Statistical Analysis
All statistical analyses were performed with R software (version 4.1.2,
R Core Team,2021).
To investigate tDCS-induced changes in HRV, we calculated the change
from baseline to the tDCS periods (5-min tDCS, during 0-back, and last
5-min tDCS) using delta scores (i.e: Δ score = RMSSD of 5-min tDCS minus
RMSSD of baseline or HF-HRV of 5-min tDCS minus HF-HRV of baseline,
etc). Outliers were inspected via boxplot distribution and three
observations were excluded for the RMSSD data, while none was excluded
for the HF-HRV (Sup. Material - Appendix 2). A linear mixed model (LMM)
(‘lme4’ package) was used to assess the effects of tDCS on the HRV. The
model included either delta RMSSD or HF-HRV scores as the dependent
variable, with a fixed effect of tDCS protocol, time and their
interaction (tDCS protocol*time). ‘Subject’ was employed as a random
intercept (full statistical model: ‘RMSSD/HF-HRV ~ tDCS
protocol*time + (1|Subject))‘. Pairwise analyses were performed
using the ‘emmeans’ function.
Secondly, we investigated whether the HRV changes were associated with
E-field magnitude induced by tDCS in the brain regions of interest. In
the first step we employed LMM models having the RMSSD/HF-HRV delta
scores as dependent variables, whereas the mean E-field in each region
of interest were the fixed factors. ‘Subject’ was considered a random
intercept. To investigate whether the mean individual E-fields were
associated with the outcome, the same model was used but the variable
‘protocol’ was included as a fixed factor (RMSSD/HF-HRV
~E-field + tDCS protocol + (1|Subject)). In
total, eight models were performed per analysis (4 brain regions of
interest x 2 hemispheres). Therefore, multiple comparison corrections
were conducted using the false discovery rate (‘stats’ package). Only
the corrected p-values are presented here.
Exploratory analyses investigated whether the HRV changes could be
influenced by baseline mood or state anxiety. For this analysis a LMM
was fitted with the RMSSD/HF-HRV as the dependent variable and the
interaction between mood and tDCS protocol (i.e: VAS baseline*tDCS
protocol and STAI*tDCS protocol) as the independent variable (full
model: RMSSD/HF-HRV ~ mood baseline*tDCS protocol +
(1|Subject)). For all statistical tests, the significance level
was set to alpha = 0.05.
Results
A total of 40 healthy volunteers were included. Two dropped-out after
the first session and one was excluded due to non-normal alterations in
the heart rate as indicated by a cardiologist. The final analyses were
performed with 37 subjects (mean age = 24.3 years, standard deviation
(SD) = 4.8), representing a total of 111 sessions being performed. Due
to electrocardiogram artifacts during data collection (especially during
the 0-back) a few time points were dropped from our dataset. Therefore,
our analyses were conducted with a total of 103 time-points each for
sham and 1.5mA, and 96 time-points for the 3.0mA protocol. For more
information, a comprehensive table can be found in the Sup. Material -
Appendix 3.