Resting state EEG power spectrum and functional connectivity in autism: a cross-sectional analysis

Abstract

Background: Understanding the development of the neuronal circuitry underlying autism spectrum disorder (ASD) is critical to shed light into its etiology and for the development of treatment options. Resting state EEG provides a window into spontaneous local and long-range neuronal synchronization and has been investigated in many ASD studies, but results are inconsistent. Unbiased investigation in large and comprehensive samples focusing on replicability is needed.

Methods: We quantified resting state EEG alpha peak metrics, power spectrum (PS, 2-32 Hz) and functional connectivity (FC) in 411 children, adolescents and adults (n = 212 ASD, n = 199 neurotypicals [NT], all with IQ > 75). We performed analyses in source-space using individual head models derived from the participants' MRIs. We tested for differences in mean and variance between the ASD and NT groups for both PS and FC using linear mixed effects models accounting for age, sex, IQ and site effects. Then, we used machine learning to assess whether a multivariate combination of EEG features could better separate ASD and NT participants. All analyses were embedded within a train-validation approach (70%-30% split).

Results: In the training dataset, we found an interaction between age and group for the reactivity to eye opening (p = .042 uncorrected), and a significant but weak multivariate ASD vs. NT classification performance for PS and FC (sensitivity 0.52-0.62, specificity 0.59-0.73). None of these findings replicated significantly in the validation dataset, although the effect size in the validation dataset overlapped with the prediction interval from the training dataset.

Limitations: The statistical power to detect weak effects-of the magnitude of those found in the training dataset-in the validation dataset is small, and we cannot fully conclude on the reproducibility of the training dataset's effects.

Conclusions: This suggests that PS and FC values in ASD and NT have a strong overlap, and that differences between both groups (in both mean and variance) have, at best, a small effect size. Larger studies would be needed to investigate and replicate such potential effects.

Keywords: Autism spectrum disorder; EEG; Functional connectivity; Power spectrum; Resting state.

Fig. 1

Overview of the statistical analysis approach. Univariate and multivariate statistics were performed in the training dataset, as well as control comparisons to evaluate the sensitivity of the results to pipeline choices. From this, concrete hypotheses are generated and tested in the validation dataset

Fig. 2

Alpha peak measures. For each measure, the scatter plot of the raw values as a function of age is shown in the left side and the residuals of the linear mixed effects model y ~ 1 + age + sex + IQ + (1|site) on the right side. All plots derived from the training dataset (147 ASD and 140 NT)

Fig. 3

Reactivity to eye opening: distribution and age trends in the training (A) and the validation dataset (B). The scatter plot of the raw values as a function of age is shown in the left side, along with the regression lines for each group. The distribution of reactivity values for each age group along with the corresponding Cohen’s d effect size is shown on the right side. Typically, d ~ 0.20 is considered small and d ~ 0.50 a medium effect size. PI indicates the 95% prediction interval from the training dataset to the validation dataset, and it was calculated following [38] based on the training dataset effects and the sample size of both datasets

Fig. 4

Classification performance of the multivariate models in internal cross-validation and external test in the validation dataset. For each of the four models subjected to testing in the validation dataset, the cross-validation performance of each of the repeated random splits within the training dataset is shown as a gray dot along with the performance in the validation dataset (blue line). All and Child indicate models trained and tested in the whole cohort and children cohort, respectively. enet and SVC represent elastic net and linear support vector classifier models, respectively