# Effects of Missing Data on Heart Rate Variability Metrics

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

#### 1.1. Related Work

#### 1.2. Aims of the Study

## 2. Materials and Methods

#### 2.1. Simulation of Missing Beats

#### 2.2. Apple Watch Dataset

#### 2.3. Missing Data Detection

#### 2.4. Correction Methods

#### 2.5. HRV Metrics

- Time domain: Mean heart rate (MHR), standard deviation of the normal-to-normal interval (SDNN) and root mean square of successive differences (RMSSD), as described in [1].
- Frequency domain: LF and HF powers (${P}_{LF}$,${P}_{HF}$); LF power measured in normalized units (${P}_{LFn}$); and ${P}_{LF}$/${P}_{HF}$ ratio. Only relative errors of ${P}_{LF}$ and ${P}_{HF}$ are presented, as the other two are derived from them. While all subjects are included when measuring relative errors, not all of them could be included when measuring the ability to distinguish sympathovagal balance. For this comparison, only subjects with respiratory rates above the classic LF band (>0.15 Hz) were selected, thus allowing a correct frequency component separation [34]. Therefore, simulation dataset is reduced from 16 to 9 subjects (age $28.3\pm 2.6$ years, 5 males). This selection only applies when comparing metrics in the frequency domain. No selection is made in the Apple Watch dataset. In addition, respiratory rate does not exceed 0.4 Hz—the classic HF band upper limit—in any case.Spectral estimation is performed via Fast Fourier Transform (FFT) and Lomb’s methods. FFT estimations are made on the evenly-sampled instantaneous heart rate signal, $r\left(t\right)$, obtained from the IPFM model [17]. This model assumes the ANS modulates the sinoatrial node by a band-limited zero-mean signal [35]. In [36], it is shown that spectra derived from $r\left(t\right)$ are a more accurate estimator for HRV than spectra derived from evenly-sampled interval series, avoiding spurious components and low-pass filtering effects. Welch’s method is used for periodogram averaging using 60 s Hamming windows with 50% overlap. For 120 s signals, three periodograms are averaged. Powers are computed using trapezoidal integration and classic windows (0.04–0.15 Hz for LF and 0.15–0.4 Hz for HF). This FFT-based approach has been tested using both model-based and gap-filling correction.On the other hand, Lomb’s periodograms can be computed from unevenly spaced signals, even in the presence of missing beats. Therefore, this method has been tested both using $\mathcal{OR}$ and gap-filling correction. It is demonstrated that the estimates on the heart rate representations are more accurate than on the beat interval representations [36]; therefore, Lomb’s periodograms are computed on the inverse interval function$${d}_{IIF}\left({t}_{k}\right)=\sum _{k=1}^{K}\frac{1}{({t}_{k}-{t}_{k-1})}\delta (t-{t}_{k})$$
- Poincaré plots: SD1, SD2, SD1/SD2, ellipse area ($S=\pi \xb7SD1\xb7SD2$), mean distance to the ellipse centroid (Md) and standard deviation to the ellipse centroid (Sd) have been computed using the ellipse fitting method [37]. As S and SD1/SD2 are computed from SD1 and SD2, relative errors are not shown for these metrics. The reliability of Poincaré plots in ultra-short term segments—less than 5 min, as this case—has been demonstrated recently [38].

