Secondary Logo

Journal Logo

Validity of the Polar S810 Heart Rate Monitor to Measure R-R Intervals at Rest


Medicine & Science in Sports & Exercise: May 2006 - Volume 38 - Issue 5 - p 887-893
doi: 10.1249/01.mss.0000218135.79476.9c
BASIC SCIENCES: Original Investigations

Purpose: This study was conducted to compare R-R intervals and the subsequent analysis of heart rate variability (HRV) obtained from the Polar S810 heart rate monitor (HRM) (Polar Electro Oy) with an electrocardiogram (ECG) (Physiotrace, Estaris, Lille, France) during an orthostatic test.

Methods: A total of 18 healthy men (age: 27.1 ± 1.9 yr; height: 1.82 ± 0.06 m; mass 77.1 ± 7.7 kg) performed an active orthostatic test during which R-R intervals were simultaneously recorded with the HRM and the ECG recorder The two signals were synchronized and corrected before a time domain analysis, the fast Fourier transform (FFT) and a Poincaré plot analysis. Bias and limits of agreement (LoA), effect size (ES), and correlation coefficients were calculated.

Results: R-R intervals were significantly different in the supine and standing position between the ECG and the HRM uncorrected and corrected signal (P < 0.05, ES = 0.000 and 0.006, respectively). The bias ± LoA, however, were 0.9 ± 12 ms. HRV parameters derived from both signals in both positions were not different (P > 0.05) and well correlated (r > 0.97, P < 0.05), except root mean square of difference (RMSSD) and SD1 in standing position (P < 0.05, ES = 0.052 and 0.057; r = 0.99 and 0.98, respectively).

Conclusion: Narrow LoA, good correlations, and small effect sizes support the validity of the Polar S810 HRM to measure R-R intervals and make the subsequent HRV analysis in supine position. Caution must be taken in standing position for the parameters sensitive to the short-term variability (i.e., RMSSD and SD1).

1Faculty of Sport Sciences, University of Lille, Ronchin, FRANCE; and 2Department of Kinesiology, University of Montreal, Montreal, CANADA

Address for correspondence: Laurent Bosquet, Department of Kinesiology, University of Montreal, CP 6128, succ. centre ville, Montreal, QC, Canada H3C 3J7; E-mail:

Submitted for publication September 2005.

Accepted for publication December 2005.

Measurement of heart rate variability (HRV) has become a common tool in the clinical domain because it appears sensitive to both physiological (13,21) and psychological (5,6) disorders. In sports medicine, it is generally used to assess adaptation (10,12,22) or maladaptation to endurance training (7,14). As such, it is a promising tool that may expand in the follow-up of elite athletes.

Measurement of HRV usually requires a high-quality electrocardiogram (ECG) with a sampling rate above 250 Hz and an accurate algorithm to detect the QRS complex (19). Over recent years, a number of ambulatory ECG recorders or Holter monitors that satisfy these requirements have been developed, permitting the use out of laboratory. The cost and the complexity of this equipment, however, made the HRV analysis difficult outside the laboratory and particularly in the physical training field conditions.

The development of wireless heart rate monitoring (HRM) with elastic electrode belt allowing the detection of R-R intervals with a resolution of 1 ms (9) represents an interesting alternative to classic fix or ambulatory ECG for coaches and physicians. It remains to determine the accuracy of this device before using it in regular basis. Kingsley et al. (8) reported good accuracy of the Polar S810 HRM (Polar Electro Oy) when compared with an ambulatory ECG during exercise at low intensity. Supine is the recommended position to detect both overreaching and overtraining (7); however, the accuracy of the HRM in this position is lacking. Moreover, the impact of differences between the signals on HRV parameters is not known, either in the time or frequency domain.

Thus, this study was conducted to compare (a) raw data obtained in a supine and standing position from the Polar S810 HRM and an ECG recorder; and (b) the HRV parameters derived from both signals in the time and frequency domain.

Back to Top | Article Outline



A total of 18 active men (age: 27.1 ± 1.9 yr; height: 1.82 ± 0.06 m; mass 77.1 ± 7.7 kg) with no smoking history and no known cardiovascular disease gave their written, informed consent to participate in the study. All of the subjects submitted to an inclusion protocol before the start of the study. This consisted of an information session about the nature, the potential risks involved, and the benefits of the study, followed by a complete medical screening when the subjects were interested in participating to the study. The protocol has been reviewed and approved by the consultative committee for the protection of human subjects in biomedical research of the Nord-Pas de Calais (France) before the start of the study.

