Effect of missing RR-interval data on nonlinear heart rate variability analysis

Abstract

The effects of missing RR-interval data on nonlinear heart rate variability (HRV) analysis were investigated using simulated missing data in actual RR-interval tachograms and actual missing RR-interval data. For the simulation study, randomly selected data (ranging from 0 to 100s) were removed from actual data in the MIT-BIH normal sinus rhythm RR-interval database. The selected data are considered as a simulated artefact section. In all, 7182 tachograms of 5-min duration were used for this analysis. For each missing interval, the analysis was performed by 100 Monte Carlo runs. Poincaré plot, detrended fluctuation, and entropy analysis were executed for the nonlinear HRV parameters in each run, and the normalized errors between the data with and without the missing data duration for these parameters, were calculated. In this process, the usefulness of reconstruction was considered, for which bootstrapping and several interpolation methods (nearest neighbour, linear, cubic spline, and piecewise cubic Hermite) were used. The rules for the reconstruction, derived from the results of these simulations, were evaluated with actual missing RR-interval data obtained from a capacitive-coupled ECG during sleep. In conclusion, nonlinear parameters, excepting Poincaré-plot-analysis parameters, may not be appropriate for the accurate HRV analysis with missing data, since these parameters have relatively larger error values than time- or frequency-domain HRV parameters. However, the analysis of the long-term variation for nonlinear HRV values can be available through applying the rules for the reconstruction obtained in this study.