#### 2.6. Statistical Analysis

## 3. Results

#### 3.1. Time-Domain Metrics

#### 3.2. Frequency-Domain Metrics Computed via FFT

#### 3.3. Frequency-Domain Metrics Computed via Lomb’s Method

#### 3.4. Poincaré Plots

## 4. Discussion

#### Limitations

## 5. Conclusions

## Author Contributions

## Funding

## Informed Consent Statement

## Acknowledgments

## Conflicts of Interest

## Abbreviations

HRV | Heart rate variability |

PRV | Pulse rate variability |

ANS | Autonomic nervous system |

ECG | Electrocardiography |

PPG | Pulse photoplethysmography |

IBI | Inter-beat interval |

SNR | Signal-to-noise ratio |

OD | Outlier detection |

OR | Outlier removal |

L | Linear (correction method) |

NL | Non-linear (correction method) |

M | Model-based (correction method) |

## References

- Camm, A.J.; Malik, M.; Bigger, J.T.; Brethardt, G.; Cerutti, S.; Cohen, R.J.; Coumel, P.; Fallen, E.L.; Kennedy, H.L.; Kleiger, R.E. Heart rate variability: Standards of measurement, physiological interpretation, and clinical use. Circulation
**1996**, 93, 1043–1065. [Google Scholar] - Casolo, G.C.; Stroder, P.; Signorini, C.; Calzolari, F.; Zucchini, M.; Balli, E.; Lazzerini, S. Heart rate variability during the acute phase of myocardial infarction. Circulation
**1992**, 85, 2073–2079. [Google Scholar] [CrossRef] [PubMed][Green Version] - Pagani, M.; Malfatto, G.; Pierini, S.; Casati, R.; Masu, A.M.; Poli, M.; Malliani, A. Spectral analysis of heart rate variability in the assessment of autonomic diabetic neuropathy. J. Auton. Nerv. Syst.
**1988**, 23, 143–153. [Google Scholar] [CrossRef] - Guzzetti, S.; Piccaluga, E.; Casati, R.; Cerutti, S.; Lombardi, F.; Pagani, M.; Malliani, A. Sympathetic predominance in essential hypertension: A study employing spectral analysis of heart rate variability. J. Hypertens.
**1988**, 6, 711–717. [Google Scholar] [CrossRef] - Malliani, A.; Pagani, M.; Lombardi, F.; Cerutti, S. Cardiovascular neural regulation explored in the frequency domain. Circulation
**1991**, 84, 482–492. [Google Scholar] [CrossRef][Green Version] - Sands, K.E.; Appel, M.L.; Lilly, L.S.; Schoen, F.J.; Mudge, G.H., Jr.; Cohen, R.J. Power spectrum analysis of heart rate variability in human cardiac transplant recipients. Circulation
**1989**, 79, 76–82. [Google Scholar] [CrossRef][Green Version] - Rechlin, T.; Weis, M.; Spitzer, A.; Kaschka, W.P. Are affective disorders associated with alterations of heart rate variability? J. Affect. Disord.
**1994**, 32, 271–275. [Google Scholar] [CrossRef] - Gil, E.; Orini, M.; Bailón, R.; Vergara, J.M.; Marozas, V.; Mainardi, L.; Laguna, P. Photoplethysmography pulse rate variability as a surrogate measurement of heart rate variability during non-stationary conditions. Physiol. Meas.
**2010**, 31, 1271–1290. [Google Scholar] [CrossRef] - Schäfer, A.; Vagedes, J. How accurate is pulse rate variability as an estimate of heart rate variability?: A review on studies comparing photoplethysmographic technology with an electrocardiogram. Int. J. Cardiol.
**2013**, 166, 15–29. [Google Scholar] [CrossRef] - Sacha, J.; Pluta, W. Alterations of an average heart rate change heart rate variability due to mathematical reasons. Int. J. Cardiol.
**2008**, 128, 444–447. [Google Scholar] [CrossRef] - Sacha, J. Interplay between heart rate and its variability: A prognostic game. Front. Psychol.
**2014**, 5, 347. [Google Scholar] [CrossRef][Green Version] - Billman, G.E.; Huikuri, H.V.; Sacha, J.; Trimmel, K. An introduction to heart rate variability: Methodological considerations and clinical applications. Front. Psychol.
**2015**, 6, 55. [Google Scholar] [CrossRef][Green Version] - Buchheit, M. Monitoring training status with HR measures: Do all roads lead to Rome? Front. Psychol.
**2014**, 5, 73. [Google Scholar] [CrossRef][Green Version] - Valenza, G.; Garcia, R.G.; Citi, L.; Scilingo, E.P.; Tomaz, C.A.; Barbieri, R. Nonlinear digital signal processing in mental health: Characterization of major depression using instantaneous entropy measures of heartbeat dynamics. Front. Psychol.
**2015**, 6, 74. [Google Scholar] [CrossRef][Green Version] - Weippert, M.; Behrens, K.; Rieger, A.; Stoll, R.; Kreuzfeld, S. Heart rate variability and blood pressure during dynamic and static exercise at similar heart rate levels. PLoS ONE
**2013**, 8, e83690. [Google Scholar] [CrossRef] - Kemper, K.J.; Hamilton, C.; Atkinson, M. Heart rate variability: Impact of differences in outlier identification and management strategies on common measures in three clinical populations. Pediatr. Res.
**2007**, 62, 337–342. [Google Scholar] [CrossRef][Green Version] - Mateo, J.; Laguna, P. Analysis of heart rate variability in the presence of ectopic beats using the heart timing signal. IEEE Trans. Biomed. Eng.
**2003**, 50, 334–343. [Google Scholar] [CrossRef][Green Version] - McNames, J.; Thong, T.; Aboy, M. Impulse rejection filter for artifact removal in spectral analysis of biomedical signals. In Proceedings of the 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, San Francisco, CA, USA, 1–5 September 2004; Volume 1, pp. 145–148. [Google Scholar]
- Lee, M.Y.; Yu, S.N. Improving discriminality in heart rate variability analysis using simple artifact and trend removal preprocessors. In Proceedings of the 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, Buenos Aires, Argentina, 31 August–4 September 2010; pp. 4574–4577. [Google Scholar]
- Begum, S.; Islam, M.S.; Ahmed, M.U.; Funk, P. K-NN based interpolation to handle artifacts for heart rate variability analysis. In Proceedings of the 2011 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), Bilbao, Spain, 14–17 December 2011; pp. 387–392. [Google Scholar]
- Al Osman, H.; Eid, M.; El Saddik, A. A pattern-based windowed impulse rejection filter for nonpathological HRV artifacts correction. IEEE Trans. Instrum. Meas.
**2014**, 64, 1944–1957. [Google Scholar] [CrossRef] - Giles, D.A.; Draper, N. Heart rate variability during exercise: A comparison of artefact correction methods. J. Strength Cond. Res.
**2018**, 32, 726–735. [Google Scholar] [CrossRef] - Baek, H.J.; Shin, J. Effect of missing inter-beat interval data on heart rate variability analysis using wrist-worn wearables. J. Med. Syst.
**2017**, 41, 147. [Google Scholar] [CrossRef] - Morelli, D.; Rossi, A.; Cairo, M.; Clifton, D.A. Analysis of the impact of interpolation methods of missing RR-intervals caused by motion artifacts on HRV features estimations. Sensors
**2019**, 19, 3163. [Google Scholar] [CrossRef] [PubMed][Green Version] - Benchekroun, M.; Chevallier, B.; Istrate, D.; Zalc, V.; Lenne, D. Preprocessing Methods for Ambulatory HRV Analysis Based on HRV Distribution, Variability and Characteristics (DVC). Sensors
**2022**, 22, 1984. [Google Scholar] [CrossRef] [PubMed] - Królak, A.; Wiktorski, T.; Bjørkavoll-Bergseth, M.F.; Ørn, S. Artifact correction in short-term hrv during strenuous physical exercise. Sensors
**2020**, 20, 6372. [Google Scholar] [CrossRef] [PubMed] - Berwal, D.; Vandana, C.R.; Dewan, S.; Jiji, C.V.; Baghini, M.S. Motion artifact removal in ambulatory ECG signal for heart rate variability analysis. IEEE Sens. J.
**2019**, 19, 12432–12442. [Google Scholar] [CrossRef] - Aygun, A.; Ghasemzadeh, H.; Jafari, R. Robust interbeat interval and heart rate variability estimation method from various morphological features using wearable sensors. IEEE J. Biomed. Health Inform.
**2019**, 24, 2238–2250. [Google Scholar] [CrossRef] [PubMed] - Martínez, J.P.; Almeida, R.; Olmos, S.; Rocha, A.P.; Laguna, P. A wavelet-based ECG delineator: Evaluation on standard databases. IEEE. Trans. Biomed. Eng.
**2004**, 51, 570–581. [Google Scholar] [CrossRef] [PubMed] - Hernando, D.; Roca, S.; Sancho, J.; Alesanco, Á.; Bailón, R. Validation of the apple watch for heart rate variability measurements during relax and mental stress in healthy subjects. Sensors
**2018**, 18, 2619. [Google Scholar] [CrossRef][Green Version] - Luczak, H.; Laurig, W. An analysis of heart rate variability. Ergonomics
**1973**, 16, 85–97. [Google Scholar] [CrossRef] - Rompelman, O.; Coenen, A.J.R.M.; Kitney, R.I. Measurement of heart-rate variability: Part 1—Comparative study of heart-rate variability analysis methods. Med. Biol. Eng. Comput.
**1977**, 15, 233–239. [Google Scholar] [CrossRef] - Kim, K.K.; Kim, J.S.; Lim, Y.G.; Park, K.S. The effect of missing RR-interval data on heart rate variability analysis in the frequency domain. Physiol. Meas.
**2009**, 30, 1039. [Google Scholar] [CrossRef] - Hernando, A.; Lazaro, J.; Gil, E.; Arza, A.; Garzon, J.M.; Lopez-Anton, R.; de la Camara, C.; Laguna, P.; Aguilo, J.; Bailon, R. Inclusion of Respiratory Frequency Information in Heart Rate Variability Analysis for Stress Assessment. IEEE J. Biomed. Health Inform.
**2016**, 20, 1016–1025. [Google Scholar] [CrossRef] - Bayly, E.J. Spectral analysis of pulse frequency modulation in the nervous systems. IEEE. Trans. Biomed. Eng.
**1968**, BME-15, 257–265. [Google Scholar] [CrossRef] - Mateo, J.; Laguna, P. Improved heart rate variability signal analysis from the beat occurrence times according to the IPFM model. IEEE. Trans. Biomed. Eng.
**2000**, 47, 985–996. [Google Scholar] [CrossRef] - Koichubekov, B.; Riklefs, V.; Sorokina, M.; Korshukov, I.; Turgunova, L.; Laryushina, Y.; Kultenova, M. Informative nature and nonlinearity of lagged poincaré plots indices in analysis of heart rate variability. Entropy
**2017**, 19, 523. [Google Scholar] [CrossRef][Green Version] - Nardelli, M.; Greco, A.; Bolea, J.; Valenza, G.; Scilingo, E.P.; Bailón, R. Reliability of lagged poincaré plot parameters in ultrashort heart rate variability series: Application on affective sounds. IEEE. J. Biomed. Health Inform.
**2017**, 22, 741–749. [Google Scholar] [CrossRef] - Laguna, P.; Moody, G.B.; Mark, R.G. Power spectral density of unevenly sampled data by least-square analysis: Performance and application to heart rate signals. IEEE. Trans. Biomed. Eng.
**1998**, 45, 698–715. [Google Scholar] [CrossRef]

