Towards the determination of the optimal recording duration for Heart Rate Variability Applications. A bootstrap study on the ‘’Fantasia’’ HRV database

Abstract

In order to determine the shortest duration (L) of a cardiotachogram
(CTG) that allowed statistically reliable analysis of heart rate variability (HRV) the CTGs
corresponding to five young (Y) and to five elderly (O) healthy subjects were downloaded
from the Fantasia database; data from Fantasia is freely available at
www.physionet.org. Recording conditions:
2 h of continuous recording in supine
position. Bootstrap analysis was applied
for generating a large amount of replicate
experiments to obtain empirical distributions
corresponding to three different
time-domain HRV indices. PNN50, and
SDNN are generally accepted definition
of HRV indices; the Power Law Index
(PLI), a measure of fractal self-similarity
was also used. The selected indices accounted
for group separation when large
duration segments were used (L = 4 000
heartbeats). The discriminative capability,
however, was reduced as L was shortened.
The application of the bootstrap method
revealed that significant differences (PI <
0.05) may be found for all the three indices
when cardiotachograms as short as 70-
150 heartbeats are compared. Duration as
short as 15 heartbeats are sufficient for
obtaining faithful discriminations based
on SDNN, and PNN50, which is in agreement
with recent studies by other authors
using conventional ECG traces for HRV
studies The probability of group overlapping
assessed by the bootstrap method
was high for short duration segments. This
probability decreased as L increased, in
an HRV index-dependent way. Optimal L
values corresponding to minimal group
overlapping were around 1 000 heartbeats
for SDNN, 2 000 heartbeats for PNN50, and
close to 4 000 for PLI. With the application
of a computer intensive statistical method
it was able to extract relevant informationabout optimal L values for HRV analysis.
Hopefully, these results may be used for
other HRV studies. In particular, for optimally
designing group as well as individual
HRV studies. Short-term HRV is
supported for research aimed at finding
differences between groups. For clinically
relevant assessment of the HRV indices
of an individual patient, at least 1 h of recording
seems to be required.