Comparison of novel versus commercial HRV softwares CHAPTER 11 COMPARISON OF NOVEL VERSUS COMMERCIAL HRV SOFTWARES 174 Comparison of novel versus commercial HRV softwares 175 Introduction In this chapter, foetal HRV data derived from F-EXTRACT was compared to that obtained from a commercial HRV software available for public use. This commercial HRV software was the Nevrokard HRV System (Medistar Inc., Slovenia). Differences in the algorithms used in these systems for the computation and derivation of HRV, and for rejection and replacement of ectopic beats and other artifacts, may lead to variations in the HRV data generated from the same ECG recordings. Jung J et al. (Jung J et al., 1996) has shown that the results generated by different commercial HRV systems on the same Holter ECG recordings were significantly different. Method 2.1 Nevrokard system description/operation The Nevrokard HRV system is a commercial software package designed to perform analysis of ECG data in the time and frequency domains. It is able to analyze data series in single-column format such as those which were earlier exported (and stored in the computer hard disk drive) from FEMO with the extension *.frr. The basic steps for HRV analysis on the Nevrokard system were as follows: The Nevrokard system was first started and data format to be analyzed was selected to be *.frr. This additional step was required as the software also recognizes other file formats. After selecting a particular data file from the stored data files, the graph of RR-interval versus time for the selected file then appeared on the screen. Any missing Comparison of novel versus commercial HRV softwares 176 or spurious beats may be viewed (Figure 11-1). The Nevrokard software does not contain any automated algorithms to detect and correct spurious beats, although it does allow the operator to manually modify any suspicious beats by dragging it with the mouse key to the correct location (Figure 11-2). For example, say the RRintervals of a particular recording were all around the value of 400 ms except for one which was found to be 900 ms, one can then move the beat from 900 ms to 400 ms before proceeding with HRV analysis. The next step was to select an appropriate interval of time free of erroneous beats for performing the HRV evaluation. This was done simply by setting the start and end points of the selected duration. The software is only able to analyze epoch lengths of 2x, where x is any interger ≥ 6. Hence, the shortest analyzable epoch is 64 seconds (s), the next length being 128 s, followed by 256 s, 512 s, 1028 s and so on. If other epoch lengths were selected using the mouse keys, the system would automatically choose the nearest available lengths of 2x. For example, if 300 s or 800 s of data were manually selected, then the system would analyze the first 256 s and 512 s of data, respectively. In this study, epoch lengths of 256 s were analyzed for reasons mentioned in Chapter 10. One main difference between the two HRV systems is that the Nevrokard system plots the RR-interval against time whereas F-EXTRACT plots it against beat number. Thus, after selecting a period of 256 s on the Nevrokard system, its corresponding period in terms of beat number was to be selected on F-EXTRACT. Comparison of novel versus commercial HRV softwares 177 Figure 11-1: A screen on Nevrokard HRV software showing the graph of foetal RR-interval versus time. Comparison of novel versus commercial HRV softwares 178 ← (a) ↓ (b) Figure 11-2: An artifact RR-interval (indicated by red arrow) measuring 900 ms (a) is dragged by the cursor to an estimated value of 400 ms (b) on the Nevrokard program. Comparison of novel versus commercial HRV softwares 179 This was done by integrating into F-EXTRACT an algorithm that allowed the conversion of the selected period’s start and end points in seconds to the corresponding start and end points in beat numbers. This ensured that both the Nevrokard and F-EXTRACT performed the HRV analyses on the same selected period of RR-intervals. For F-EXTRACT, the results of the time- and frequencydomain analyses on the selected epoch were generated by using programming commands that included entering the start and end beat numbers on the MatLab software. The details of operating F-EXTRACT are described in Chapter 9. As for the Nevrokard system, after selection of the RR-interval epoch, timedomain and frequency-domain analyses were performed. The information in the generated output for time-domain analysis is shown in Table 11-1. Figure 11-3 shows the generated FFT spectrum, as well as the values of absolute and normalized