**Figure 1.**Example of simulation with a segment of 40 beats. Deleted beats are displayed in red. (

**a**) Random distributed missing beats, $p=0.25$. (

**b**) Bursts of missing beats. The elements at the ends (green) cannot be deleted.

**Figure 2.**Process flow. $\mathcal{OD}$ = Outlier Detection; $\mathcal{OR}$ = Outlier Rejection; $\mathcal{L}$ = Linear; $\mathcal{NL}$ = Non-Linear; $\mathcal{M}$ = Model-based.

**Figure 5.**

**Relax (green)/stress (blue) discrimination of time-domain metrics from Apple Watch dataset.**(

**a**) MHR. (

**b**) SDNN. (

**c**) RMSSD. *: Significant differences ($p<0.05$) between relax and stress groups. **: Significant differences ($p<0.001$) between relax and stress groups.

**Figure 6.**

**Coverage of frequency-domain metrics computed via FFT from Apple Watch dataset.**(

**a**) ${P}_{LF}$. (

**b**) ${P}_{HF}$.

**Figure 7.**

**Relax (green)/Stress (blue) discrimination of frequency-domain metrics computed via FFT from Apple Watch dataset.**(

**a**) ${P}_{LF}$. (

**b**) ${P}_{HF}$. *: Significant differences ($p<0.05$) between relax and stress groups. **: Significant differences ($p<0.001$) between relax and stress groups.

**Figure 8.**

**Coverage of frequency-domain metrics computed via Lomb’s method from Apple Watch dataset.**(

**a**) ${P}_{LF}$. (

**b**) ${P}_{HF}$.

**Figure 9.**

**Relax (green)/Stress (blue) discrimination of frequency-domain metrics computed via Lomb’s method from Apple Watch dataset.**(

**a**) ${P}_{LF}$. (

**b**) ${P}_{HF}$. **: Significant differences ($p<0.001$) between relax and stress groups.

**Figure 11.**

**Relax (green)/stress (blue) discrimination of Poincaré metrics from Apple Watch dataset.**(

**a**) SD1. (

**b**) SD2. (

**c**) S. (

**d**) Md. (

**e**) Sd. *: Significant differences ($p<0.05$) between relax and stress groups. **: Significant differences ($p<0.001$) between relax and stress groups.

**Table 1.**

**Relative error (%) of time-domain metrics. (a)**Scattered missing beats.

**(b)**Bursts. †: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

(a) | |||||
---|---|---|---|---|---|

Method | Metric | Deletion Probability (%) | |||

5 | 15 | 25 | 35 | ||

$\mathcal{OR}$ | MHR | 0.13 (0.05–0.24) | 0.25 (0.12–0.48) ${}^{\u2020}$ | 0.39 (0.20–0.74) ${}^{\u2020}$ | 0.54 (0.28–1.02) ${}^{\u2020}$ |

SDNN | 1.80 (0.85–3.07) ${}^{\u2020}$ | 3.61 (1.76–6.03) ${}^{\u2020}$ | 5.10 (2.34–9.18) ${}^{\u2020}$ | 7.32 (3.11–14.93) ${}^{\u2020}$ | |

RMSSD | 2.09 (0.95–4.03) ${}^{\u2020}$ | 5.40 (2.12–9.23) ${}^{\u2020}$ | 8.90 (3.96–14.55) | 10.84 (5.21–23.97) | |

