proper identification of the seizure onset zone (SOZ). Stereo-electroencephalography (sEEG) can be used to localize SOZs but must be hypothesis driven. Proper utilization of phase 1, noninvasive studies can maximize sEEG planning. Methods: We performed a retrospective chart review of pediatric patients who underwent sEEG implantation for DRE at a single institution and then subsequently had treatment for an identified SOZ. Each sEEG lead was identified by phase 1, noninvasive data indicating possible SOZ localization. SOZ and patient outcomes were correlated with phase 1 study findings. Results: One hundred patients with a total of 1777 leads implanted over the span of 10 years were analyzed. A total of 242 SOZs were identified; 41.5% of patients were seizure-free at 1 year, and 75.4% had at least a 50% reduction in seizure frequency. Multivariate modeling showed that anatomical findings such as lesions (odds ratio [OR] = 1.6) and calcifications (OR = 2.5), as well as magnetoencephalography (OR = 1.5) and semiology (OR = 1.7), were the most predictive of SOZ. Predictive power varied with the underlying seizure etiology. Significance: These results highlight the importance of a multimodal approach to SOZ identification in the noninvasive evaluation phase. A deeper understanding of the potential of each individual preoperative testing modality can guide sEEG placement to minimize surgical risk while maximizing diagnostic yield.

Commentary

Localization of the seizure onset zone (SOZ) is the foundation of effective surgery for focal epilepsies. ¹ The phase 1 evaluation is a standard approach for noninvasive SOZ localization that includes analysis of semiology, long-term video EEG, epilepsy protocol, MR neuroimaging, and neuropsychological evaluation. ² When these techniques are insufficient, adjuncts such as nuclear imaging (FDG-PET and SPECT), magnetoencephalography (MEG), or fMRI are applied to gather additional evidence supporting seizure focus localization. Consequently, phase 1 evaluation is highly resource intensive, carrying significant cost in time and money for patients, caregivers, and the epilepsy team alike. When phase 1 results are inconclusive—which is often the case, particularly in MR-negative epilepsies—intracranial monitoring is required, adding further cost and resource utilization. ³ Importantly, despite this resource burden, epilepsy surgery for focal epilepsy is well-established as both clinically effective and cost-effective. ⁴

Clinical prediction models offer the ability to identify patients most likely to benefit from a given diagnostic test or treatment. Given the resource-intensive nature of presurgical epilepsy evaluation, there is significant interest in tools that predict which patients will benefit from phase 1 evaluation, proceed to phase 2 invasive monitoring, and ultimately achieve seizure freedom with surgery. Models that assign predictive value to individual components of the phase 1 evaluation are therefore of particular interest. In 2021, Astor-Rohracher et al ¹ described the 5-SENSE score, a clinical prediction model that estimates SOZ focality as measured by sEEG invasive monitoring based on five features: (1) focal lesion on neuroimaging, (2) bilateral independent spikes on scalp EEG, (3) localizing neuropsychological findings, (4) localizing semiology, and (5) regional ictal scalp EEG onset. This score demonstrated reasonable discriminative performance (AUC 0.83, specificity 76.3%, and sensitivity 83.3%) and represents a meaningful advance in identifying patients unlikely to have a focal SOZ on invasive monitoring. However, the 5-SENSE score was developed in a cohort of 128 patients across multiple centers, and questions remain about its generalizability—particularly in younger pediatric populations.

Recently, Katlowitz et al ⁵ investigated which features of the phase 1 evaluation best predict SOZ identification during invasive monitoring in a pediatric epilepsy cohort. This work analyzed 100 pediatric patients who underwent invasive monitoring over 10 years at a single institution. Multivariable modeling showed that the presence of a lesion (odds ratio [OR] = 1.6), calcification (OR = 2.5), MEG cluster (OR = 1.5), and localizing semiology (OR = 1.7) were the strongest predictors of SOZ localization. Similar to the 5-SENSE score, these results highlight the dominant role that MR-visible lesions—and perhaps calcification as a subclass of imaging finding—play in predicting SOZ identification by invasive monitoring. This is not surprising: lesions visible on MRI, such as cavernomas, tumors, and focal cortical dysplasia (FCD, particularly FCD2B), are frequently co-localized with the SOZ, lesionectomy is associated with good surgical outcomes, and many such patients can proceed directly to surgery when phase 1 findings are concordant. ⁶ An equally important takeaway from Katlowitz et al is the value of multimodal phase 1 assessment. The similar odds ratios observed across semiology, lesion, and MEG findings suggest that no single modality is sufficient—rather, it is the convergence of evidence across modalities that builds a compelling hypothesis for invasive monitoring. It is well accepted that invasive monitoring is most effective when guided by hypotheses derived from noninvasive phase 1 data—the anatomoelectroclinical hypothesis ⁷ —and both the Katlowitz and 5-SENSE models now provide quantitative support for this principle.

An important caveat to the Katlowitz et al study is its unique population, in which 59% of patients carried a diagnosis of TSC. Generalizing these findings to a broader pediatric focal epilepsy population is therefore difficult. Children with TSC characteristically have multiple brain lesions competing as potential SOZ candidates, which may have paradoxically biased the lesion OR downward relative to what one might expect in a cohort enriched with FCD or tumoral epilepsy, where a single dominant lesion more reliably identifies the SOZ. Conversely, the dominance of calcification as the strongest predictor (OR = 2.5) almost certainly reflects the high prevalence of calcified tubers in TSC, and may not be informative in other pediatric epilepsy populations. Taken together, the Katlowitz model is perhaps best understood as a well-executed, TSC-weighted analysis that offers valuable hypothesis-generating insights, but requires validation in a more etiologically diverse cohort before its conclusions can be broadly applied.

Building reliable prediction models for focal epilepsy surgery is inherently difficult. Phase 1 evaluation generates high-dimensional data—multiple modalities, not all obtained in every patient, and many intercorrelated with one another and with underlying etiology—within a relatively small patient cohort. The correlation between lesion and calcification on MRI is one straightforward example, but etiology-driven correlations are subtler and more consequential: in the mesial temporal lobe epilepsy subgroup of the Katlowitz cohort, semiology appeared uninformative on univariate analysis but emerged as a strong independent predictor once anatomy and PET were included in the multivariable model, illustrating how correlated predictors can mask one another's true contributions. More broadly, drug-resistant focal epilepsy is a highly heterogeneous condition with etiology-specific drivers of SOZ localization, treatment selection, and surgical outcome—a reality that no single-institution retrospective cohort can fully capture. The population most affected by these limitations is arguably those with MR-negative focal epilepsy, who represent fewer than 10% of the Katlowitz cohort yet face the greatest diagnostic uncertainty and the greatest unmet need for better phase 1 prediction tools. ⁸ Addressing this gap will require prospective, multi-institution studies with sufficient etiologic diversity and sample size.

In conclusion, Katlowitz et al have taken an important step toward understanding the predictive value of phase 1 evaluation for SOZ localization in pediatric invasive monitoring. Like the 5-SENSE score, their findings reinforce that neuroimaging plays a central and dominant role, and that no single modality is sufficient—it is the convergence of evidence across semiology, neuroimaging, and electrophysiology that builds a compelling case for the SOZ. However, both models were developed in cohorts with limited etiologic diversity, and their generalizability to the full spectrum of pediatric focal epilepsy—particularly MR-negative drug-resistant epilepsy—remains to be established. As the field moves toward more precise, individualized presurgical evaluation, prospective multi-institution studies with etiologically diverse pediatric cohorts will be essential to building prediction tools that can guide clinical decision-making for all children with drug-resistant focal epilepsy.