Back to Top | Article Outline

Experimental Design

Two weeks after the inclusion visit, the subjects reported to the laboratory within 2 h of waking (between 6:00 and 10:00 a.m.). Subjects were asked to abstain from caffeine-containing foods and beverages on the day before the test. Before the 17-min recording, the skin of the subject was cleaned and prepared for the attachment of surface electrodes (Blue Sensor, Medicotest Ltd, Ølstykke, Denmark). The electrodes of the ECG were placed in such a way not to prevent the installation of the HRM elastic electrode belt (T61, Polar Electro Oy). The electrode belt was placed just below the chest muscles with conductive gel being applied as described by the manufacturer.

The subject rested comfortably during the recording for at least 10 min in a supine position and 7 min in a standing position in a quiet, semidark laboratory room, maintained at a temperature of 19-21°C. To control the respiratory influence on HRV, the subjects matched their breathing frequency to an auditory metronome set at 0.20 Hz (12 breaths·min−1). No attempt was made to control the tidal volume.

Back to Top | Article Outline

Data Acquisition

R-R intervals were recorded simultaneously with a Polar S810 HRM and a two-lead ECG recorder (Physiotrace Estaris, Lille, France) at a sampling frequency of 1000 Hz for both devices.

R-wave peaks were detected automatically in the ECG series using a detection algorithm supplied by the manufacturer (Estaris). Following the recordings and storage of the raw ECG data, ECG signals were replayed to verify and validate visually each R-wave peak by an accustomed person. A vertical mark on the ECG indicated the detection of an R-wave. If the detection was incorrect, R-wave peak was determined manually by replacing the vertical marks on the correct R-wave peak. Subsequently, R-R intervals were exported under the ASCII format. The HRM signal was transferred to the Polar Precision Performance Software (release 3.00; Polar Electro Oy) and R-R intervals were exported under ASCII format.

Back to Top | Article Outline

Data Analysis

R-R interval comparison.

The ECG and HRM signals were synchronized for further analysis by marking the data using the temporal "event" marker available in both systems. Raw R-R intervals from both acquisition systems were edited and compared to discriminate error caused by the HRM acquisition or by a nonsinus beat. Nonsinus beats, which were present in both signals, were replaced by interpolated data derived from adjacent normal R-R intervals.

An error caused by the HRM acquisition was considered when the difference between ECG and HRM interval exceeded 20 ms (11). Then the HRM interval was labeled anomalous and later assigned to one of five identified error categories (11). A type 1 error was defined as a single point of discrepancy, either positive or negative between the ECG and HRM R-R interval. A type 2 error was defined as a long interval immediately followed by a short interval and the magnitude of the difference between the two ECG and HRM R-R intervals were similar. When a short interval was immediately followed by a long interval and the magnitude of the difference between the two ECG and HRM R-R intervals were similar, this error was defined as a type 3 error. A type 4 error was defined when the HRM R-R interval was equivalent to two or three ECG R-R intervals. Finally, a type 5 error occurred when the HRM detected two or more short R-R intervals, whereas the ECG detected one interval. Generally, the addition of these short intervals corresponded to the ECG interval.

To conserve time synchrony between the two data series and to allow the comparison between the ECG and the uncorrected HRM data, an ECG R-R interval of 0 ms was inserted when a type 5 error was present. On the contrary, a Polar R-R interval of 0 ms was inserted when a type 4 error was present. The correction algorithm for HRM data was the following: when a type 1 error was present, the R-R interval was replaced by interpolated value from the two adjacent R-R intervals. When type 2 or type 3 errors were present, the two uncorrected R-R intervals were averaged. When a type 4 error was present, the R-R interval was divided by two or three, according to the number of R-waves undetected. Finally, when a type 5 error occurred, anomalous HRM short R-R intervals were combined to approach the corresponding ECG value. Once noisy complexes were replaced, the signal was considered to be normal, and to provide normal-to-normal (NN) intervals.

Back to Top | Article Outline

Time domain analysis.

A corresponding segment of 256 s was selected within the last 300 s of the supine and standing corrected Polar and ECG recordings. The mean NN interval, the standard deviation of all NN intervals (SDNN), the root mean square of differences (RMSSD) of successive NN intervals, and the proportion of differences between adjacent NN intervals of more than 50 ms (pNN50) were computed.

Back to Top | Article Outline

Frequency domain analysis.

