ECG collection and RSA-R quantification
RSA-R was extracted from each participant’s electrocardiogram (ECG) using the following steps. ECG was acquired continuously at 2000 Hz using a BIOPAC MP150 system with ECG 100C amplifiers and analyzed offline using the AcqKnowledge 5.0.6 software (Biopac Systems, 2018). The ECG signal was recorded during the MacArthur Reactivity Protocol using the ECG100C amplifier with three pre-gelled Ag/AgCl electrodes with a lead II configuration.
Offline, ECG signals were bandpass filtered using an FIR filter from 0.5 to 35 Hertz. Next, R-R intervals were extracted for the ECG signal using a modified Pan-Tompkins QRS detector (Pan & Tompkins, 1985) and resampled to 8hz after cubic spline interpolation. Data were then visually inspected for movement artifacts or incorrect identification of the QRS complex by the scoring algorithm.
RSA values were computed using a fast Fourier transform with a hamming window within the respiratory frequency band that corresponded to the participant’s age (0.2-1.0Hz for children under six, 0.15-0.5Hz for children six and older; Boyce et al., 2001, Gentzler et al. 2009). RSA values were computed in 60 second epochs and log transformed. RSA during the first and last 60 seconds was averaged for each baseline and challenge task, respectively. RSA reactivity (RSA-R) reflected the difference score for each challenge task minus its corresponding comparison task11The psychometric properties of using a regression residual are preferable, but more difficult to interpret. We computed all analyses with both difference scores and with the regression residual and all results were identical. Difference scores are reported to increase ease of interpretation..
Analytic plan
We characterized the feasibility of ECG collection in this sample (Aim 1) by computing the percent of collected, and subsequently usable, ECG data for each MRP task across both visits. For participants with usable ECG data at both visits, we ran intraclass correlations with RSA values for corresponding tasks at visits 1 and 2 to assess the test-retest stability of each RSA measure (Aim 1). ICC values were interpreted as poor (< 0.4), fair (0.4-0.59), good (0.6-0.74), and excellent (0.75-1.0; Cicchetti, 1994). Finally, for tasks with adequate ICCs, we tested whether RSA-R related with concurrent externalizing behavior (Aim 2) at visit 1 via bivariate Pearson’s correlations. Before conducting analyses, we first confirmed that age and developmental level were not related to RSA or externalizing scores.
According to Bujang & Baharum (2017) with  80% power, 15 subjects with data at both timepoints are required to detect an ICC of 0.6 (i.e., good). Seven subjects with data at both timepoints are required to detect an ICC of 0.8 (i.e., excellent reliability). For Aim 2, data from 25 children would enable detection of a large effect size (ƒ2 = 0.35) with power of 0.8 and α = 0.05. Fewer children with required language levels were available to participate given the COVID pandemic, so we may not detect all large effects for concurrent validity estimates. Given the study was designed to pilot the initial feasibility and stability, we believe the data reported could still be useful for planning future studies.
Results