Skip to main content
NCBI home page
Annals of Noninvasive Electrocardiology logoLink to Annals of Noninvasive Electrocardiology
. 2017 Jul 2;23(1):e12483. doi: 10.1111/anec.12483

The sensitivity of 38 heart rate variability measures to the addition of artifact in human and artificial 24‐hr cardiac recordings

Nicolas J C Stapelberg 1,2,✉, David L Neumann 3, David H K Shum 3, Harry McConnell 4, Ian Hamilton‐Craig 5
PMCID: PMC7313264  PMID: 28670841

Abstract

Background

Artifact is common in cardiac RR interval data derived from 24‐hr recordings and has a significant impact on heart rate variability (HRV) measures. However, the relative impact of progressively added artifact on a large group of commonly used HRV measures has not been assessed. This study compared the relative sensitivity of 38 commonly used HRV measures to artifact to determine which measures show the most change with increasing increments of artifact. A secondary aim was to ascertain whether short‐term and long‐term HRV measures, as groups, share similarities in their sensitivity to artifact.

Methods

Up to 10% of artifact was added to 20 artificial RR (ARR) files and 20 human cardiac recordings, which had been assessed for artifact by a cardiac technician. The added artifact simulated deletion of RR intervals and insertion of individual short RR intervals. Thirty‐eight HRV measures were calculated for each file. Regression analysis was used to rank the HRV measures according to their sensitivity to artifact as determined by the magnitude of slope.

Results

RMSSD, SDANN, SDNN, RR triangular index and TINN, normalized power and relative power linear measures, and most nonlinear methods examined are most robust to artifact.

Conclusion

Short‐term time domain HRV measures are more sensitive to added artifact than long‐term measures. Absolute power frequency domain measures across all frequency bands are more sensitive than normalized and relative frequency domain measures. Most nonlinear HRV measures assessed were relatively robust to added artifact, with Poincare plot SD1 being most sensitive.

Keywords: artifact; clinical, noninvasive techniques—Holter/event recorders; noninvasive technique—heart rate variability

1. INTRODUCTION

Heart rate varies over different timescales due to different physiological influences on heart rate, such as diurnal rhythm, Mayer waves, and respiratory activity (e.g., Berntson et al., 1997). Measures of heart rate variability (HRV) assess the beat‐to‐beat variation in heart rate over time. The measures often analyze the RR intervals, which are the intervals between successive R‐wave peaks in the cardiac signal (NASPE & T.F.o.t.E.a.t, 1996). It is recommended that HRV measures are performed on cardiac recordings collected through short‐term recordings in controlled settings (e.g., 5 min) or long‐term 24‐hr cardiac recordings (NASPE & T.F.o.t.E.a.t, 1996). Long‐term recordings are frequently used in research and applied settings, but are more likely than short‐term recordings to contain artifact.

Artifact in cardiac recordings can arise through two sources. Electrocardiography (ECG) artifact can arise from movement of the electrodes, muscle movements being misinterpreted as a cardiac signal, or line noise in the recording circuitry (Friesen et al., 1990; Manis, Alexandridi, Nikolopoulos, & Davos, 2005; Peltola, 2012). Twenty‐four‐hour cardiac recordings are usually ambulatory and can contain significant amounts of ECG artifact (Berntson, Quigley, Jang, & Boysen, 1990; Berntson & Stowell, 1998). HRV (or physiological) artifact (e.g., Peltola, 2012) can be caused by physiological cardiac events such as ventricular ectopic beats (VEBs) (Barbieri & Brown, 2006; Lippman, Stein, & Lerman, 1994; Wen & He, 2011) or arrhythmias (Sapoznikov, Luria, Mahler, & Gotsman, 1992).

HRV measures are divided up into time domain, frequency domain, and nonlinear types (NASPE & T.F.o.t.E.a.t, 1996). It is established that HRV measures are sensitive to artifact (e.g., NASPE & T.F.o.t.E.a.t, 1996; Peltola, 2012). However, the comparative sensitivity of HRV measures among several time domain, frequency domain, and nonlinear methods has not been established.

Time domain measures comprise statistical measures performed on the time series of cardiac beat‐to‐beat intervals. The measures include the SD of all RR intervals (SDNN), the square root of the mean‐squared differences of the successive RR interval (RMSSD) and geometric HRV measures, such as TINN and RR triangular index. Frequency domain HRV measures are based on calculations of the power component of the cardiac signal across a frequency spectrum. The frequency components of a cardiac RR interval series are measured by either nonparametric methods such as fast Fourier transform (FFT) or parametric methods such as autoregressive (AR) calculation. Both methods yield power for discreet frequency bands. These bands are, by convention, divided into high‐frequency (HF, 0.15–0.4 Hz), low‐frequency (LF, 0.04–0.15 Hz), very low‐frequency (VLF, 0.0033–0.04 Hz), and ultralow‐frequency (ULF, <0.0033 Hz) components (NASPE & T.F.o.t.E.a.t, 1996). HF and LF measures can be normalized, such that LF (normalized power) = LF/(LF + HF) and HF (normalized power) = HF/(LF + HF). Relative power is the ratio of either the VLF, LF, or HF power components divided by the total power (VLF + LF + HF) and expressed as a percentage.

Nonlinear HRV measures assess the complexity of variability in a cardiac recording. High complexity is generally associated with robust cardiac control by the autonomic nervous system (e.g., Rottenberg, 2007; Sassi et al., 2015) and has been linked to normal physical and even mental health. Lower complexity of HRV is linked to medical illness such as coronary heart disease (CHD) (e.g., Bigger et al., 1996). Nonlinear HRV measures include the Poincare plot, recurrence plot analysis, and nonlinear entropy measures.