$\mathcal{L}$ | MHR | 0.00 (0.00–0.01) ${}^{\u25b5}$ | 0.01 (0.01–0.03) ${}^{\u25b5}$ | 0.03 (0.01–0.49) ${}^{\u25b5}$ | 0.08 (0.03–0.75) ${}^{\u25b5}$ |

SDNN | 0.43 (0.19–0.81) ${}^{\u25b5}$ | 1.85 (0.81–3.37) ${}^{\u25b5}$ | 4.71 (2.59–8.25) ${}^{\u25b5}$ | 8.09 (4.18–14.48) ${}^{\u25b5}$ | |

RMSSD | 1.07 (0.41–2.05) ${}^{\u25b5}$ | 2.68 (1.14–6.09) ${}^{\u25b5}$ | 7.90 (2.72–19.56) ${}^{\u25b5}$ | 13.98 (5.21–37.84) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | MHR | 0.00 (0.00–0.00) | 0.00 (0.00–0.02) | 0.02 (0.01–0.09) | 0.05 (0.01–0.70) ${}^{\mathrm{\S}}$ |

SDNN | 0.16 (0.05–0.44) ${}^{\mathrm{\S}}$ | 0.77 (0.24–2.34) ${}^{\mathrm{\S}}$ | 2.63 (0.85–6.00) ${}^{\mathrm{\S}}$ | 5.42 (2.10–10.18) ${}^{\mathrm{\S}}$ | |

RMSSD | 1.13 (0.44–2.28) ${}^{\mathrm{\S}}$ | 4.14 (2.13–7.43) ${}^{\mathrm{\S}}$ | 9.42 (4.95–16.70) ${}^{\mathrm{\S}}$ | 14.50 (7.54–29.35) ${}^{\mathrm{\S}}$ | |

(b) | |||||

Method | Metric | Burst duration (s) | |||

5 | 10 | 15 | 20 | ||

$\mathcal{OR}$ | MHR | 0.16 (0.07–0.28) | 0.22 (0.11–0.44) ${}^{\u2020}$ | 0.31 (0.14–0.56) ${}^{\u2020}$ | 0.40 (0.17–0.71) ${}^{\u2020}$ |

SDNN | 1.69 (0.84–2.43) ${}^{\u2020}$ | 2.12 (1.10–3.65) ${}^{\u2020}$ | 3.06 (1.40–4.87) ${}^{\u2020}$ | 3.55 (1.72–5.96) ${}^{\u2020}$ | |

RMSSD | 1.66 (0.83–2.44) ${}^{\u2020}$ | 2.43 (1.14–3.74) ${}^{\u2020}$ | 3.15 (1.58–5.38) ${}^{\u2020}$ | 4.08 (2.07–7.08) ${}^{\u2020}$ | |

$\mathcal{L}$ | MHR | 0.01 (0.00–0.03) ${}^{\u25b5}$ | 0.03 (0.01–0.53) ${}^{\u25b5}$ | 0.47 (0.02–0.76) ${}^{\u25b5}$ | 0.73 (0.09–0.98) ${}^{\u25b5}$ |

SDNN | 1.39 (0.57–2.66) ${}^{\u25b5}$ | 3.41 (1.51–5.32) ${}^{\u25b5}$ | 4.83 (2.63–7.82) ${}^{\u25b5}$ | 6.38 (3.14–9.87) ${}^{\u25b5}$ | |

RMSSD | 1.66 (0.75–3.38) ${}^{\u25b5}$ | 3.60 (1.83–6.23) ${}^{\u25b5}$ | 4.87 (2.84–8.41) ${}^{\u25b5}$ | 6.97 (3.95–10.86) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | MHR | 0.01 (0.00–0.09) ${}^{\mathrm{\S}}$ | 0.02 (0.01–0.57) ${}^{\mathrm{\S}}$ | 0.09 (0.01–0.72) | 0.55 (0.03–0.83) |

SDNN | 1.24 (0.44–3.02) ${}^{\mathrm{\S}}$ | 2.84 (1.04–4.82) ${}^{\mathrm{\S}}$ | 4.46 (2.15–7.02) ${}^{\mathrm{\S}}$ | 5.80 (2.86–8.85) ${}^{\mathrm{\S}}$ | |

RMSSD | 1.77 (0.90–3.87) ${}^{\mathrm{\S}}$ | 3.77 (2.13–6.51) ${}^{\mathrm{\S}}$ | 5.80 (3.40–9.01) ${}^{\mathrm{\S}}$ | 7.78 (4.36–12.17) ${}^{\mathrm{\S}}$ |

**Table 2.**

**p**

**-values of ranked signed test for supine/tilt discrimination of time-domain metrics.**N.S.: Not significant ($p>0.05$).

Method | Metric | Reference | Deletion Probability (%) | Burst Duration (s) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

5 | 15 | 25 | 35 | 5 | 10 | 15 | 20 | |||||

$\mathcal{OR}$ | MHR | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SDNN | 0.020 | 0.011 | 0.016 | N.S. | N.S. | 0.011 | 0.011 | 0.011 | 0.011 | |||

RMSSD | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{L}$ | MHR | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SDNN | 0.020 | 0.014 | 0.008 | N.S. | N.S. | 0.011 | 0.011 | 0.011 | 0.004 | |||

RMSSD | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{NL}$ | MHR | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SDNN | 0.020 | 0.012 | 0.011 | 0.021 | 0.014 | 0.014 | 0.007 | 0.012 | 0.026 | |||

RMSSD | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ |

**Table 3.**

**Relative error (%) of time-domain metrics from Apple Watch dataset.**†: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

Metric | Method | ||
---|---|---|---|

$\mathcal{OR}$ | $\mathcal{L}$ | $\mathcal{NL}$ | |

MHR | 0.12 (0.04–0.47) | 0.03 (0.01–0.52) ${}^{\u25b5}$ | 0.03 (0.01–0.66) ${}^{\mathrm{\S}}$ |

SDNN | 3.36 (1.97–7.47) ${}^{\u2020}$ | 2.92 (1.51–9.55) ${}^{\u25b5}$ | 2.96 (1.31–8.62) |

RMSSD | 7.84 (4.29–15.90) ${}^{\u2020}$ | 8.56 (3.99–20.22) ${}^{\u25b5}$ | 8.61 (3.74–17.69) ${}^{\mathrm{\S}}$ |

**Table 4.**

**Relative error (%) of frequency-domain metrics computed via FFT. (a)**Scattered missing beats.

**(b)**Bursts. †: Significant differences ($p<0.05$) between $\mathcal{M}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{M}$.