The same Polar corrected and ECG segments of 256 s were resampled at 2 Hz and detrended for subsequent analysis. As recommended by the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology (19), spectral analysis was performed with the fast Fourier transform (FFT) to quantify the power spectral density of the very low frequency (VLF; 0.00-0.04 Hz), the low frequency (LF; 0.04-0.15 Hz), and the high frequency (HF; 0.15-0.40 Hz) bands. Additional calculations included LF + HF, LF, and HF expressed in normalized unit (i.e., in a percentage of LF + HF) and the ratio LF:HF.

Back to Top | Article Outline

Quantitative beat-to-beat analysis.

The Poincaré plot is a scattergram in which each NN interval is plotted as a function of the previous one. The Poincaré plot provides both a qualitative and a quantitative analysis of HRV. The shape of the plot can be used to classify the signal into one of various classes (14,23), but also to fit an ellipse, which enables us to quantify the parameters SD1 and SD2. SD1 represents the dispersion of the points perpendicular to the line of identity, and it is thought to be an index of the instantaneous beat-to-beat variability of the data. SD2 represents the dispersion of the points along the line of identity, and represents the slow variability of heart rate (3,23).

Back to Top | Article Outline

Statistical Analysis

Standard statistical methods were used to calculate the means and standard deviations. Normal Gaussian distribution and homogeneity of variance were verified by the Shapiro-Wilk and the Levenne tests, respectively. Homoscedasticity was checked with a modified Levenne test. A paired t-test or, when appropriate, a Wilcoxon matched-pairs test, was used to detect the presence of a systematic difference in R-R interval or HRV indices calculated from both systems. Effect size (ES), which represents the ratio of the mean difference over the pooled variance (20), was used to estimate the magnitude of the difference. As proposed by Cohen (4), the difference was considered small when ES ≤ 0.2, moderate when ES ≤ 0.5, and great when ES > 0.8. Relative reliability, defined as the degree to which individuals maintain their position in a sample with repeated measurements (1), was assessed by the Pearson's product-moment correlation coefficient or, when appropriate, by the Spearman rank-order correlation. Finally, Bland-Altman plots of all measures from both systems were constructed and the 95% limits of agreement (LoA) were computed. As recommended by Bland-Altman (2), data were log-transformed before the calculation of the LoA when heteroscedasticity was present. Statistical significance was set at P = 0.05 level for all analysis. All calculations were made with Statistica (Release 6.0, Statsoft, Tulsa, OK).

Back to Top | Article Outline


The number of R-R intervals detected was 11,353 and 9,878 in supine and standing position, respectively, for the ECG, and 11,335 and 9,857 for the HRM. The degree and the type of error are described in Table 1. The t-test revealed that uncorrected and corrected R-R intervals were different from ECG R-R intervals in supine position (P < 0.05, ES = 0.025 and 0.000, respectively). Figures 1 and 2 represent Bland-Altman plots for combined ECG and uncorrected R-R intervals and the ECG and corrected R-R intervals. The correlations were 0.88 and 0.99 for the uncorrected and the corrected HRM R-R intervals with the ECG in supine position, respectively (P < 0.001). In standing position, coefficients of correlation with ECG R-R intervals were 0.88 and 0.99 for uncorrected and corrected HRM data, respectively (P < 0.001). No significant differences were noted for time domain, FFT, and Poincaré plot parameters obtained from the corrected Polar and ECG signals, except for RMSSD, SD1 in standing position (P < 0.05). The correlation of HRM with ECG parameters as well as the coefficient of variation, the bias, the 95% confidence interval for the bias, and the magnitude of the difference are presented in Tables 2 and 3.











Back to Top | Article Outline


This study compared raw data and the HRV parameters derived from a Polar S810 HRM and a two-lead ECG recorder. The present results demonstrate the HRM can provide HRV measurements consistent with an ECG recorder in healthy subjects during an active orthostatic test.

The error rate in detection of R-waves for Polar compared with the ECG system was 0.40%. This is in accordance with previous studies that reported a rate of 0.32-2.8% (8,17). The most common error occurring in the uncorrected HRM signal was a type 4 error (too few R-R intervals detected). It represented 75 and 56% of the total errors in supine and standing position, respectively. The origin of this error is not known, but a lack of contact between the skin and the elastic electrode belt could cause a decrease in R-wave amplitude and the inability to detect it. The type 5 error (too many R-R intervals detected) was the second most common error. It seems that this error results from multiple triggering during a single cardiac contraction. This error may have been caused by the HRM registering a T-wave, a P-wave registering as an R-wave, or both (11).