Among time domain, frequency domain, and nonlinear HRV measures, several measures are highly correlated (NASPE & T.F.o.t.E.a.t, 1996; Sassi et al., 2015). Such correlations reflect the time scales used to derive the metrics. For example, RMSSD, pNN50, NN50 count, and HF power all measure changes over short time scales. These “short‐term HRV measures” are highly correlated (NASPE & T.F.o.t.E.a.t, 1996). SDNN, TINN, RR triangular index, VLF power, and LF power are also highly correlated and can be grouped as “long‐term HRV measures” (NASPE & T.F.o.t.E.a.t, 1996). Several nonlinear HRV measures are correlated with linear measures. SD1 (Poincare plot) correlates with high‐frequency power, while SD2 correlates with longer term components of HRV (Acharya, Joseph, Kannathal, Lim, & Suri, 2006; Contreras, Canetti, & Migliaro, 2007). Two measures of detrended fluctuation analysis (DFA), alpha1 and alpha2, are highly correlated with frequency domain measures such that alpha1 is approximately equal to 2LF/(LF + HF) and alpha2 is approximately equal to 2VLF/(VLF + LF) (Willson & Francis, 2003). Alpha1 is regarded as a short‐term HRV measure and alpha2 a long‐term HRV measure (Sassi et al., 2015). One implication of a correlation of HRV measures may be a shared sensitivity or robustness to artifact between highly correlated HRV measures.

2. AIM

Given the potential impact of artifact on the calculation of different HRV measures in 24‐hr ambulatory cardiac recordings, it is important to know which HRV measures are comparatively the most robust to artifact (Stapelberg, Neumann, Shum, McConnell, & Hamilton‐Craig, 2016). This study aimed to compare the relative sensitivity of 38 commonly used HRV measures to artifact. A secondary aim was to ascertain whether short‐term and long‐term HRV measures, as groups, share similarities in their sensitivity to artifact.

3. DATA AND METHOD

3.1. Artificial RR interval files and human 24‐hr cardiac recordings

Twenty artificial RR (ARR) files were generated using software written in Matlab (The MathWorks, 2012) by McSharry and colleagues (Clifford, Azuaje, & McSharry, 2006; McSharry, Clifford, Tarassenko, & Smith, 2002). These ARR files simulate a 24‐hr human recording and have been used previously in research (Aziz, Abbas, & Arif, 2005; Stapelberg et al., 2016). They have a similar structure to human RR interval recordings, leading to comparable measures of HRV between the ARR files and human recordings (Aziz et al., 2005). These ARR files are free of artifact and have recently been used to validate a preprocessing tool for RR interval recordings (Stapelberg et al., 2016). Twenty ARR files were generated using the Matlab code available at http://www.robots.ox.ac.uk/~gari/CODE/RRGEN/cinc2005/.

In addition, 20 human 24‐hr Holter recordings were used. These recordings were randomly chosen from participants in the Heart and Mind Study, a prospective cohort study examining the relationship between psychometric measures of major depressive disorder (MDD) and measures of HRV. The 20 human recordings were manually examined by a cardiac technician to annotate artifact intervals as well as normal physiological RR intervals. The files were preprocessed to remove this artifact using a preprocessing tool (described in Stapelberg et al., 2016). The preprocessing tool interpolated a mean of 84.32% (SD = 8.6%) of the technician‐assessed artifact. Overall, the preprocessing tool had a mean sensitivity across the 20 technician‐assessed cardiac recordings of 0.84 (SD = 0.09) and a specificity of 1.00 (SD = 0.01) (Stapelberg et al., 2016). The aim was to have the human recordings as free of artifact as possible, before artificially adding artifact in specified percentages.

3.2. The addition of artifact to artificial RR interval files and human 24‐hr cardiac recordings

A program was written in Matlab added a specified percentage of artifact (1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, and 10%) to each of the 20 ARR files and the 20 human recordings. The addition of artifact was achieved by adding a fixed number of ARR intervals proportionate to the total number or RR intervals in a data file. The location of artifact to be inserted was randomly selected across the entire time series. The amplitude of artifact was derived from a histogram of artifact intervals calculated from the 20 technician‐assessed human recordings as described in Stapelberg et al. (2016).

ECG artifact intervals identified by the cardiac technician fell into two groups. The first group consisted of a false R‐wave peak being introduced between two physiological R‐wave peaks. This resulted in two RR intervals of shorter duration than the surrounding physiological RR intervals. This was simulated by shortening the randomly chosen RR interval and then inserting an adjacent short interval into the time series such that the two adjacent intervals added up to the magnitude of the surrounding physiological RR intervals.

The second group of artifact consisted of very long duration RR intervals resulting from the nonidentification of one or more adjacent physiological RR intervals due to overlying artifact (e.g., motion artifact). This was simulated by converting the chosen RR interval in the time series to one of long duration. A number of adjacent physiological RR intervals which closely approximated the magnitude of the artifact interval were then removed from the RR interval series. This preserved the total length of the recording, but disrupted the time series of the recording, as occurs in reality (e.g., Peltola, 2012). Each percentage of artifact was independently added to each of the 20 ARR files and 20 human cardiac recordings. This resulted in 400 artifact‐added files, while the 40 files without added artifact were also examined.

3.3. Calculation of HRV measures and comparison of measures with added artifact

Thirty‐eight HRV measures were calculated using Kubios HRV software (Tarvainen, Niskanen, Lipponen, Ranta‐Aho, & Karjalainen, 2014). The relative change in HRV was used to examine the change in the HRV measure as a function of the amount of artifact added. Relative change in HRV (HRVR) was calculated as:

HRVA=HRVA−HRVRef/HRVRef

HRVA is the HRV result with a certain percentage of added artifact and HRVRef is the reference HRV result, without added artifact.

Linear regression was performed for model fitting in SPSS (IBMCorp, 2011). The slope of the model fit (B value of the regression) for all HRV measures was ranked according to magnitude. HRV measures for each group of measures (time domain, frequency domain, and nonlinear) were individually ranked using absolute values of B. HRV measures known to be correlated were ranked by the same method.

4. RESULTS

For the majority of HRV measures examined, the addition of increasing amounts of artifact resulted in a linear relationship when the percentage of artifact added (horizontal axis) was plotted against the relative change in HRV (vertical axis). Examples for plots with linear regression lines for a time domain measure (SDANN), a frequency domain measure (FFT HF, absolute power), and a nonlinear measure (Poincare plot SD1) are shown in Figure 1. Table 1 displays the ranked slopes (B values) for all the HRV measures. Table 1 also shows adjusted R 2 values and p values for each regression model. Table 2 shows HRV measures separated by class of HRV measure and ranked by absolute value of B.