(a) | |||||
---|---|---|---|---|---|

Method | Metric | Deletion Probability (%) | |||

5 | 15 | 25 | 35 | ||

$\mathcal{M}$ | ${P}_{LF}$ | 8.08 (3.46–19.10) ${}^{\u2020}$ | 20.29 (11.37–32.72) ${}^{\u2020}$ | 36.36 (22.36–53.64) ${}^{\u2020}$ | 55.16 (29.10–161.86) ${}^{\u2020}$ |

${P}_{HF}$ | 15.37 (7.00–30.10) ${}^{\u2020}$ | 32.89 (20.17–45.24) ${}^{\u2020}$ | 50.12 (36.46–63.16) ${}^{\u2020}$ | 59.41 (42.65–73.21) ${}^{\u2020}$ | |

$\mathcal{L}$ | ${P}_{LF}$ | 0.99 (0.39–2.34) ${}^{\u25b5}$ | 4.28 (2.04–9.24) ${}^{\u25b5}$ | 10.94 (5.66–18.71) ${}^{\u25b5}$ | 15.98 (8.81–28.74) ${}^{\u25b5}$ |

${P}_{HF}$ | 2.81 (1.19–5.47) ${}^{\u25b5}$ | 10.83 (5.54–17.64) ${}^{\u25b5}$ | 22.69 (11.98–41.40) ${}^{\u25b5}$ | 34.04 (19.54–61.20) | |

$\mathcal{NL}$ | ${P}_{LF}$ | 0.41 (0.15–1.11) ${}^{\mathrm{\S}}$ | 1.44 (0.48–4.71) ${}^{\mathrm{\S}}$ | 4.24 (1.41–12.57) ${}^{\mathrm{\S}}$ | 8.96 (2.20–21.22) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 1.63 (0.71–4.16) ${}^{\mathrm{\S}}$ | 6.88 (2.45–14.99) ${}^{\mathrm{\S}}$ | 18.97 (9.80–37.55) ${}^{\mathrm{\S}}$ | 29.20 (17.06–54.77) ${}^{\mathrm{\S}}$ | |

(b) | |||||

Method | Metric | Burst duration (s) | |||

5 | 10 | 15 | 20 | ||

$\mathcal{M}$ | ${P}_{LF}$ | 10.20 (3.53–20.75) ${}^{\u2020}$ | 14.62 (6.52–26.29) ${}^{\u2020}$ | 21.90 (9.95–32.57) ${}^{\u2020}$ | 26.50 (14.26–39.04) ${}^{\u2020}$ |

${P}_{HF}$ | 12.62 (6.38–28.40) ${}^{\u2020}$ | 17.99 (9.32–32.90) ${}^{\u2020}$ | 22.82 (14.13–36.28) ${}^{\u2020}$ | 28.65 (18.53–43.37) ${}^{\u2020}$ | |

$\mathcal{L}$ | ${P}_{LF}$ | 4.94 (1.70–12.34) ${}^{\u25b5}$ | 10.89 (4.67–19.12) ${}^{\u25b5}$ | 15.45 (7.13–26.16) ${}^{\u25b5}$ | 19.25 (8.88–31.24) |

${P}_{HF}$ | 6.81 (2.92–11.42) ${}^{\u25b5}$ | 9.95 (4.98–17.20) ${}^{\u25b5}$ | 14.02 (7.06–22.67) ${}^{\u25b5}$ | 18.26 (9.41–28.11) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | ${P}_{LF}$ | 4.72 (1.56–12.18) ${}^{\mathrm{\S}}$ | 10.01 (4.34–17.88) ${}^{\mathrm{\S}}$ | 13.31 (6.35–25.36) ${}^{\mathrm{\S}}$ | 19.34 (9.28–30.70) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 6.82 (3.36–11.76) ${}^{\mathrm{\S}}$ | 11.02 (5.73–17.59) ${}^{\mathrm{\S}}$ | 15.19 (7.85–23.70) ${}^{\mathrm{\S}}$ | 19.10 (10.94–29.80) ${}^{\mathrm{\S}}$ |

**Table 5.**

**p**

**-values of ranked signed test for supine/tilt discrimination of frequency-domain metrics computed via FFT.**N.S.: Not significant ($p>0.05$).

Method | Metric | Reference | Deletion Probability (%) | Burst Duration (s) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

5 | 15 | 25 | 35 | 5 | 10 | 15 | 20 | |||||

$\mathcal{M}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | N.S. | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | 0.005 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{L}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{NL}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ |

**Table 6.**

**Relative error (%) of frequency-domain metrics computed via FFT from Apple Watch dataset.**†: Significant differences ($p<0.05$) between $\mathcal{M}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{M}$.

