Effect of missing RR-interval data on heart rate variability analysis in the time domain.

In this study, the effects of missing RR-interval data on time-domain analysis were investigated using simulated missing data in real RR-interval tachograms and actual missing RR data in an ECG obtained by an unconstrained measurement. For the simulation, randomly selected data (0-100 s) were removed from real RR data obtained from the MIT-BIH normal sinus rhythm database. In all, 2615 tachograms of 5 min durations were used for this analysis. For certain durations of missing data, the analysis was performed by 1000 Monte Carlo runs. MeanNN, SDNN, SDSD, RMSSD and pNN50 were calculated as the time-domain parameters in each run, and the relative errors between the original and the incomplete tachograms for these parameters were computed. The results of the simulation revealed that MeanNN is the parameter most robust to missing data; this feature can be explained by the theory of finite population correction (FPC). pNN50 is the parameter most sensitive to missing data. MeanNN was also found to be the most robust to real missing RR data derived from a capacitive-coupled ECG recorded during sleep; furthermore, the parameter patterns for the missing data were considerably similar to those for the original RR data, although the relative errors may exceed those of the simulation results.