Figure 1.

Figure 1

Open in a new tab

Examples for plots of percentage artifact added (horizontal axis) against relative difference in HRV measure (vertical axis) with linear regression for a time domain measure (SDANN), a frequency domain measure (FFT HF, absolute power), and a nonlinear measure (Poincare plot SD1)

Table 1.

HRV measures ranked by slope (B value) on linear regression

4.

Open in a new tab

Table 2.

Time domain, frequency domain, and nonlinear HRV measures ranked by absolute slope

4.

Open in a new tab

Time domain measures and frequency domain measures predominantly show an increase in variability with increasing amounts of added artifact, resulting in a positive slope. Among the time domain measures, short‐term HRV metrics such as RMSSD, pNN50, and NN50 show a greater change with added artifact (i.e., are more sensitive to added artifact) than long‐term metrics such as SDANN, SDNN, and the geometric measures TINN and RR triangular index. Among the frequency domain measures, sensitivity to added artifact is differentiated by method of calculation. Absolute power metrics of both long‐term and short‐term frequency bands show the most sensitivity to added artifact, whereas normalized and relative metrics of HF, LF, and VLF bands are more tolerant to added artifact.

Among the nonlinear HRV measures, adding increasing amounts of artifact generally results in a reduction of variability in those measures as indicated by a negative slope. Poincare plot SD1 is significantly sensitive to added artifact, whereas most other nonlinear measures assessed are relatively artifact tolerant, having slopes below 0.10. Among the nonlinear measures, the short‐term HRV measures for Poincare plot and DFA are less sensitive to added artifact than the other measures.

A minority of HRV measures did not show a linear change with increasing artifact. Correlation dimension (Figure 2) produces a marked widening of spread of data points with increasing amounts of added artifact, but with a relatively minimal slope (B = 0.019 for human data). All the RPA metrics, as well as sample entropy and DFA alpha2 produce curves rather than linear plots. While most of these have a curvature small enough to accommodate a linear curve fit with a high significance (see Table 1), some do not conform well to straight‐line fitting. The plots for two such measures, mean line length and sample entropy, are shown in Figure 2.

Figure 2.

Figure 2

Open in a new tab

Examples for plots of percentage artifact added (horizontal axis) against relative difference in HRV measure (vertical axis) for HRV measures which did not show a linear change with increasing artifact (correlation dimension, mean line length, and sample entropy)

Table 3 shows the results for correlated HRV measures in order of rank. The results fall into groups of short‐term and long‐term HRV measures. While such measures are highly correlated using data with low levels of artifact (e.g., NASPE & T.F.o.t.E.a.t, 1996; Sassi et al., 2015), the strength of the correlations show greater variation as more artifact is added due to differences in their derivation. In Table 3, HRV measures are not differentiated by whether they are time domain, frequency domain, or nonlinear.

Table 3.

Short term, long term and other HRV measures ranked by absolute slope

4.

Open in a new tab

5. DISCUSSION AND CONCLUSION

The results showed considerable variation in the sensitivity of HRV measures to the addition of artifact. In both human cardiac recordings and ARR files, long‐term time domain HRV measures were more tolerant to artifact than short‐term measures. For frequency domain measures, there is no clear differentiation between long‐ and short‐term HRV measures. However, normalized HF and LF measures, VLF relative power, and LF/HF ratio are the most artifact tolerant for FFT and AR measures, whereas absolute power metrics are the most sensitive to added artifact for all frequency bands.

Most nonlinear methods show lower values with added artifact, producing a negative slope. RPA measures, sample entropy, and DFA alpha2 produce curves which flatten with increasing artifact or may even have a U‐shaped quadratic curve beyond 10% added artifact (see Figure 2). While the derivation of these measures differs markedly (see Acharya et al., 2006; NASPE & T.F.o.t.E.a.t, 1996; Tarvainen & Niskanen, 2008; Voss, Schulz, Schroeder, Baumert, & Caminal, 2009), they reflect a loss of time series complexity with the increased addition of artifact.

Despite the significant differences in the way HRV measures are calculated, it is useful to have a ranking of different HRV measures according to their sensitivity to artifact for applications such as ambulatory 24‐hr Holter recordings (e.g., Rand et al., 2007) or recordings where participants have high levels of physical activity (Citi, Brown, & Barbieri, 2012). These results are also significant in the context of the proliferation of wearable electronic devices which frequently record ambulatory biological signals with a single sensor, potentially giving rise to significant artifact in the recorded data (e.g., Heathers, 2013; Hegde, Kumar, Rai, Mathur, & Varadan, 2012; Muaremi, Arnrich, & Tröster, 2013). In cases where artifact is likely to be present, it is recommended that HRV measures robust to artifact are used.

5.1. The sensitivity of different HRV measures to the types of added artifact

Two types of artifact were added to cardiac recordings and ARR files. VEBs result in a short duration RR interval, followed by a longer compensatory RR interval. VEBs were not simulated in this study. However, the short duration RR intervals that were introduced share some similarity with the initial shortened RR interval which occurs in VEBs, and so similar changes to some HRV measures are possible, especially with regard to short‐term HRV measures. The sensitivity of statistical time domain and all frequency domain HRV measures to artifact has previously been established, especially the sensitivity of short‐term HRV measures to artifacts and editing (NASPE & T.F.o.t.E.a.t, 1996; Peltola, 2012). Factors arising from assumptions in the calculation of certain HRV measures can also be significant, such as the effect of nonstationarities on frequency domain power calculations.

5.2. The effects of time series disruption on HRV measures

Time series derived from cardiac control systems show complex structures on multiple spatiotemporal scales (e.g., Costa, Goldberger, & Peng, 2002). Disruption to the RR interval time series by artifact changes these spatiotemporal cycles or structures. Artifact arising from the nonidentification of R‐wave peaks due to motion artifact (e.g., Friesen et al., 1990) can result in large artifact intervals in a cardiac recording, followed by normally detected physiological RR intervals. Missed intervals are lost in the resultant RR interval series, resulting in a “frameshift” of all resultant recorded intervals (e.g., Citi et al., 2012).