Metric | Method | ||
---|---|---|---|

$\mathcal{M}$ | $\mathcal{L}$ | $\mathcal{NL}$ | |

${P}_{LF}$ | 0.09 (0.04–0.30) ${}^{\u2020}$ | 0.08 (0.03–0.22) ${}^{\u25b5}$ | 0.08 (0.03–0.17) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 0.14 (0.07–0.31) ${}^{\u2020}$ | 0.16 (0.07–0.30) ${}^{\u25b5}$ | 0.17 (0.07–0.31) ${}^{\mathrm{\S}}$ |

**Table 7.**

**Relative error (%) of frequency-domain metrics computed via Lomb’s method. (a)**Scattered missing beats.

**(b)**Bursts. †: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

(a) | |||||
---|---|---|---|---|---|

Method | Metric | Deletion Probability (%) | |||

5 | 15 | 25 | 35 | ||

$\mathcal{OR}$ | ${P}_{LF}$ | 10.75 (4.77–18.84) ${}^{\u2020}$ | 23.45 (9.90–40.99) ${}^{\u2020}$ | 34.71 (17.49–66.27) ${}^{\u2020}$ | 58.11 (24.93–123.23) ${}^{\u2020}$ |

${P}_{HF}$ | 23.01 (11.38–45.74) ${}^{\u2020}$ | 79.42 (46.93–155.29) ${}^{\u2020}$ | 160.28 (87.39–296.90) ${}^{\u2020}$ | 304.57 (142.89–665.16) ${}^{\u2020}$ | |

$\mathcal{L}$ | ${P}_{LF}$ | 0.89 (0.35–1.90) ${}^{\u25b5}$ | 3.75 (1.60–7.49) ${}^{\u25b5}$ | 9.90 (4.56–18.19) ${}^{\u25b5}$ | 15.77 (7.56–28.59) ${}^{\u25b5}$ |

${P}_{HF}$ | 2.62 (1.10–4.81) ${}^{\u25b5}$ | 8.94 (4.93–16.22) ${}^{\u25b5}$ | 21.27 (11.03–37.15) ${}^{\u25b5}$ | 30.78 (17.23–61.62) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | ${P}_{LF}$ | 0.37 (0.13–1.06) ${}^{\mathrm{\S}}$ | 1.36 (0.44–3.95) ${}^{\mathrm{\S}}$ | 3.43 (1.17–11.70) ${}^{\mathrm{\S}}$ | 7.58 (2.28–22.67) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 1.45 (0.51–3.31) ${}^{\mathrm{\S}}$ | 5.46 (1.92–12.01) ${}^{\mathrm{\S}}$ | 16.22 (8.12–31.57) ${}^{\mathrm{\S}}$ | 28.33 (14.72–52.65) ${}^{\mathrm{\S}}$ | |

(b) | |||||

Method | Metric | Burst Duration (s) | |||

5 | 10 | 15 | 20 | ||

$\mathcal{OR}$ | ${P}_{LF}$ | 11.19 (6.69–17.18) ${}^{\u2020}$ | 18.66 (10.87–28.82) ${}^{\u2020}$ | 25.33 (12.89–38.36) ${}^{\u2020}$ | 29.23 (14.57–48.91) ${}^{\u2020}$ |

${P}_{HF}$ | 14.06 (7.55–19.79) ${}^{\u2020}$ | 22.86 (12.70–34.06) ${}^{\u2020}$ | 30.88 (18.39–45.27) ${}^{\u2020}$ | 39.17 (24.01–60.79) ${}^{\u2020}$ | |

$\mathcal{L}$ | ${P}_{LF}$ | 4.48 (1.65–11.24) ${}^{\u25b5}$ | 9.99 (3.64–19.00) ${}^{\u25b5}$ | 13.85 (6.53–23.87) ${}^{\u25b5}$ | 17.38 (7.14–28.54) ${}^{\u25b5}$ |

${P}_{HF}$ | 5.51 (2.26–11.05) ${}^{\u25b5}$ | 8.74 (3.94–18.11) ${}^{\u25b5}$ | 13.18 (5.10–21.58) ${}^{\u25b5}$ | 16.22 (7.42–26.31) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | ${P}_{LF}$ | 4.58 (1.68–11.53) ${}^{\mathrm{\S}}$ | 8.43 (3.55–17.21) ${}^{\mathrm{\S}}$ | 13.49 (5.93–23.56) ${}^{\mathrm{\S}}$ | 18.41 (9.23–28.86) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 6.00 (2.79–11.69) ${}^{\mathrm{\S}}$ | 10.92 (5.24–19.42) ${}^{\mathrm{\S}}$ | 14.60 (7.65–23.30) ${}^{\mathrm{\S}}$ | 18.46 (9.45–29.31) ${}^{\mathrm{\S}}$ |

**Table 8.**

**p**

**-values of ranked signed test for supine/tilt discrimination of frequency-domain metrics computed via Lomb’s method.**N.S.: Not significant ($p>0.05$).

Method | Metric | Reference | Deletion Probability (%) | Burst Duration (s) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

5 | 15 | 25 | 35 | 5 | 10 | 15 | 20 | |||||

$\mathcal{OR}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{L}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

$\mathcal{NL}$ | ${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

${P}_{LFn}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

${P}_{LF}$/${P}_{HF}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ |

**Table 9.**

**Relative error (%) of frequency-domain computed via Lomb’s method metrics from Apple Watch dataset.**†: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

Metric | Method | ||
---|---|---|---|

$\mathcal{OR}$ | $\mathcal{L}$ | $\mathcal{NL}$ | |

${P}_{LF}$ | 0.10 (0.05–0.24) ${}^{\u2020}$ | 0.08 (0.03–0.23) ${}^{\u25b5}$ | 0.08 (0.03–0.18) ${}^{\mathrm{\S}}$ |

${P}_{HF}$ | 0.20 (0.08–0.56) ${}^{\u2020}$ | 0.15 (0.06–0.32) ${}^{\u25b5}$ | 0.13 (0.06–0.25) ${}^{\mathrm{\S}}$ |

**Table 10.**

**Relative error (%) of Poincaré metrics. (a)**Scattered missing beats.

**(b)**Bursts. †: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

(a) | |||||
---|---|---|---|---|---|

Method | Metric | Deletion Probability (%) | |||

5 | 15 | 25 | 35 | ||

$\mathcal{OR}$ | SD1 | 2.13 (0.94–3.96) ${}^{\u2020}$ | 5.33 (2.24–9.25) ${}^{\u2020}$ | 9.06 (4.10–14.36) | 10.66 (5.29–23.67) |

SD2 | 2.58 (1.31–4.28) ${}^{\u2020}$ | 4.93 (2.04–8.53) ${}^{\u2020}$ | 7.20 (3.22–13.05) ${}^{\u2020}$ | 10.75 (4.93–20.32) ${}^{\u2020}$ | |