In this study, we observed a significant difference between the uncorrected HRM and the ECG R-R intervals in supine and standing position. The decrease of limits of agreement and the increase of the correlation coefficient after the correction of errors caused by the HRM acquisition demonstrates that correction protocols were successful when applied to the current data. It is worth noting that these correction protocols can be applied without ECG in regard to the frequency and the recognizable pattern of the errors caused by the Polar detection. In fact, the most common error (i.e., a type 4 error) involves anomalous R-R intervals two times longer than normal adjacent R-R intervals. In this way, the adapted correction protocol is easy to apply, even for the other error types. Nevertheless, this correction protocol remains to be validated. Besides, the difference observed between the corrected HRM and ECG R-R intervals remained significant. The many observations in supine and standing position (N = 11,353 and 9,878, respectively) may produce this significant statistical difference because the magnitude of the difference (i.e., the effect size) was very small (ES = 0.000 and 0.006, respectively). As already reported by Kingsley et al. (8), the bias was less than 1 ms in the current study. Our limits of agreement, however, were wider than those reported by Kingsley et al. (8) in resting condition (LoA: −5.2 to 5.89 ms, P < 0.05). This difference may be explained by the method of correction, because Kingsley et al. (8) excluded artifacts and nonsinus beats from the signal, whereas we corrected them in our study. Nevertheless, the very small magnitude of the difference, together with a good correlation between the HRM and ECG data (r > 0.99, P < 0.001), suggests that the HRM is a valid tool to measure R-R intervals. Beyond the fact that the use of the elastic electrode belt for the HRM could induce artifact, the small difference in the R-wave peak detection by the two devices may be caused by the different algorithm of detection used.

As reported by Radespiel-Troger et al. (16), we found a good correlation between time domain parameters estimated from HRM and ECG signals (r > 0.97, P < 0.05). No significant differences were found between parameters estimates, excepted for RMSSD in standing position. A possible explanation for this difference is that RMSSD reflects the short-term variability of the signal (19). Therefore, it is more sensitive to light variations in the R-R interval duration between the HRM and the ECG. Nevertheless, the correlation coefficient for this parameter between the two acquisition systems is good (r = 0.99, P < 0.05) and the magnitude of the difference is small (ES < 0.052). Generally, we note that this measurement error is negligible. If we consider that 3 wk of intensive training induce a significant decrease of RMSSD (from 22.10 ± 22.33 to 13.65 ± 17.44 ms), which corresponds to an effect size of 0.42 (15), we note that the measurement error by the HRM is reasonably good.

The calculated SD1 and SD2 for the HRM and the ECG signals were similar in supine position. In standing position, SD1 estimated from the HRM signal was significantly lower than SD1 obtained from the ECG signal (Table 3). SD1 represents the standard deviation of instantaneous beat-to-beat variability (23). Then, as RMSSD, SD1 is more sensitive to the slight variations in the successive R-R intervals duration between the two acquisition systems. Nevertheless, a good correlation (r > 0.99, P < 0.05), together with narrow LoA (Table 3), supports the validity of the HRM to realize a Poincaré plot analysis.

In the frequency domain, the VLF, LF, and HF components were almost identical (Tables 2 3). The observed differences for these parameters were not statistically significant (P > 0.05), whatever the position. The LoA in the present study were in accordance with the lower values reported by Kingsley et al. (i.e., 8 ms2 for LF and HF vs 10 ms2) (8). Indeed, the magnitude of difference lower than 0.2 for all frequency parameters confirmed this slight difference (4). Again, the measurement error is largely acceptable when compared with the effect size of 0.80 reported in a meta-analysis by Sandercock et al. (18) for HF after training in sedentary men.

In conclusion, narrow LoA, good correlations, and small effects size support the use of the Polar S810 HRM signal to measure HRV in supine position after data correction. Caution must be taken for its use in the standing position for the parameters sensitive to the short-term variability (i.e., RMSSD and SD1). Nevertheless, the slight differences obtained with the Polar S810 are negligible when compared with training or overtraining effects on HRV parameters. Moreover, the use of the same device during HRV studies may help avoid this difference.

