Machine learning‐enabled epileptiform spike detection stratifies patient subpopulations with Alzheimer’s disease

Abstract

AbstractBackgroundA subpopulation of patients with Alzheimer’s disease exhibit epileptiform activity and have more rapidly progressing disease. Manual identification of these pathological events is infeasible at scale and machine learning tools may establish epileptiform activity as an actionable biomarker for stratification and endpoints within clinical trials.MethodsRoutine EEGs from 90 subjects with AD and 39 subjects with MCI from the Beacon data repository were analyzed via the Beacon Platform harnessing an ML‐enabled interictal epileptiform discharge (IED) detection algorithm, and by expert visual inspection.ResultsIEDs were found in 17 of the 90 subjects with AD (19%), and 8 out of the 39 MCI subjects (20%). In subcategorization, 13% (9 / 71) of AD subjects without epilepsy had IEDs, while 42% (8/19) of AD subjects with epilepsy had IEDs. In the AD/no epilepsy subgroup there were 24 patients taking an Anti‐Seizure Medication (ASM), and 4% (1/24) had IEDs (versus 17% (8/47) of the AD without epilepsy group that were not on ASMs).ConclusionMachine learning augmented detection of IEDs allows for rapid, quantitative identification of individuals with subclinical epileptiform activity in patients with AD and MCI. These results demonstrate that AI‐detectable IEDs are seen in a sizable minority of AD patients without epilepsy and in MCI patients, and more commonly in AD patients with epilepsy. Ultimately, automated analysis of EEG data is a promising way to stratify patients and track disease progression while enriching clinical trials with subjects most likely to respond to next‐generation treatments.