Deletion of RR intervals, as well as insertion of artifact intervals introduces step‐like shapes into RR interval time series, with abrupt changes in the beat‐to‐beat variability between the intervals adjacent to the missing RR interval segments (Peltola, 2012; Salo, Huikuri, & Seppanen, 2001). This affects short‐term time domain HRV measures, such as NN50, pNN50, and RMSSD. Sudden transitions introduced by deletions will falsely increase all three measures. As shown here, long‐term time domain HRV measures are more tolerant to artifact than short‐term measures, which is consistent with the literature (e.g., Peters, Vullings, Bergmans, Oei, & Wijn, 2008; Salo et al., 2001).

Among frequency domain measures, VEBs introduce sharp transients into the time series, erroneously increasing HF power (Mateo & Laguna, 2003; Peltola, 2012) and a sensitivity of both HF and LF absolute power to increased artifact was shown by Lippman et al. (1994). In the results presented here, absolute power of HF and LF increased significantly, whereas VLF absolute power was less affected. Despite these differences, absolute power was disrupted across all (long‐term and short‐term) frequency bands. Normalized power and relative power were robust to added artifact. Normalized HF power is calculated in relation to LF power and normalized LF power in relation to VLF power. Given changes across all frequency bands, some of the changes should cancel each other out, resulting in a smaller B value. Normalized HF power should still have a greater slope than normalized LF power due to the greater sensitivity of the HF band. This is observed in human and ARR data, although the differences in slope are small. Normalized spectral HRV measures are algebraically redundant with respect to each other and with respect to the LF/HF ratio (Burr, 2007). Relative power metrics are derived by dividing HF, LF, and VLF by the total spectral power. As with normalized power, relative power has a reduced slope due to changes across the frequency spectrum canceling each other out to some extent.

The presence of VEBs reduces DFA alpha1, and approximate entropy (Tarkiainen et al., 2007). In our study, DFA alpha1 and approximate entropy were both reduced by the addition of artifact. Tarkiainen et al. (2007) showed that Poincare plot SD1 and SD2 increased with the presence of VEBs. Similarly, our results show increases in both SD1 and SD2 with added artifact. When the VEBs were deleted, disrupting the time series (Tarkiainen et al., 2007), alpha1 was increased relative to the interpolated recordings, while SD1 and SD2 were reduced. Similarly, deletion or RR intervals resulted in a false increase in alpha1 in cardiac patients (Peltola, Seppänen, Mäkikallio, & Huikuri, 2004).

Twenty‐four‐hour cardiac recordings are subject to slow linear trends known as nonstationarities (Peltola, 2012; Tarvainen, Ranta‐aho, & Karjalainen, 2002) (Berntson et al., 1997). Frequency domain calculations inherently assume that the signal is at least weakly stationary (Tarvainen et al., 2002) and thus nonstationarity of an RR interval series can significantly effect frequency domain calculations. Both the FFT power spectrum (Peltola, 2012) and AR spectrum (Tarvainen et al., 2002) are affected. Nonstationarities can lead to a relative increase in the power of the VLF band, and an underestimation of power in the HF band (Tarvainen et al., 2002). In the present study, no detrending was performed and it is possible that the results obtained for frequency domain measures are in part due to the effects of nonstationarities.

5.3. Biological implications

Biologically, short‐term components of HRV are thought to reflect vagal function (Akselrod, Gordon, Ubel, Shannon, Berger, & Cohen, 1981; Malliani, Pagani, Lombardi, & Cerutti, 1991; Pomeranz et al., 1985), although this relationship has been shown to be more complex than originally proposed (e.g., Pyetan & Akselrod, 2003; Rottenberg, 2007). Long‐term HRV measures are thought to represent a mixture of sympathetic and parasympathetic activity (Akselrod, Gordon, & Ubel, 1981; Appel, Berger, Saul, Smith, & Cohen, 1989). Our findings that short‐term time domain measures are more sensitive to artifact than long‐term measures and that absolute power calculations in the frequency domain have significant sensitivity to artifact are significant, as both time domain and frequency domain measures have been extensively used in studies as biomarkers for general medical conditions as well as mental illnesses. For example, vagal function has been shown to be important in illnesses such as MDD (e.g., Stapelberg, Neumann, Shum, McConnell, & Hamilton‐Craig, 2011, 2015). Short‐term time domain HRV measures have also been used to examine the relationship between HRV and MDD in 24‐hr recordings, for example, pNN50 (Sayar, Güleç, Gökçe, & Ak, 2002) or RMSSD (Boettger et al., 2008; Sayar et al., 2002). Using these measures derived from 24‐hr ambulatory cardiac data may introduce error into the results obtained.

Absolute power frequency domain measures have also been used in studies of 12‐ to 24‐hr cardiac recordings using HRV as predictors of adverse events in medical illnesses, such as cardiac events in CHD (e.g., Bigger et al., 1992, 1995; Stein, Domitrovich, Huikuri, & Kleiger, 2005). Similar studies also used short‐term time domain measures sensitive to artifact such as RMSSD (e.g., Stein et al., 2005; Zuanetti et al., 1996).

Several studies have used HRV to examine links between MDD and CHD using 24‐hr cardiac recordings. Among these are studies which have employed short‐term time domain measures, for example, PNN50 (Carney et al., 1995; Stein et al., 2000) or RMSSD (Carney et al., 1995; Martens, Nyklíček, Szabo, & Kupper, 2008; Stein et al., 2000), as well as absolute power frequency domain measures (Gehi, Mangano, Pipkin, Browner, & Whooley, 2005; Martens et al., 2008; Stein et al., 2000). Given their widespread use in research and their biological importance, recommendations for the selection of HRV measures in 24‐hr recordings are as follows:

  • SDANN, and possibly SDNN, along with the geometric measures RR triangular index and TINN, should be used as long‐term measures of time domain HRV. RMSSD is recommended for use as a short‐term time domain HRV measure above pNN50 and NN50. This recommendation coincides with the recommendation by the Task Force of the European Society of Cardiology & the North American Society of Pacing and Electrophysiology to use four “standard measures” of time domain together (SDNN, SDANN, RMSSD, and HRV triangular index) (NASPE & T.F.o.t.E.a.t, 1996).

