The past decade has witnessed a dramatic increase in the prevalence of autism spectrum disorder (ASD), a neurodevelopmental disorder characterized by social communication and repetitive and restrictive behaviors (1). The CDC estimates that one in 54 children has an ASD diagnosis (2), up from the one in 88 prevalence reported about a decade ago (2). Currently, ASD is diagnosed using behavioral measures, so a diagnosis cannot be made until toddlerhood or later when behavioral symptoms are reliably observable (3). However, there is strong support for the assertion that early intervention leads to better intellectual and behavioral outcomes (4,5). Therefore, a central focus for the field has been to develop identification tools using biological markers to facilitate earlier detection and subsequent intervention of ASD.
Neuroimaging measures provide strong candidate tools for early identification as they can be obtained from the newborn period onwards. For example, several recent studies have used magnetic resonance imaging (MRI) data collected in infancy to predict ASD diagnoses(6,7). However, MRI has several drawbacks, including expense and participant restrictions, making it a less feasible general screening tool. Electroencephalography (EEG), on the other hand, may prove to be a more scalable tool, given its low cost and ease of acquisition in awake and sleeping infants without participant restrictions. Moreover, EEG is known to be sensitive to brain-related changes in ASD before behavioral symptoms are observable (8–12). Initial efforts to predict ASD diagnoses using baseline (i.e. resting-state) EEG early in life have shown promise (13–16). However, diagnostic prediction using EEG recorded during tasks related to ASD symptoms has yet to be attempted and may outperform prior baseline EEG-based classification.
Language is frequently delayed or impaired in ASD (17–21), which may result from atypical peak synaptic sensitivity (22) and cortical excitatory and inhibitory imbalance (23) that disrupt neural circuits typically involved in language development. Therefore, focusing on the brain’s electrical activity during language processing may facilitate improved diagnostic prediction accuracy relative to baseline conditions, and provide insights into the neurobiology of language processing deficits within ASD. Notably, EEG has been used to measure differences in language processing in children with ASD who are older than 12 months (24–26), suggesting EEG is sensitive to atypical neural processing of language stimuli in ASD.
Perceptual narrowing of phoneme discrimination is a critical first stage in language acquisition (26,27) and may be processed differently in infants with ASD (28–30). That is, very young infants can discriminate between native and non-native phonemes better than adults, but they lose this ability over the first year of life as their phoneme perception is tuned to the language(s) experienced in daily life (31). Therefore, this study focused on the phoneme learning window over the first year of life as a potential source of early indicators of subsequent ASD diagnosis.
There were two overarching goals of the present study. First, we aimed to evaluate whether EEG data collected during a language phoneme task at either 6- or 12-months of age in infants with familial risk for ASD can accurately predict later ASD diagnosis. We utilized EEG data collected from high familial risk infant siblings as part of a prospective longitudinal study, where diagnosis of ASD was determined at 2–3 years of age. Though power analysis of EEG is most common, nonlinear measures can capture dynamical properties of the brain that power analysis is not able to quantify. Nonlinear measures of adult EEG have accurately classified other clinical conditions, including depression (32–34), schizophrenia (35–37), and epilepsy (38–40). Our lab previously found that these measures computed from resting-state EEG are useful in predicting ASD outcome (15), and we now aim to improve predictive capacity by evaluating these measures on task-related data. Second, given that expected perceptual narrowing of phoneme discrimination occurs between 6 and 12 months of age, we aimed to compare the EEG features most predictive of diagnosis and determine whether there are developmental differences in which features are most important during versus after the language phoneme learning period.