Infant EEG activity as a biomarker for autism: a promising approach or a false promise?

The ability to determine an infant's likelihood of developing autism via a relatively simple neurological measure would constitute an important scientific breakthrough. In their recent publication in this journal, Bosl and colleagues claim that a measure of EEG complexity can be used to detect, with very high accuracy, infants at high risk for autism (HRA). On the surface, this appears to be that very scientific breakthrough and as such the paper has received widespread media attention. But a close look at how these high accuracy rates were derived tells a very different story. This stems from a conflation between "high risk" as a population-level property and "high risk" as a property of an individual. We describe the approach of Bosl et al. and examine their results with respect to baseline prevalence rates, the inclusion of which is necessary to distinguish infants with a biological risk of autism from typically developing infants with a sibling with autism. This is an important distinction that should not be overlooked.