Automated algorithms for seizure forecast: a systematic review and meta-analysis.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epileptic seizures pose significant challenges for patient management and quality of life.
• Automated seizure forecasting aims to predict seizures, enabling proactive interventions.
• Existing algorithms vary in methodology and performance, necessitating a comprehensive review.

Purpose of the Study:

• To systematically review and characterize the methodologies and performance of automated seizure forecast algorithms.
• To establish a benchmark for current seizure forecasting technology.
• To identify gaps and propose guidelines for future research and development.

Main Methods:

• Systematic literature review of studies published up to May 10, 2024.
• Inclusion criteria: original, patient-specific algorithms for human epileptic seizure forecast using intraindividual cyclic event distribution and/or surrogate preictal state measures.
• Two meta-analyses were performed: one for Area Under the ROC Curve (AUC) and another for Brier Skill Score (BSS).

Main Results:

• Eighteen studies met eligibility criteria, encompassing 43 unique algorithms and data from 419 patients with 19,442 reported seizures.
• The overall mean AUC across eligible algorithms was 0.71, with similar performance regardless of input data type (cyclic events, surrogate measures, or combined).
• The overall mean Brier Skill Score (BSS) was 0.13, also showing consistency across different input data strategies.

Conclusions:

• Automated seizure forecast algorithms demonstrate a consistent, moderate level of performance, indicated by the mean AUC of 0.71.
• A significant lack of standardization in study design and performance evaluation was identified.
• Guidelines are proposed to promote standardization and advance the development of reliable seizure forecasting solutions.