Reducing the impact of low sampling frequency on heart rate variability

Background: Heart rate variability (HRV) provides a view of a patient's sympathetic and parasympathetic balance. Low values of HRV are abnormal, therefore factors that falsely increase the apparent HRV value will reduce the sensitivity to detect abnormal HRV. In this study, we focus on the impact of additive noise on RR interval accuracy and HRV for two methods of recovering high resolution RR from low sample rate ECG. Methods: The two methods of RR interval recovery were based on up sampling. The first method chooses R-wave peaks. The second method uses template matching to align beats. Twenty minute samples of single lead ECG (n = 1000) were generated by the Physionet HRV ECG simulator. Each ECG sample had a random combination of heart rate standard deviation and additive noise. ECG was decimated to 125sps from 1000sps to test the RR interval recovery methods. RR interval reference was the sequence of R-wave markers from the simulator. HRV standard deviation of RR intervals (SDNN)was calculated for each record using three RR interval series, 125sps R-wave peaks, up-sampled 1000sps peaks and up-sampled 1000sps template-match-corrected RR. Bland-Altman analysis was used to assess SDNN bias. Impact of noise level on RR interval error was assessed by standard deviation of RR interval error (RRerrSD) in noise level bins. Results: Tables 1 shows the bias of SDNN at low values. Table 2 shows RR interval error by noise level. The up-sampled methods have much lower SDNN error in the low range of SDNN. RRerrSD increases with increasing noise for the peak picking methods. RRerrSD was constant across noise levels for template matching but higher than the up-sampled R peaks method at lower noise levels. Conclusion: The up-sampled R peak method of RR interval recovery showed lower RR interval and SDNN error in general but the template match method was better at higher noise levels.