Tinnitus distress targeted the frequency bands beta3 (23.5–35 Hz) and gamma (35.5–45 Hz) in right frontal regions, whereas delta (0.5–4 Hz) exhibited significant changes in temporal-parietal sources. Further, 57.8% of the total variance in EEG power was subject-specific and acknowledged by the LMM framework and its prediction.

Accounting for Heterogeneity: Mixed-Effects Models in Resting-State EEG Data in a Sample of Tinnitus Sufferers:

Abstract

In neuroscience, neural oscillations and other features of brain activity recorded by electroencephalography (EEG) are typically statistically assessed on the basis of the study’s population mean to identify possible blueprints for healthy subjects, or subjects with diagnosable neurological or psychiatric disorders. Despite some inter-individual similarities, there is reason to believe that a discernible portion of the individual brain activity is subject-specific. In order to encompass the potential individual source of variance in EEG data and psychometric parameters, we introduce an innovative application of linear mixed-effects models (LMM) as an alternative procedure for the analysis of resting-state EEG data. Using LMM, individual differences can be modelled through the assumptions of idiosyncrasy of all responses and dependency among data points (e.g., from the same subject within and across units of time) via random effects parameters. This report provides an example of how LMM can be used for the statistical analysis of resting-state EEG data in a heterogeneous group of subjects; namely, people who suffer from tinnitus (ringing in the ear/s). Results from 49 participants (38 male, mean age of 46.69 ± 12.65 years) revealed that EEG signals were not only associated with specific recording sites, but exhibited regional specific oscillations in conjunction to symptom severity. Tinnitus distress targeted the frequency bands beta3 (23.5–35 Hz) and gamma (35.5–45 Hz) in right frontal regions, whereas delta (0.5–4 Hz) exhibited significant changes in temporal-parietal sources. Further, 57.8% of the total variance in EEG power was subject-specific and acknowledged by the LMM framework and its prediction. Thus, a deeper understanding of both the underlying statistical and physiological patterns of EEG data was gained.