  • Among the frequency domain HRV measures, normalized power measures and relative power measures are recommended above measures of absolute power for both FFT and AR calculation methods. The LF/HF ratio is relatively tolerant to artifact. The findings of this study suggest that neither FFT nor AR is superior in terms of tolerance to artifact.

  • Most of the nonlinear methods examined here are tolerant to artifact. However, caution is advised in interpreting the SD1 measure of the Poincare plot, as well as recurrence rate and possibly correlation dimension due to their greater sensitivity to artifact.

Finally, the selection of HRV measures which are more robust to artifact should be considered in the context of methods to remove artifact from cardiac recordings. Human assessment of cardiac recordings to remove artifact remains the gold standard, but is error prone and dependent on the level of skill of the assessor (Berntson et al., 1990). Automated methods may either miss some artifact or may interpolate some normal physiological RR intervals, effectively altering the cardiac recording. Robust HRV measures would be less sensitive to any remaining artifact after preprocessing and may also be less sensitive to interpolation of normal physiological intervals during preprocessing. It is not suggested that reprocessing of cardiac recordings should be ignored if a robust HRV measure is used, but the use of good artifact correction methods with HRV measures known to be robust to artifact.

5.4. Study limitations

While most fitted linear models had a level of significance below 0.05 for the regression equation on ANOVA, three regressions among the ARR files did not reach significance. In some cases this is due to the regression analysis method, with little or no change with increasing percentages of added artifact and a significant spread of data points resulting in little or no relationship between the IV and the DV. As noted earlier, some HRV measures show a curved distribution with increasing artifact, but these plots all showed a significance below 0.05 on linear regression. A straight‐line fit was used for all HRV metrics as it described most of the data well and provided consistency across measures.

Slow linear or complex nonstationaries, discussed earlier, can influence the calculation of frequency domain measures and reference HRV values did not use detrending for nonstationaries. Therefore, reference values of VLF, LF, and HF could be distorted (Tarvainen et al., 2002). However, not all HRV measures (e.g., nonlinear measures) require detrending, and detrending may have biased results for HRV calculations with added artifact. Thus, no detrending was performed on any measures to maintain consistency throughout the study.

A further limitation of the study is the selection of HRV measures used, which are those available in the Kubios HRV software. We acknowledge that several relevant HRV measures, or categories of HRV measure, are missing, such as measures based on symbolic dynamics (Kurths et al., 1995; Porta et al., 2001; Voss, Kurths, Kleiner, Witt, & Wessel, 1995; Voss et al., 1996; Wessel et al., 2000), multiscale entropy (Costa et al., 2002), nonlinear prediction (Fortrat, Yamamoto, & Hughson, 1997; Porta et al., 2007; Sugihara, Allan, Sobel, & Allan, 1996), or time irreversibility analysis (Porta et al., 2008). These HRV metrics should be assessed for their sensitivity to artifact in a future study.

CONFLICTS OF INTEREST

The authors declare that they have no conflict of interest.

ACKNOWLEDGMENTS

We gratefully acknowledge the assistance of Mr Luke Shanahan, Advanced Cardiac Technician and Cardiac Scientist, Gold Coast Hospital and Health Service.

Stapelberg NJC, Neumann DL, Shum DHK, McConnell H, Hamilton‐Craig I. The sensitivity of 38 heart rate variability measures to the addition of artifact in human and artificial 24‐hr cardiac recordings. Ann Noninvasive Electrocardiol. 2018;23:e12483 10.1111/anec.12483

