Applying fractal analysis to short sets of heart rate variability data.

ABSTRACT

The aim of this study was to explore the interchangeability of fractal scaling exponents derived from short- and long-term recordings of real and synthetic data. We compared the alpha(1) exponents as obtained by detrended fluctuation analysis from RR-interval series (9 am to 6 pm) of 54 adults in normal sinus rhythm, and the alpha(1) estimated from shorted segments of these series involving only 50, 100, 200 and 300 RR intervals. Three series of synthetic data were also analysed. The principal finding of this study is the lack of individual agreement between alpha(1) derived from long and short segments of HRV data as indicated by the existence of bias and low intraclass correlation coefficient (r(i) = 0.158). The extent of variation in the estimation of alpha(1) from real data does not only appear related to segments' length, but also to different dynamics among subjects or lack of uniform scaling behaviour. However, we did find statistical agreement between the means of alpha(1) exponents from long and short segments, even for segments involving just 50 RR intervals. According to results of synthetic series, the 95% confidence interval found for the variation of alpha(1) using segments with 300 samples is [-0.1783 + 0.1828]. Caution should be taken concerning the use of short segments to obtain representative exponents of fractal RR dynamics; a circumstance not fully considered in several studies.