Procedure for spectral estimation and RSA estimation

The error correction of RR series was based on outliers detection. The median of the last 10 RR intervals was used. The outliers were classified in false positives, false negatives or ectopic beats. A correction in the RR series was applied to keep the total duration of the recording²⁷. After the artifact removal, the RR series were resampled to 4 Hz with a shape-preserving piecewise cubic interpolation for the spectral estimation. For each paced respiratory frequency interval, the corresponding RR interval was extracted from the whole RR resampled series. The segment was windowed with a Hann window and zero padded to a number of samples power of 2. The power spectrum was estimated by using an Fast Fourier Transform (FFT). The RSA was identified by looking for the maximum of the spectrum in a frequency interval ± 20 of the paced frequency. If a discrepancy greater than 5% was detected, we considered that the subject was not following the paced rhythm and the record was discarded.