Early Prediction of Drug-resistance from the Child’s Drug-naïve EEG: a New EEG Network Analysis and Prediction Tool
Abstract number :
1.211
Submission category :
2. Translational Research / 2C. Biomarkers
Year :
2024
Submission ID :
1268
Source :
www.aesnet.org
Presentation date :
12/7/2024 12:00:00 AM
Published date :
Authors :
Presenting Author: Vitor Pimenta de Figueiredo, BS – Boston Children's Hospital
Navaneethakrishna Makaram, PhD, MS – Boston Children's Hospital, Harvard Medical School, Boston, MA, USA
Matthew Pesce, BS – Boston Childrens Hospital
Ellen Grant, MD – Boston Children's Hospital
Phillip Pearl, MD – Boston Children’s Hospital
Alexander Rotenberg, MD PhD – Boston Children's Hospital - Harvard Medical School
Eleonora Tamilia, PhD – Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA
Rationale:
About 30% of patients with epilepsy develop drug-resistant epilepsy (DRE). This significantly increases the risks of morbidity and mortality, making early DRE prediction crucial for improving outcomes. However, in clinical practice, it is hard to predict who will develop DRE at the time of epilepsy diagnosis; therefore, most of these children expose precious developmental years to recurrent seizures and medication side effects, before considering alternative treatments.
Given that brain connectivity is altered in the epileptic brain, we hypothesize that deconstructing the child’s brain network will enable early DRE diagnosis. In this study, we tested whether a network analysis (functional connectivity, FC) of the interictal scalp EEG could predict DRE in drug-naïve children with new-onset seizures.
Methods:
We analyzed the medical records and initial interictal EEG of 97 drug-naïve children with new-onset epilepsy (Fig 1A), who later received diagnosis of drug-resistant (DRE) or drug-sensitive epilepsy (DSE). We performed FC analysis in 5 frequency bands (delta, theta, alpha, beta, gamma; Fig 1B-C). We then characterized each frequency-specific network via a set of features (Fig 1D) that quantify the network distribution and architecture. These were compared between DRE and DSE with a 2-way ANOVA, where the effect of brain structural abnormality (normal vs abnormal MRI) was also tested. Spectral power at various frequencies was also computed and compared between groups.
Finally, we used stepwise logistic regression, followed by ROC curve, to predict DRE based on the patient’s network or spectral (power) properties separately.
Results:
Our cohort included 97 children: 29 with DRE (30%; age=7.9±5.8 years) and 78 with DSE (70%; age=6.9±4.4 years). Abnormal MRI findings were more common in DRE than DSE (54% vs. 21%).
The ANOVA (Fig 2A) showed that the gamma brain network of children with DRE presents higher strength, variability and modularity, but lower clustering and efficiency compared to children who responded to anti-seizure medications (DSE). Beta also showed increased strength and decreased clustering in DRE. Other frequency bands did not show differences and no differences were found in the power features.
From the multi-feature logistic regression, we found that gamma EEG network properties, plus MRI findings, predicted DRE likelihood with an area under ROC curve of 83%, sensitivity of 76% and specificity of 83% (p< 0.0001) (Fig 2B). Abnormal MRI findings alone were not predictive (AUC=0.66, p >0.05), while gamma network features alone were adequate to predict DRE (AUC=0.72; p< 0.001).
Conclusions:
We show that deconstructing the child’s EEG network at epilepsy onset may allow early DRE prediction. Our EEG-based classification model predicted DRE in drug-naïve children with high accuracy when using gamma network properties.
This novel approach could improve outcomes by avoiding ineffective anti-seizure medication trials and expediting the initiation of alternative treatments.
Funding: RSZ TNC Pilot Award
Translational Research