Computational Decision Support in Epilepsy Using Retrospective EEG

Epilepsy Clinical Trial

Official title: Retrospective Analysis of Resting-State EEG in the Diagnosis of Epilepsy to Validate a Computational Biomarker for Seizure Susceptibility

Status Completed

Clinical Trial Summary

The primary aim is to validate a set of computational biomarkers as potential decision support in epilepsy on a large cohort of study participants that were diagnosed with epilepsy and controls that ended up with another diagnosis (such as syncope or non-epileptic seizures). The goal is to examine if the methodology works robustly on this large cohort, and can theoretically contribute to the reduction of misdiagnosis rates.

The secondary aim is to examine whether the computational biomarkers could contribute to reducing the waiting time and the number of clinical appointments needed before a final diagnosis is made.

Clinical Trial Description

Mathematical models provide a powerful and useful tool with which to identify and understand biological mechanisms that may lead to the risk of having seizures as well as how they generate, propagate and terminate (Wendling, 2005). Mathematical models that combine experimental and clinical detail at diverse scales have revealed the importance of many microscopic and macroscopic mechanisms in the generation of seizure-like activity, ranging from genetic and molecular mechanisms to changes in the excitability of neural populations leading to the generation of pathological oscillations (for review see Woldman & Terry (2015); Soltesz & Staley (2008)). Due to the increased availability of data recordings (EEG, MRI, MEG, CT, PET), there has been a significant increase in research studies that aim to identify novel biomarkers from these recordings with potential clinical value, using various different techniques (e.g. time-series analysis, computational modelling, machine learning).

By combining mathematical and computational techniques, we have identified properties in the resting-state EEG (eyes closed, relaxed) of people with epilepsy that differ from those of controls as well as their first-degree relatives (Chowdhury et al., 2014). Developing these approaches and applying them to routine recordings from individuals with epilepsy against a control cohort (Schmidt et al., 2016) revealed levels of diagnostic accuracy similar to current general (i.e. non-specialist) neurology practices (60% sensitivity, 87% specificity, N=68). Crucially, our method correctly classified several subjects using their first EEG, whereas clinical diagnosis was confirmed only after prolonged telemetric recordings over many months.

Since our methods and analysis depend on short segments of resting-state EEG only, its accuracy and efficacy do not rely on capturing epileptiform abnormalities, in contrast to the current use of EEG in diagnosing epilepsy. Since many EEGs return negative, clinicians are often faced with the problem of deciding on whether to opt for longer recordings of EEG or ambulatory or video EEG, which is currently the final method in the diagnostic cascade. This is time-consuming, expensive and relies on the availability and expertise of trained EEG-readers. By optimally interrogating short segments of background activity with mathematical and computational analysis, our methods, in the short term, provide additional evidence that could guide clinicians in future diagnostic steps. ;

Related Conditions & MeSH terms

• Epilepsy

• NCT number: NCT05384782
• Study type: Observational
• Source: Cornwall Partnership NHS Foundation Trust
• Status: Completed
• Start date: December 1, 2019
• Completion date: March 31, 2022