Computational protocol: Performance Monitoring in Medication-Naïve Children with Tourette Syndrome

Similar protocols

Protocol publication

[…] We preprocessed the EEG in Matlab (Mathworks, Natick, MA, USA) using the EEGLAB toolbox (Delorme and Makeig, ) and in-house scripts.The continuous EEG data were resampled to 500 Hz. The data were then re-referenced to common average reference, and filtered from 0.5 to 45 Hz using a finite impulse response filter generated with the firfilt plugin (Widmann, ).For artifact removal/reduction, the data were segmented into stimulus-locked (−0.5 to +1 s), and response-locked epochs (−1 to 0.5 s). The prestimulus period was used as baseline for both epochs. Epochs were excluded when exceeding a ±300 μV amplitude criterion. The remaining epochs were sorted using a summary score of root mean square amplitude across all channels and time points, spatial SD, power spectrum ratio between low and high frequencies, skewness and kurtosis, normalized to unit variance across epochs. Only epochs within ±1 SD were retained for further analysis. These epochs were concatenated and subjected to temporal independent component analysis (ICA) using the infomax algorithm (Bell and Sejnowski, ), and 32 components were estimated. We used spatial templates to identify horizontal and vertical eye movements and ECG artifacts, and removed these automatically (Viola et al., ). Following the rationale presented in COMPASS (Wessel and Ullsperger, ), we assumed that components of interest were broad, dipolar topographies with time-locked event-related responses, and we therefore generated scores based on the spatial smoothness of the component scalp maps and the root mean square of the event related average, and retained the top 15 components. These were then visually cross-checked, and components reminiscent of artifacts were marked. Between 10 and 15 components were kept and back-projected in this manner. [...] Statistics were performed in Matlab and Statistica (Statsoft, Tulsa, OK, USA). Repeated measure analyses were conducted to test outcome effects in the behavioral and the ERP data (congruent vs. incongruent vs. error trials) and “group.” Additional univariate Analysis of Covariance (ANCOVA) were conducted for behavioral measures and ERP components as dependent variable, group as categorical factor and covariates as continuous predictors to test group differences. Significant or trend-significant effects were followed-up with additional post-hoc tests. All statistics were considered significant at p < 0.05. The effect size indicator partial eta squared (ηp2) is reported for each significant/trend-significant statistical comparison as a measure of the strength of the effect, with of 0.01 representing a small effect, 0.06 a medium effect, and 0.14 a large effect (Cohen, ). To demonstrate the adequacy of pooling children with TS with and without comorbid ADHD, we also performed ANCOVAs with four groups, separating TS only and TS+ADHD, control group, ADHD, with the main behavioral and ERP result.Response times (RT) and response accuracy (RACC) averages were generated for all possible outcomes. Premature responses faster than 200 ms and slow responses >2000 ms were not considered in the averages. RTs were analyzed with covariates:Age: Because of substantial speeding of RT, and improvement of accuracy with age across the entire sample regardless of group, all analyses included age as a covariate.FSIQ: We decided to analyze the behavioral data with FSIQ as covariate for the sake of consistency across behavioral- and ERP analyses. This appears to be the most sound practice in our case, however see relevant publications for a discussion on this issue (Willcutt et al., ; Dennis et al., ).ERP components were analyzed with covariates:Age: Groups did not differ in mean age. However, to control within group variation of electrophysiological measures we followed current guidelines (Picton et al., ). Age in particular influences many features in the EEG, resulting also in prominent maturational changes of ERP amplitudes and latencies (Davies et al., ; Wiersema et al., ; Brydges et al., ; Rojas-Benjumea et al., ).FSIQ: Earlier research has also shown that IQ differences account for variability of ERP measures. We therefore decided to include IQ as a covariate in line with other studies in the field (Pelosi et al., ; Deary and Caryl, ; Jausovec and Jausovec, ; Ramchurn et al., ).RT/ IIV: Response times and their variability substantially affect ERP features (Eichele et al., ). This is partly due to task-induced amplitude modulation, and partly nuisance variability due to spatio-temporal overlap of stimulus and response-related components, see also (Ramchurn et al., ).RACC: Average accuracy provides a gross measure of the effort that an individual invests in a task, therefore adjusting for ACC is useful to account for state and trait factors not specifically related to diagnosis/group.ADHD symptom scores were included initially as a covariate in the statistical models for the behavioral and the ERP correlates, but proved non-significant and were subsequently removed from both models. Pairwise correlations were used to further investigate significant effects of the group factor and covariates. To test for post-error slowing (PES) and to compensate for confounders, we conducted a pairwise comparison of post-error and pre-error trials around each error (Dutilh et al., ) followed by an ANCOVA, including the covariates age and FSIQ. […]

Pipeline specifications

Software tools EEGLAB, Statistica
Applications Miscellaneous, Clinical electrophysiology
Organisms Homo sapiens
Diseases Tourette Syndrome