Back to Top | Article Outline


1. Baugmarter, T. A. Norm-referenced measurement: reliability. In: Measurement Concepts in Physical Education and Exercise Science, M. J. Safrit and T. M. Wood. Champaign, IL: Human Kinetics 1989, pp. 45-72.
2. Bland, J. M., and D. G. Altman. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307-310, 1986.
3. Brennan, M., M. Palaniswami, and P. Kamen. Do existing measures of Poincare plot geometry reflect nonlinear features of heart rate variability? I.E.E.E. Trans. Biomed. Eng. 48:1342-1347, 2001.
4. Cohen, J. Statistical Power Analysis for the Behavioral Sciences. Hillsdale: Lawrence Erlbaum, 1988, pp. 1-599.
5. Dishman, R. K., Y. Nakamura, M. E. Garcia, R. W. Thompson, A. L. Dunn, and S. N. Blair. Heart rate variability, trait anxiety, and perceived stress among physically fit men and women. Int. J. Psychophysiol. 37:121-133, 2000.
6. Friedman, B. H., and J. F. Thayer. Autonomic balance revisited: panic anxiety and heart rate variability. J. Psychosom. Res. 44:133-151, 1998.
7. Hedelin, R., U. Wiklund, P. Bjerle, and K. Henriksson-Larsen. Cardiac autonomic imbalance in an overtrained athlete. Med. Sci. Sports Exerc. 32:1531-1533, 2000.
8. Kingsley, M., M. J. Lewis, and R. E. Marson. Comparison of polar 810 s and an ambulatory ECG system for RR interval measurement during progressive exercise. Int. J. Sports Med. 26:39-44, 2005.
9. Kinnunen, H., and I. Heikkila. The timing accuracy of the Polar Vantage NV heart rate monitor. J. Sports Sci. 16:S107-S110, 1998.
10. Levy, W. C., M. D. Cerqueira, G. D. Harp,et al. Effect of endurance exercise training on heart rate variability at rest in healthy young and older men. Am. J. Cardiol. 82:1236-1241, 1998.
11. Marchant-Forde, R. M., D. J. Marlin, and J. N. Marchant-Forde. Validation of a cardiac monitor for measuring heart rate variability in adult female pigs: accuracy, artifacts and editing. Physiol. Behav. 80:449-458, 2004.
12. Melanson, E. L., and P. S. Freedson. The effect of endurance training on resting heart rate variability in sedentary adult males. Eur. J. Appl. Physiol. 85:442-449, 2001.
13. Molgaard, H., K. E. Sorensen, and P. Bjerregaard. Attenuated 24-h heart rate variability in apparently healthy subjects, subsequently suffering sudden cardiac death. Clin. Auton. Res. 1:233-237, 1991.
14. Mourot, L., M. Bouhaddi, S. Perrey,et al. Decrease in heart rate variability with overtraining: assessment by the Poincare plot analysis. Clin. Physiol. Funct. Imaging. 24:10-18, 2004.
15. Pichot, V., F. Roche, J. M. Gaspoz, et al. Relation between heart rate variability and training load in middle-distance runners. Med. Sci. Sports Exerc. 32:1729-1736, 2000.
16. Radespiel-Troger, M., R. Rauh, C. Mahlke, T. Gottschalk, and M. Muck-Weymann. Agreement of two different methods for measurement of heart rate variability. Clin. Auton. Res. 13:99-102, 2003.
17. Ruha, A., S. Sallinen, and S. Nissila. A real-time microprocessor QRS detector system with a 1-ms timing accuracy for the measurement of ambulatory HRV. I.E.E.E. Trans. Biomed. Eng. 44:159-167, 1997.
18. Sandercock, G. R., P. D. Bromley, and D. A. Brodie. Effects of exercise on heart rate variability: inferences from meta-analysis. Med. Sci. Sports Exerc. 37:433-439, 2005.
19. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability: standards of measurement, physiological interpretation and clinical use. Circulation 93:1043-1065, 1996.
20. Thomas, J. R. Research methods in physical activity. In: Human Kinetics, Champaign, IL: Human Kinetics 2001, pp. 1-449.
21. Tsuji, H., F. J. Venditti, E. S. Jr., J. C. Manders, M. G. Evans, C. L. Larson, and D. Feldman. Reduced heart rate variability and mortality risk in an elderly cohort. The Framingham Heart Study. Circulation 90:878-883, 1994.
22. Tulppo, M. P., A. J. Hautala, T. H. Makikallio, et al. Effects of aerobic training on heart rate dynamics in sedentary subjects. J. Appl. Physiol. 95:364-372, 2003.
23. Tulppo, M. P., T. H. Makikallio, T. E. Takala, T. Seppanen, and H. V. Huikuri. Quantitative beat-to-beat analysis of heart rate dynamics during exercise. Am. J. Physiol. 271:H244-H252, 1996.


©2006The American College of Sports Medicine