Quantifying errors in spectral estimates of HRV due to beat replacement and resampling

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Heart rate variability (HRV) analysis commonly uses spectral estimation techniques like the fast Fourier transform (FFT).
• FFT requires evenly sampled time series data, but cardiac RR intervals are inherently irregular.
• Traditional methods involve resampling RR intervals, potentially introducing inaccuracies.

Purpose of the Study:

• To evaluate the impact of interpolation and resampling on spectral estimates of HRV.
• To compare the performance of the Lomb-Scargle (LS) periodogram against traditional FFT-based methods for irregularly sampled RR interval data.
• To assess the necessity and impact of ectopy removal on spectral estimation accuracy.

Main Methods:

• Utilized a realistic artificial RR interval generator to simulate cardiac data.
• Applied linear and cubic spline resampling techniques prior to FFT-based spectral estimation.
• Employed the Lomb-Scargle (LS) periodogram, designed for unevenly sampled data.
• Investigated the effects of ectopy removal and phantom beat replacement on power spectral density (PSD) estimates.
• Compared spectral estimation techniques on real RR interval data during sleep.

Main Results:

• Interpolation and resampling consistently led to over-estimations of the power spectral density (PSD) compared to theoretical values.
• The Lomb-Scargle (LS) periodogram provided a superior PSD estimate by utilizing the original, unevenly sampled data.
• Ectopy removal is crucial for accurate spectral estimation, irrespective of the technique used.
• Resampling and phantom beat replacement reduced PSD estimation accuracy, even with minimal ectopy or artefacts.
• A linear relationship was observed between ectopy/artefact frequency and PSD estimate error (mean and variance).
• On real sleep data, the LS periodogram yielded a less noisy spectral estimate of HRV.

Conclusions:

• Standard resampling methods introduce significant errors in HRV spectral analysis.
• The Lomb-Scargle periodogram is a more appropriate and accurate method for spectral estimation of HRV from irregularly sampled data.
• Accurate ectopy handling is essential for reliable HRV spectral analysis, and artificial beat replacement can degrade results.