REFERENCES

  1. Acharya, U. R. , Joseph, K. P. , Kannathal, N. , Lim, C. M. , & Suri, J. S. (2006). Heart rate variability: A review. Medical and Biological Engineering and Computing, 44(12), 1031–1051. 10.1007/978-3-540-36675-1_5 [DOI] [PubMed] [Google Scholar]
  2. Akselrod, S. , Gordon, D. , & Ubel, F. A. (1981). Power spectrum analysis of heart rate fluctuation: A quantitative probe of beat‐to‐beat cardiovascular control. Science, 213(4504), 220–222. [DOI] [PubMed] [Google Scholar]
  3. Akselrod, S. , Gordon, D. , Ubel, F. A. , Shannon, D. C. , Berger, A. , & Cohen, R. J. (1981). Power spectrum analysis of heart rate fluctuation: A quantitative probe of beat‐to‐beat cardiovascular control. Science, 213(4504), 220–222. [DOI] [PubMed] [Google Scholar]
  4. Appel, M. L. , Berger, R. D. , Saul, J. P. , Smith, J. M. , & Cohen, R. J. (1989). Beat to beat variability in cardiovascular variables: Noise or music? Journal of the American College of Cardiology, 14(5), 1139–1148. [DOI] [PubMed] [Google Scholar]
  5. Aziz, W. , Abbas, R. , & Arif, M. (2005). Detrended Fluctuation Analysis of Synthetic and Real RR‐interval time Series of Healthy Subjects. Journal of Applied and Emerging Science, 1(2), 58–63. [Google Scholar]
  6. Barbieri, R. , & Brown, E.N. (2006). Correction of erroneous and ectopic beats using a point process adaptive algorithm. Paper presented at the Engineering in Medicine and Biology Society, 2006. EMBS'06. 28th Annual International Conference of the IEEE. 10.1109/iembs.2006.260325 [DOI] [PubMed]
  7. Berntson, G. G. , Bigger, J. T. Jr , Eckberg, D. L. , Grossman, P. , Kaufmann, P. G. , Malik, M. , … van der Molen, M. W. (1997). Heart rate variability: Origins, methods, and interpretive caveats. Psychophysiology, 34(6), 623–648. 10.1111/j.1469-8986.1997.tb02140.x [DOI] [PubMed] [Google Scholar]
  8. Berntson, G. G. , Quigley, K. S. , Jang, J. F. , & Boysen, S. T. (1990). An approach to artifact identification: Application to heart period data. Psychophysiology, 27(5), 586–598. 10.1111/j.1469-8986.1990.tb01982.x [DOI] [PubMed] [Google Scholar]
  9. Berntson, G. G. , & Stowell, J. R. (1998). ECG artifacts and heart period variability: Don't miss a beat!. Psychophysiology, 35(1), 127–132. [PubMed] [Google Scholar]
  10. Bigger, J. T. Jr , Fleiss, J. L. , Steinman, R. C. , Rolnitzky, L. M. , Kleiger, R. E. , & Rottman, J. N. (1992). Frequency domain measures of heart period variability and mortality after myocardial infarction. Circulation, 85(1), 164–171. [DOI] [PubMed] [Google Scholar]
  11. Bigger, J. T. , Fleiss, J. L. , Steinman, R. C. , Rolnitzky, L. M. , Schneider, W. J. , & Stein, P. K. (1995). RR variability in healthy, middle‐aged persons compared with patients with chronic coronary heart disease or recent acute myocardial infarction. Circulation, 91(7), 1936–1943. [DOI] [PubMed] [Google Scholar]
  12. Bigger, J. T. Jr , Steinman, R. C. , Rolnitzky, L. M. , Fleiss, J. L. , Albrecht, P. , & Cohen, R. J. (1996). Power law behavior of RR‐interval variability in healthy middle‐aged persons, patients with recent acute myocardial infarction, and patients with heart transplants. Circulation, 93(12), 2142–2151. [DOI] [PubMed] [Google Scholar]
  13. Boettger, S. , Hoyer, D. , Falkenhahn, K. , Kaatz, M. , Yeragani, V. K. , & Bar, K. J. (2008). Nonlinear broad band dynamics are less complex in major depression. Bipolar Disorders, 10(2), 276–284. [DOI] [PubMed] [Google Scholar]
  14. Burr, R. L. (2007). Interpretation of normalized spectral heart rate variability indices in sleep research: A critical review. Sleep, 30(7), 913. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Carney, R. M. , Saunders, R. D. , Freedland, K. E. , Stein, P. , Rich, M. W. , & Jaffe, A. S. (1995). Association of depression with reduced heart rate variability in coronary artery disease. American Journal of Cardiology, 76(8), 562–564. 10.1016/s0002-9149(99)80155-6 [DOI] [PubMed] [Google Scholar]
  16. Citi, L. , Brown, E. N. , & Barbieri, R. (2012). A real‐time automated point‐process method for the detection and correction of erroneous and ectopic heartbeats. IEEE Transactions on Biomedical Engineering, 59(10), 2828–2837. 10.1109/tbme.2012.2211356 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Clifford, G. D. , Azuaje, F. , & McSharry, P. (2006). Advanced methods and tools for ECG data analysis. Norwood, MA: Artech House Inc; 10.1186/1475-925x-6-18 [DOI] [Google Scholar]
  18. Contreras, P. , Canetti, R. , & Migliaro, E. R. (2007). Correlations between frequency‐domain HRV indices and lagged Poincaré plot width in healthy and diabetic subjects. Physiological measurement, 28(1), 85. [DOI] [PubMed] [Google Scholar]
  19. Costa, M. , Goldberger, A. L. , & Peng, C.‐K. (2002). Multiscale entropy analysis of complex physiologic time series. Physical review letters, 89(6), 068102. [DOI] [PubMed] [Google Scholar]
  20. Fortrat, J.‐O. , Yamamoto, Y. , & Hughson, R. L. (1997). Respiratory influences on non‐linear dynamics of heart rate variability in humans. Biological cybernetics, 77(1), 1–10. [DOI] [PubMed] [Google Scholar]
  21. Friesen, G. M. , Jannett, T. C. , Jadallah, M. A. , Yates, S. L. , Quint, S. R. , & Nagle, H. T. (1990). A comparison of the noise sensitivity of nine QRS detection algorithms. IEEE Transactions on Biomedical Engineering, 37(1), 85–98. 10.1109/10.43620 [DOI] [PubMed] [Google Scholar]
  22. Gehi, A. , Mangano, D. , Pipkin, S. , Browner, W. S. , & Whooley, M. A. (2005). Depression and heart rate variability in patients with stable coronary heart disease: Findings from the heart and soul study. Archives of general psychiatry, 62(6), 661–666. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Heathers, J. A. (2013). Smartphone‐enabled pulse rate variability: An alternative methodology for the collection of heart rate variability in psychophysiological research. International Journal of Psychophysiology, 89(3), 297–304. [DOI] [PubMed] [Google Scholar]
