Multivariate Resting-State EEG Organization in Autism: Shared Age-Related Structure and Diagnostic Context Dependent IQ Associations

Resting-state electroencephalography (EEG) studies of autism have often sought categoricaldifferentiation between autistic and typical neurodevelopmental groups, with limited stabilityacross features or classifiers. A neurodiversity-informed perspective views autism as part ofhuman neurodevelopmental variation, while recognizing associated disability and support needs.In this view, overlap between autistic and non-autistic EEG distributions is not merelyclassification noise, but may reflect shared dimensions of neural organization. A complementaryquestion arise is whether position along these dimensions has different functional meaning acrossdiagnostic contexts. We conducted an exploratory secondary analysis of 726 eyes-open resting-state EEG features previously extracted in a published multi-site autism EEG study, with a finalanalytic sample of 767 children and adolescents: 344 controls (CON), 347 participants classifiedby the Autism Diagnostic Observation Schedule as Autistic Disorder (ADOS-A), and 76classified as Autism Spectrum (ADOS-S). Principal component analysis was applied once to thepooled sample without including age, sex, IQ, or diagnostic group. The dominant component(PC1) was interpreted post hoc by examining its EEG feature loading pattern, its association withchronological age, the consistency of this age association across diagnostic groups, and theconcentration of age-related variance across components. We then tested whether age-adjustedposition on this component was differentially associated with general cognitive performance(i.e., IQ) across groups. PC1 explained 15.94% of feature variance and was characterized by thestrongest positive loadings on delta-band power and the strongest negative loadings on sampleentropy. PC1 showed a significant negative association with chronological age (𝑟 = −0.846,𝑝 < .001), indicating that older age was associated with relatively lower delta power and highersample entropy along this component, consistent with previously reported developmental EEGchanges. Age-related EEG variance was concentrated primarily on PC1 rather than onsubsequent components, supporting its interpretation as a shared age-related EEG dimension.After removing PC1’s linear association with age, residual PC1 scores (PC1r) were negativelyassociated with IQ within the ADOS-A group (𝑏 = −0.612, 𝑝 < .001), but not within thecontrol or ADOS-S groups. Pairwise slope contrasts indicated that this association wassignificantly more negative in ADOS-A than in CON, whereas ADOS-A vs. ADOS-S andADOS-S vs. CON contrasts were not significant. These findings shift emphasis from categoricalEEG abnormality toward individual variation within a broadly shared age-related EEGorganization, in which brain–cognition coupling depends on diagnostic context.