VLF, LF and HF power, and the LF/HF ratio. The Nevrokard software allows the frequency ranges analyzed to be defined by the operator. The frequency bands selected for foetal HRV analyses were the same for both HRV systems, i.e., very low frequency (VLF: 0.003-0.04 Hz), low frequency (LF: 0.04-0.15 Hz) and high frequency (HF: 0.15-1.0 Hz). Other frequency-domain parameters compared were total power, LF and HF power in normalized units, as well as the LF/HF ratio. In summary, after selection of the corresponding 256 s epoch and elimination of ectopic and artifact beats, foetal HRV was evaluated by both the Nevrokard and FEXTRACT. In total, 374 foetal ECG (fECG) recordings performed on 18 to 41 week- Comparison of novel versus commercial HRV softwares Table 11-1: The time-domain statistics displayed by Nevrokard software: Duration Time Range from/to No. of Samples Maximum Minimum Max./Min. Range Mean Median 95 % Conf. Interval 99 % Conf. Interval Coef. of Variance Variance Std. Deviation (SDNN) Std. Error (SENN) SDSD RMSSD NN50 Count or NN27 Count (Fetal NN) pNN50 or pNN27 (Fetal NN) 180 Comparison of novel versus commercial HRV softwares 181 Figure 11-3: A screen on Nevrokard HRV software showing the HRV power spectrum and calculated VLF, LF and HF power in absolute and normalized units. Comparison of novel versus commercial HRV softwares 182 old foetuses were subjected to analysis by both HRV systems, thereby generating two sets of HRV data from each fECG recording. Each set of HRV data included five time-domain parameters (fHR, mNN, SDNN, rMSSD and pNN27) and six frequencydomain parameters (absolute LF power, normalized LF power, absolute HF power, normalized HF power, LF/HF ratio and total power). The Nevrokard and FEXTRACT systems were compared using the HRV data determined from their algorithms. Statistics To evaluate the level of agreement between the Nevrokard and F-EXTRACT HRV systems, the Bland-Altman technique (Bland JM and Altman DG, 2003, 1999) was performed using Prism 4.03 for Windows (GraphPad Software Inc., San Diego, CA, USA). The Bland-Altman technique evaluates the agreement between two methods of measurement and consists of a graphical presentation where the differences between the two measurements (Y-axis) is plotted against their average values (Xaxis). In this study, HRV analyses were performed using Nevrokard and FEXTRACT systems on each of the 374 fECG recording. The differences in the values of each time- and frequency-domain HRV measurement generated by these two systems were plotted against their average values. For each HRV variable, the absolute difference was calculated by subtracting values obtained by F-EXTRACT Comparison of novel versus commercial HRV softwares 183 (F) from those obtained by Nevrokard (N), i.e., (N-F). This absolute difference was then plotted against the average value of N and F, i.e., N+F . In the Bland-Altman plot, the middle solid line running horizontally across the plot represents the mean of (N-F) throughout the range of mean HRV measurements. For reference, a dotted line is drawn at y=0, which represents zero difference between N and F. The upper and lower solid lines on the plots represent the upper and lower 95% limits of agreement (LoA), which delineate the range within which 95% of the differences lie. The 95% LoA were computed from the equation: đ ± 1.96 SD, where đ is the mean difference and SD is the standard deviation of the differences. The Bland-Altman plot allows the visualization of how big is the discrepancy or bias between the two methods, whether there is any consistent relation or systemic bias between the difference and the mean, and whether the assumption of constant standard deviation of the differences is satisfied. When the data points on the Bland-Altman plot spread out as the magnitude of measurements increase, the assumption of constant standard deviation is violated. A logarithmic or percent Y-scale, which is computed by log (N-F) and {[(N-F) ÷ (N+F)/2] x 100}, respectively, is thus recommended (Bland JM and Altman DG, 1999). In this study, the percent plot is preferred because it is easily interpreted as numbers can be read directly from the plot without the need for back logtransformation (Dewitte K et al., 2002). Another advantage of the percent plot is that it gives an indication of the magnitude of the difference with reference to its mean Comparison of novel versus commercial HRV softwares 187 measured by F-EXTRACT. As such, the calculated LF/HF ratios were significantly lower (averaging about one-sixth) than the LF/HF ratios calculated by F-EXTRACT. 