  24. Hegde, S. , Kumar, P.S. , Rai, P. , Mathur, G.N. , & Varadan, V.K. (2012). Music close to one's heart: heart rate variability with music, diagnostic with e‐bra and smartphone. Paper presented at the SPIE Smart Structures and Materials+ Nondestructive Evaluation and Health Monitoring.
  25. IBMCorp. (2011). IBM SPSS Statistics for Windows, Version 20.0. Armonk, NY: IBM Corp. [Google Scholar]
  26. Kurths, J. , Voss, A. , Saparin, P. , Witt, A. , Kleiner, H. , & Wessel, N. (1995). Quantitative analysis of heart rate variability. Chaos: An Interdisciplinary Journal of Nonlinear Science, 5(1), 88–94. [DOI] [PubMed] [Google Scholar]
  27. Lippman, N. , Stein, K. M. , & Lerman, B. B. (1994). Comparison of methods for removal of ectopy in measurement of heart rate variability. American Journal of Physiology‐Heart and Circulatory Physiology, 267(1), H411–H418. [DOI] [PubMed] [Google Scholar]
  28. Malliani, A. , Pagani, M. , Lombardi, F. , & Cerutti, S. (1991). Cardiovascular neural regulation explored in the frequency domain. Circulation, 84(2), 482–492. [DOI] [PubMed] [Google Scholar]
  29. Manis, G. , Alexandridi, A. , Nikolopoulos, S. , & Davos, K. (2005). The Effect of White Noise and False Peak Detection on HRV Analysis. Paper presented at the International Workshop on Biosignal Processing and Classification, 10.5220/0001195301610166 [DOI]
  30. Martens, E. , Nyklíček, I. , Szabo, B. , & Kupper, N. (2008). Depression and anxiety as predictors of heart rate variability after myocardial infarction. Psychological medicine, 38(03), 375–383. [DOI] [PubMed] [Google Scholar]
  31. Mateo, J. , & Laguna, P. (2003). Analysis of heart rate variability in the presence of ectopic beats using the heart timing signal. IEEE Transactions on Biomedical Engineering, 50(3), 334–343. 10.1109/tbme.2003.808831 [DOI] [PubMed] [Google Scholar]
  32. McSharry, P. E. , Clifford, G. , Tarassenko, L. , & Smith, L. A. (2002). Method for generating an artificial RR tachogram of a typical healthy human over 24‐hours. New York: IEEE; 10.1109/cic.2002.1166748 [DOI] [Google Scholar]
  33. Muaremi, A. , Arnrich, B. , & Tröster, G. (2013). Towards measuring stress with smartphones and wearable devices during workday and sleep. BioNanoScience, 3(2), 172–183. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. NASPE. , & T.F.o.t.E.a.t . (1996). Heart rate variability: Standards of measurement, physiological interpretation and clinical use. Task Force of the European Society of Cardiology & the North American Society of Pacing and Electrophysiology. Circulation, 93(5), 1043–1065. 10.1161/01.cir.93.5.1043 [DOI] [PubMed] [Google Scholar]
  35. Peltola, M. A. (2012). Role of editing of R–R intervals in the analysis of heart rate variability. Frontiers in physiology, 3, 1–10. 10.3389/fphys.2012.00148 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Peltola, M. A. , Seppänen, T. , Mäkikallio, T. H. , & Huikuri, H. V. (2004). Effects and Significance of Premature Beats on Fractal Correlation Properties of R‐R Interval Dynamics. Annals of noninvasive electrocardiology, 9(2), 127–135. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Peters, C. , Vullings, R. , Bergmans, J. , Oei, G. , & Wijn, P. (2008). The effect of artifact correction on spectral estimates of heart rate variability. Paper presented at the Engineering in Medicine and Biology Society, 2008. EMBS 2008. 30th Annual International Conference of the IEEE. 10.1109/iembs.2008.4649751 [DOI] [PubMed]
  38. Pomeranz, B. , Macaulay, R. , Caudill, M. A. , Kutz, I. , Adam, D. , Gordon, D. , … Cohen, R. J. (1985). Assessment of autonomic function in humans by heart rate spectral analysis. American Journal of Physiology‐Heart and Circulatory Physiology, 248(1), H151–H153. [DOI] [PubMed] [Google Scholar]
  39. Porta, A. , Casali, K. R. , Casali, A. G. , Gnecchi‐Ruscone, T. , Tobaldini, E. , Montano, N. , … Van Leeuwen, P. (2008). Temporal asymmetries of short‐term heart period variability are linked to autonomic regulation. American Journal of Physiology‐Regulatory, Integrative and Comparative Physiology, 295(2), R550–R557. [DOI] [PubMed] [Google Scholar]
  40. Porta, A. , Guzzetti, S. , Furlan, R. , Gnecchi‐Ruscone, T. , Montano, N. , & Malliani, A. (2007). Complexity and nonlinearity in short‐term heart period variability: Comparison of methods based on local nonlinear prediction. IEEE Transactions on Biomedical Engineering, 54(1), 94–106. [DOI] [PubMed] [Google Scholar]
  41. Porta, A. , Guzzetti, S. , Montano, N. , Furlan, R. , Pagani, M. , Malliani, A. , & Cerutti, S. (2001). Entropy, entropy rate, and pattern classification as tools to typify complexity in short heart period variability series. IEEE Transactions on Biomedical Engineering, 48(11), 1282–1291. [DOI] [PubMed] [Google Scholar]
  42. Pyetan, E. , & Akselrod, S. (2003). Do the high‐frequency indexes of HRV provide a faithful assessment of cardiac vagal tone? A critical theoretical evaluation. IEEE Transactions on Biomedical Engineering, 50(6), 777–783. [DOI] [PubMed] [Google Scholar]
  43. Rand, J. , Hoover, A. , Fishel, S. , Moss, J. , Pappas, J. , & Muth, E. (2007). Real‐time correction of heart interbeat intervals. IEEE Transactions on Biomedical Engineering, 54(5), 946–950. 10.1109/tbme.2007.893491 [DOI] [PubMed] [Google Scholar]
  44. Rottenberg, J. (2007). Cardiac vagal control in depression: A critical analysis. Biological psychology, 74(2), 200–211. 10.1016/j.biopsycho.2005.08.010 [DOI] [PubMed] [Google Scholar]
  45. Salo, M. A. , Huikuri, H. V. , & Seppanen, T. (2001). Ectopic beats in heart rate variability analysis: Effects of editing on time and frequency domain measures. Annals of noninvasive electrocardiology, 6(1), 5–17. 10.1111/j.1542-474x.2001.tb00080.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Sapoznikov, D. , Luria, M. H. , Mahler, Y. , & Gotsman, M. S. (1992). Computer processing of artifact and arrhythmias in heart rate variability analysis. Computer methods and programs in biomedicine, 39(1), 75–84. 10.1016/0169-2607(92)90060-k [DOI] [PubMed] [Google Scholar]
