A multimodal clinical data resource for personalized risk assessment of sudden unexpected death in epilepsy

Abstract

Epilepsy affects ~2–3 million individuals in the United States, a third of whom have uncontrolled seizures. Sudden unexpected death in epilepsy (SUDEP) is a catastrophic and fatal complication of poorly controlled epilepsy and is the primary cause of mortality in such patients. Despite its huge public health impact, with a ~1/1,000 incidence rate in persons with epilepsy, it is an uncommon enough phenomenon to require multi-center efforts for well-powered studies. We developed the Multimodal SUDEP Data Resource (MSDR), a comprehensive system for sharing multimodal epilepsy data in the NIH funded Center for SUDEP Research. The MSDR aims at accelerating research to address critical questions about personalized risk assessment of SUDEP. We used a metadata-guided approach, with a set of common epilepsy-specific terms enforcing uniform semantic interpretation of data elements across three main components: (1) multi-site annotated datasets; (2) user interfaces for capturing, managing, and accessing data; and (3) computational approaches for the analysis of multimodal clinical data. We incorporated the process for managing dataset-specific data use agreements, evidence of Institutional Review Board review, and the corresponding access control in the MSDR web portal. The metadata-guided approach facilitates structural and semantic interoperability, ultimately leading to enhanced data reusability and scientific rigor. MSDR prospectively integrated and curated epilepsy patient data from seven institutions, and it currently contains data on 2,739 subjects and 10,685 multimodal clinical data files with different data formats. In total, 55 users registered in the current MSDR data repository, and 6 projects have been funded to apply MSDR in epilepsy research, including three R01 projects and three R21 projects.

Reliable detection of generalized convulsive seizures using an off‐the‐shelf digital watch: A multisite phase 2 study

“…The new testing frameworks for GCS detection algorithms are vital in a real‐world setting due to this problem of extreme class imbalance in long‐term datasets. As described by Li et al., 22 most seizure detection studies using existing evaluation methods report near‐perfect algorithm performance but rarely result in routine clinical usage. In the context of wearable‐based seizure detection, this is exemplified by overreliance on cross‐validation algorithm testing (Table 2).…”

“…An OTS acquisition-based GCS detection algorithm was compared against the gold standard epilepsy monitoring unit (EMU) video-EEG-recorded seizure data. Previous studies have used machine learning to detect seizures using classical machine learning approaches such as support vector machine, 4,12,18,19,20 k-nearest neighbor, 5,21 random forestbased algorithms, 5,21 and deep learning algorithms, 3 and limit evaluation to leave-one-patient-out cross-validation (LOPO CV), 22 which does not capture real clinical scenarios. LOPO CV provides an overly optimistic estimate of algorithm performance due to (1) small sample size of testing datasets, (2) lack of independence in training and testing datasets, and (3) strong auto correlation in time series data.…”

A hybrid unsupervised and supervised learning approach for postictal generalized EEG suppression detection

“…Center for SUDEP Research aims to better understand cortical, subcortical, and brainstem mechanisms involved in SUDEP through a data-driven, systems biology approach that focuses on cortical influences in SUDEP. The CSR's Informatics and Data Analytics Core (IDAC; NIH U01NS090408) has developed an infrastructure for integrating and analyzing prospectively collected data related to SUDEP from different domains, such as clinical, electrophysiological, biochemical, genetic, and neuropathological fields (Li et al, 2022 ). The CSR data repository contains multimodal data from over 2,500 epilepsy patients (a broad spectrum of ages as well as social, racial, and ethnic groups), including thousands of 24-hour electrophysiological recordings in the European Data Format (Li et al, 2020 ).…”

Sudden unexpected death in epilepsy

“…It is clear from studies such as the multicentre retrospective Mortality in Epilepsy Monitoring Units Study (MORTEMUS), that multifactorial monitoring will be important to continue to understand and stratify SUDEP risk in patients with epilepsy. Such multimodal monitoring has begun to be implemented in some centres and is producing valuable information [41,42 ▪ ,43 ▪ ,44].…”