Md | 2.40 (1.16–4.15) ${}^{\u2020}$ | 4.49 (2.18–7.44) ${}^{\u2020}$ | 6.35 (2.87–10.78) ${}^{\u2020}$ | 8.70 (4.10–17.82) ${}^{\u2020}$ | |

Sd | 2.80 (1.31–4.95) ${}^{\u2020}$ | 6.19 (2.87–11.14) | 10.05 (4.47–18.67) ${}^{\u2020}$ | 14.46 (7.16–31.01) ${}^{\u2020}$ | |

$\mathcal{L}$ | SD1 | 1.07 (0.41–2.05) ${}^{\u25b5}$ | 2.68 (1.14–6.09) ${}^{\u25b5}$ | 7.90 (2.72–19.56) ${}^{\u25b5}$ | 14.00 (5.21–37.87) ${}^{\u25b5}$ |

SD2 | 0.40 (0.19–0.93) ${}^{\u25b5}$ | 1.85 (0.97–3.30) ${}^{\u25b5}$ | 4.28 (2.24–7.08) ${}^{\u25b5}$ | 6.99 (3.95–13.22) ${}^{\u25b5}$ | |

Md | 0.52 (0.22–1.08) ${}^{\u25b5}$ | 2.05 (0.85–4.18) ${}^{\u25b5}$ | 4.46 (2.07–7.23) ${}^{\u25b5}$ | 7.11 (3.54–11.93) ${}^{\u25b5}$ | |

Sd | 0.47 (0.17–0.98) ${}^{\u25b5}$ | 1.34 (0.52–3.23) ${}^{\u25b5}$ | 3.59 (1.27–12.57) ${}^{\u25b5}$ | 7.09 (1.78–33.63) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | SD1 | 1.13 (0.44–2.28) ${}^{\mathrm{\S}}$ | 4.14 (2.13–7.43) ${}^{\mathrm{\S}}$ | 9.42 (4.95–16.70) ${}^{\mathrm{\S}}$ | 14.51 (7.54–29.39) ${}^{\mathrm{\S}}$ |

SD2 | 0.12 (0.04–0.30) ${}^{\mathrm{\S}}$ | 0.56 (0.15–1.50) ${}^{\mathrm{\S}}$ | 1.67 (0.54–3.98) ${}^{\mathrm{\S}}$ | 3.39 (1.19–8.51) ${}^{\mathrm{\S}}$ | |

Md | 0.23 (0.09–0.56) ${}^{\mathrm{\S}}$ | 1.06 (0.27–3.00) ${}^{\mathrm{\S}}$ | 2.43 (1.04–5.73) ${}^{\mathrm{\S}}$ | 4.83 (1.88–9.33) ${}^{\mathrm{\S}}$ | |

Sd | 0.30 (0.10–0.76) ${}^{\mathrm{\S}}$ | 0.90 (0.26–2.46) ${}^{\mathrm{\S}}$ | 2.39 (0.77–7.89) ${}^{\mathrm{\S}}$ | 4.92 (1.60–17.24) ${}^{\mathrm{\S}}$ | |

Method | Metric | Burst Duration (s) | |||

5 | 10 | 15 | 20 | ||

$\mathcal{OR}$ | SD1 | 1.69 (0.84–2.50) ${}^{\u2020}$ | 2.45 (1.16–3.76) ${}^{\u2020}$ | 3.19 (1.63–5.39) ${}^{\u2020}$ | 4.03 (2.00–7.04) ${}^{\u2020}$ |

SD2 | 1.85 (0.89–2.74) ${}^{\u2020}$ | 2.44 (1.29–3.92) ${}^{\u2020}$ | 3.24 (1.57–5.11) ${}^{\u2020}$ | 3.94 (1.89–6.39) ${}^{\u2020}$ | |

Md | 1.91 (0.94–3.05) ${}^{\u2020}$ | 2.54 (1.04–4.59) ${}^{\u2020}$ | 3.49 (1.26–5.86) ${}^{\u2020}$ | 4.14 (2.01–7.40) ${}^{\u2020}$ | |

Sd | 1.50 (0.90–2.49) | 2.32 (1.21–3.65) | 2.98 (1.65–4.54) | 3.57 (2.02–5.67)${}^{\u2020}$ | |

$\mathcal{L}$ | SD1 | 1.67 (0.75–3.38) ${}^{\u25b5}$ | 3.60 (1.83–6.23) ${}^{\u25b5}$ | 4.88 (2.84–8.41) ${}^{\u25b5}$ | 6.97 (3.95–10.86) ${}^{\u25b5}$ |

SD2 | 1.31 (0.56–2.70) ${}^{\u25b5}$ | 3.40 (1.36–5.37) ${}^{\u25b5}$ | 4.90 (2.49–7.88) ${}^{\u25b5}$ | 6.26 (3.10–10.14) ${}^{\u25b5}$ | |

Md | 2.33 (0.97–4.11) ${}^{\u25b5}$ | 5.65 (2.90–8.72) ${}^{\u25b5}$ | 8.44 (4.31–11.97) ${}^{\u25b5}$ | 10.53 (4.53–14.29) ${}^{\u25b5}$ | |

Sd | 1.97 (0.75–4.06) ${}^{\u25b5}$ | 3.24 (1.57–6.84) | 4.08 (1.92–7.50) | 4.68 (2.37–8.05) ${}^{\u25b5}$ | |

$\mathcal{NL}$ | SD1 | 1.77 (0.90–3.87) ${}^{\mathrm{\S}}$ | 3.78 (2.13–6.52) ${}^{\mathrm{\S}}$ | 5.80 (3.41–9.02) ${}^{\mathrm{\S}}$ | 7.78 (4.36–12.17) ${}^{\mathrm{\S}}$ |

SD2 | 1.16 (0.32–2.87) ${}^{\mathrm{\S}}$ | 2.53 (0.89–4.96) ${}^{\mathrm{\S}}$ | 4.42 (2.18–6.88) ${}^{\mathrm{\S}}$ | 5.60 (2.61–8.70) ${}^{\mathrm{\S}}$ | |