  47. Sassi, R. , Cerutti, S. , Lombardi, F. , Malik, M. , Huikuri, H. V. , Peng, C.‐K. , … Lip, G.H. (2015). Advances in heart rate variability signal analysis: Joint position statement by the e‐Cardiology ESC Working Group and the European Heart Rhythm Association co‐endorsed by the Asia Pacific Heart Rhythm Society. EP Europace, 17(9), 1341–1353. [DOI] [PubMed] [Google Scholar]
  48. Sayar, K. , Güleç, H. , Gökçe, M. , & Ak, I. (2002). Heart rate variability in depressed patients. Bulletin of Clinical Psychopharmacology, 12(3), 130–133. [Google Scholar]
  49. Stapelberg, N. J. , Neumann, D. L. , Shum, D. H. , McConnell, H. , & Hamilton‐Craig, I. (2011). A topographical map of the causal network of mechanisms underlying the relationship between major depressive disorder and coronary heart disease. Australian and New Zealand Journal of Psychiatry, 45(5), 351–369. 10.3109/00048674.2011.570427 [DOI] [PubMed] [Google Scholar]
  50. Stapelberg, N. J. , Neumann, D. L. , Shum, D. H. , McConnell, H. , & Hamilton‐Craig, I. (2015). From Physiome to Pathome: A Systems Biology Model of Major Depressive Disorder and the Psycho‐Immune‐Neuroendocrine Network. Current Psychiatry Reviews, 11(1), 32–62. [Google Scholar]
  51. Stapelberg, N. J. , Neumann, D. L. , Shum, D. H. , McConnell, H. , & Hamilton‐Craig, I. (2016). A preprocessing tool for removing artifact from cardiac RR‐interval recordings using three‐dimensional spatial distribution mapping. Psychophysiology, 53(4), 482–492. [DOI] [PubMed] [Google Scholar]
  52. Stein, P. K. , Carney, R. M. , Freedland, K. E. , Skala, J. A. , Jaffe, A. S. , Kleiger, R. E. , & Rottman, J. N. (2000). Severe depression is associated with markedly reduced heart rate variability in patients with stable coronary heart disease. Journal of Psychosomatic Research, 48(4–5), 493–500. [DOI] [PubMed] [Google Scholar]
  53. Stein, P. K. , Domitrovich, P. P. , Huikuri, H. V. , & Kleiger, R. E. (2005). Traditional and nonlinear heart rate variability are each independently associated with mortality after myocardial infarction. Journal of cardiovascular electrophysiology, 16(1), 13–20. [DOI] [PubMed] [Google Scholar]
  54. Sugihara, G. , Allan, W. , Sobel, D. , & Allan, K. D. (1996). Nonlinear control of heart rate variability in human infants. Proceedings of the National Academy of Sciences, 93(6), 2608–2613. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Tarkiainen, T. H. , Kuusela, T. A. , Tahvanainen, K. U. , Hartikainen, J. E. , Tiittanen, P. , Timonen, K. L. , & Vanninen, E. J. (2007). Comparison of methods for editing of ectopic beats in measurements of short‐term non‐linear heart rate dynamics. Clinical Physiology and Functional Imaging, 27(2), 126–133. [DOI] [PubMed] [Google Scholar]
  56. Tarvainen, M.P. , & Niskanen, J.‐P. (2008). Kubios HRV version 2.0 user's guide. Kuopio, Finland: Department of Physics, University of Kuopio; http://kubios.uef.fi/media/Kubios_HRV_Users_Guide.pdf [Google Scholar]
  57. Tarvainen, M. P. , Niskanen, J.‐P. , Lipponen, J. A. , Ranta‐Aho, P. O. , & Karjalainen, P. A. (2014). Kubios HRV–Heart rate variability analysis software. Computer methods and programs in biomedicine, 113(1), 210–220. 10.1016/j.cmpb.2013.07.024 [DOI] [PubMed] [Google Scholar]
  58. Tarvainen, M. P. , Ranta‐aho, P. O. , & Karjalainen, P. A. (2002). An advanced detrending method with application to HRV analysis. IEEE Transactions on Biomedical Engineering, 49(2), 172–175. [DOI] [PubMed] [Google Scholar]
  59. The MathWorks . (2012). MATLAB and Statistics Toolbox Release 2012b. Natick, MA: The MathWorks. [Google Scholar]
  60. Voss, A. , Kurths, J. , Kleiner, H. , Witt, A. , & Wessel, N. (1995). Improved analysis of heart rate variability by methods of nonlinear dynamics. Journal of Electrocardiology, 28, 81–88. [DOI] [PubMed] [Google Scholar]
  61. Voss, A. , Kurths, J. , Kleiner, H. , Witt, A. , Wessel, N. , Saparin, P. , … Dietz, R. (1996). The application of methods of non‐linear dynamics for the improved and predictive recognition of patients threatened by sudden cardiac death. Cardiovascular Research, 31(3), 419–433. [PubMed] [Google Scholar]
  62. Voss, A. , Schulz, S. , Schroeder, R. , Baumert, M. , & Caminal, P. (2009). Methods derived from nonlinear dynamics for analysing heart rate variability. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 367(1887), 277–296. 10.1098/rsta.2008.0232 [DOI] [PubMed] [Google Scholar]
  63. Wen, F. , & He, F.‐T. (2011). An efficient method of addressing ectopic beats: New insight into data preprocessing of heart rate variability analysis. Journal of Zhejiang University SCIENCE B, 12(12), 976–982. 10.1631/jzus.b1000392 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Wessel, N. , Ziehmann, C. , Kurths, J. , Meyerfeldt, U. , Schirdewan, A. , & Voss, A. (2000). Short‐term forecasting of life‐threatening cardiac arrhythmias based on symbolic dynamics and finite‐time growth rates. Physical Review E, 61(1), 733. [DOI] [PubMed] [Google Scholar]
  65. Willson, K. , & Francis, D. P. (2003). A direct analytical demonstration of the essential equivalence of detrended fluctuation analysis and spectral analysis of RR‐interval variability. Physiological measurement, 24(1), N1. [DOI] [PubMed] [Google Scholar]
  66. Zuanetti, G. , Neilson, J. M. , Latini, R. , Santoro, E. , Maggioni, A. P. , & Ewing, D. J. (1996). Prognostic Significance of Heart Rate Variability in Post–Myocardial Infarction Patients in the Fibrinolytic Era The GISSI‐2 Results. Circulation, 94(3), 432–436. [DOI] [PubMed] [Google Scholar]

Articles from Annals of Noninvasive Electrocardiology are provided here courtesy of International Society for Holter and Noninvasive Electrocardiology, Inc. and Wiley Periodicals, Inc.

RESOURCES