Md | 1.16 (0.32–2.87) ${}^{\mathrm{\S}}$ | 2.53 (0.89–4.96) ${}^{\mathrm{\S}}$ | 4.42 (2.18–6.88) ${}^{\mathrm{\S}}$ | 5.60 (2.61–8.70) ${}^{\mathrm{\S}}$ | |

Sd | 1.81 (0.61–4.33) | 2.97 (1.21–5.62) | 3.53 (1.67–7.63) | 4.17 (1.99–7.03) |

**Table 11.**

**p**

**-values of ranked signed test for supine/tilt discrimination of Poincaré metrics.**N.S.: Not Significative ($p>0.05$).

Method | Metric | Reference | Deletion Probability (%) | Burst Duration (s) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

5 | 15 | 25 | 35 | 5 | 10 | 15 | 20 | |||||

$\mathcal{OR}$ | SD1 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SD2 | 0.031 | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | |||

SD12 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

S | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | 0.012 | 0.002 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

Md | <${10}^{-3}$ | 0.001 | 0.002 | 0.155 | 0.016 | 0.001 | 0.002 | 0.002 | 0.002 | |||

Sd | 0.039 | 0.009 | N.S. | N.S. | N.S. | 0.024 | 0.026 | 0.022 | 0.015 | |||

$\mathcal{L}$ | SD1 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SD2 | 0.031 | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | |||

SD12 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

S | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | 0.007 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

Md | <${10}^{-3}$ | <${10}^{-3}$ | 0.001 | 0.010 | 0.012 | <${10}^{-3}$ | <${10}^{-3}$ | 0.002 | 0.002 | |||

Sd | 0.039 | 0.034 | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | |||

$\mathcal{NL}$ | SD1 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | ||

SD2 | 0.031 | 0.043 | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | N.S. | |||

SD12 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

S | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | 0.004 | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | |||

Md | <${10}^{-3}$ | <${10}^{-3}$ | <${10}^{-3}$ | 0.002 | 0.002 | 0.002 | 0.001 | 0.002 | 0.013 | |||

Sd | 0.039 | 0.033 | N.S. | N.S. | N.S. | 0.041 | 0.041 | 0.027 | 0.049 |

**Table 12.**

**Relative error (%) of Poincaré metrics from Apple Watch dataset.**†: Significant differences ($p<0.05$) between $\mathcal{OR}$ and $\mathcal{L}$. △: Significant differences ($p<0.05$) between $\mathcal{L}$ and $\mathcal{NL}$. §: Significant differences ($p<0.05$) between $\mathcal{NL}$ and $\mathcal{OR}$.

Metric | Method | ||
---|---|---|---|

$\mathcal{OR}$ | $\mathcal{L}$ | $\mathcal{NL}$ | |

SD1 | 7.83 (4.15–15.91) ${}^{\u2020}$ | 8.56 (3.99–20.22) ${}^{\u25b5}$ | 8.61 (3.73–17.70) ${}^{\mathrm{\S}}$ |

SD2 | 3.22 (1.59–6.38) ${}^{\u2020}$ | 2.61 (1.06–7.98) ${}^{\u25b5}$ | 2.35 (1.06–6.08) |

Md | 3.97 (2.21–8.28) ${}^{\u2020}$ | 3.55 (2.03–9.24) ${}^{\u25b5}$ | 3.40 (1.67–8.42) |

Sd | 4.20 (1.66–9.42) | 3.96 (1.87–10.00) ${}^{\u25b5}$ | 3.44 (1.49–10.06) ${}^{\mathrm{\S}}$ |

**Table 13.**Summary of findings.

**(a)**Best correction method.

**(b)**Maximum acceptable missing beats for a relative error less than 20% in the third quartile.

(a) | ||
---|---|---|

Metric | Scattered Missing Beats | Bursts |

MHR | $\mathcal{NL}$ | $\mathcal{NL}$ |

SDNN | $\mathcal{NL}$ | $\mathcal{OR}$ |

RMSSD | $\mathcal{L}$ | $\mathcal{OR}$ |

${P}_{LF}$ (FFT) | $\mathcal{NL}$ | $\mathcal{NL}$ |

${P}_{HF}$ (FFT) | $\mathcal{NL}$ | $\mathcal{L}$ |

${P}_{LF}$ (Lomb) | $\mathcal{NL}$ | $\mathcal{NL}$ |

${P}_{HF}$ (Lomb) | $\mathcal{NL}$ | $\mathcal{L}$ |

SD1 | $\mathcal{L}$ | $\mathcal{OR}$ |

SD2 | $\mathcal{NL}$ | $\mathcal{OR}$ |

Md | $\mathcal{NL}$ | $\mathcal{OR}$ |

Sd | $\mathcal{NL}$ | $\mathcal{OR}$/$\mathcal{NL}$ |

(b) | ||

Metric | Scattered Missing Beats | Bursts |

MHR | 35% | 20 s |

SDNN | 35% | 20 s |

RMSSD | 25% | 20 s |

${P}_{LF}$ (FFT) | 25% | 10 s |

${P}_{HF}$ (FFT) | 15% | 10 s |

${P}_{LF}$ (Lomb) | 25% | 10 s |

${P}_{HF}$ (Lomb) | 15% | 10 s |

SD1 | 25% | 20 s |

SD2 | 35% | 20 s |

Md | 35% | 20 s |

Sd | 35% | 20 s |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Cajal, D.; Hernando, D.; Lázaro, J.; Laguna, P.; Gil, E.; Bailón, R.
Effects of Missing Data on Heart Rate Variability Metrics. *Sensors* **2022**, *22*, 5774.
https://doi.org/10.3390/s22155774

**AMA Style**

Cajal D, Hernando D, Lázaro J, Laguna P, Gil E, Bailón R.
Effects of Missing Data on Heart Rate Variability Metrics. *Sensors*. 2022; 22(15):5774.
https://doi.org/10.3390/s22155774

**Chicago/Turabian Style**

Cajal, Diego, David Hernando, Jesús Lázaro, Pablo Laguna, Eduardo Gil, and Raquel Bailón.
2022. "Effects of Missing Data on Heart Rate Variability Metrics" *Sensors* 22, no. 15: 5774.
https://doi.org/10.3390/s22155774