4.2 Comparison of time-domain parameters between Nevrokard and F-EXTRACT using Bland-Altman method Differences between foetal HRV measured by the Nevrokard and FEXTRACT systems were analyzed using the Bland-Altman method. Table 11-2 displays the mean differences and LoA of HRV variables while Table 11-3 displays the mean percent difference and the percent LoA of HRV variables. Figure 11-6 shows the comparison of differences in time-domain HRV parameters using the two HRV systems. For fHR, the mean difference (N-F) was -1.5 bpm (-1.3%), suggesting that the Nevrokard system generally measured 1.5 bpm (1.3%) lower than FEXTRACT. The upper and lower LoA were -20.2 bpm (-15.6%) and 17.2 bpm (12.9%), which means that majority (95%) of the differences in fHR measured by the Nevrokard system could range from being 20.2 bpm (15.6%) less than to 17.2 bpm (12.9%) more than that measured by F-EXTRACT. Similar to fHR, the mNN measured by the two HRV systems did not vary much, as shown by the mean difference (N-F) of 7.1 ms, which is a negligible 1.3% difference. The LoA ranged from -61.1 ms (-13%) to 75.3 ms (15.7%) (Tables 11-2 and 11-3). Figure 11-6 shows that as the values of SDNN, rMSSD and pNN27 increased, there was an increase in the differences. Figure 11-6 also shows an increase in the standard deviation of the differences (indicated by the spreading out of Comparison of novel versus commercial HRV softwares 188 Table 11-2: Bland-Altman analysis (mean difference) of time- and frequencydomain variables Timedomain fHR (bpm) mNN (ms) SDNN (ms) rMSSD (ms) pNN27 (%) Mean difference (N-F) -1.5 7.1 20.6 30.3 13.2 Frequency -domain LF power (ms2) LF norm (n.u.) HF power (ms2) HF norm (n.u.) LF/HF ratio Total power (ms2) -53.4 -33.1 910.2 37.3 -1.2 1236.4 HRV parameter N- measurements obtained from Nevrokard F- measurements obtained from F-EXTRACT 9.5 34.8 25.7 19.0 11.4 Lower limit of agreement -20.2 -61.1 -29.7 -6.9 -9.1 Upper limit of agreement 17.2 75.3 70.9 67.4 35.5 310.7 15.4 869.0 20.0 1.2 2199.7 -662.5 -63.2 -793.1 -1.8 -3.6 -3075.0 555.7 -2.9 2613.5 76.4 1.1 5547.8 SD of difference 189 Comparison of novel versus commercial HRV softwares Table 11-3: Bland-Altman analysis (mean percent difference) of time- and frequency-domain variables Timedomain fHR (bpm) mNN (ms) SDNN (ms) rMSSD (ms) pNN27 (%) Mean % difference (N-F) -1.3 1.3 47.9 87.9 79.0 Frequency -domain LF power (ms2) LF norm (n.u.) HF power (ms2) HF norm (n.u.) LF/HF ratio Total power (ms2) -36.0 -92.9 101.6 61.0 -131.3 45.2 HRV parameter 7.3 7.3 47.0 45.8 67.6 Lower limit of agreement -15.6 -13.0 -44.3 -1.8 -53.6 Upper limit of agreement 12.9 15.7 140.1 177.6 211.6 63.1 33.6 56.1 28.9 39.4 55.2 -159.7 -158.8 -8.3 4.4 -208.5 -62.9 87.7 -27.0 211.5 117.6 -54.1 153.3 SD of difference Mean difference, SD of bias and limits of agreements are in %. N- measurements obtained from Nevrokard F- measurements obtained from F-EXTRACT 190 Comparison of novel versus commercial HRV softwares ms 50 D if f e r e n c e in m N N D Iffer en ce in fH R bpm 25 -25 -50 90 160 120 80 40 -40 -80 -120 110 130 150 170 400 190 ms 150 100 50 -50 -100 25 50 75 700 50 -50 600 100 D Iffer en ce in r M S S D D ifference in S D N N ms 500 Average mNN (ms) Average fHR (bpm) 100 125 25 50 75 100 Average rMSSD (ms) Average SDNN (ms) D iffer en ce in p N N 27 % 60 40 20 -20 -40 10 20 30 40 50 Average pNN27 (%) Figure 11-6: Bland-Altman plots of absolute difference (N-F) against average for time-domain variables (ms). The middle solid line represents mean bias, surrounded by upper and lower limits of agreement. N and F= measurements obtained from Nevokard and F-EXTRACT, respectively. Comparison of novel versus commercial HRV softwares 191 data points) with increasing values of SDNN, rMSSD and pNN27. As recommended, Bland-Altman plots of percent difference were plotted when the criteria of constant standard deviation is not met (Figure 11-7). For SDNN, the mean percent difference was 47.9% and the LoA ranged from -44.3% to 140.1%. This indicates that the values of SDNN measured by the Nevrokard system were nearly 50% higher than those measured by the F-EXTRACT system. The rMSSD values measured by the Nevrokard system were nearly 88% higher than those measured by the F-EXTRACT system with LoA ranging from –1.8% to177.6%. As for pNN27, the values measured by the Nevrokard system were 79% higher than those measured by the F-EXTRACT system and the LoA ranged from –53.6% to 211.6% (Table 11-3). 4.3 Comparison of frequency-domain parameters between Nevrokard and F-EXTRACT using Bland-Altman method Figures 11-8 and 11.9 display the Bland-Altman plots of absolute and percent differences, respectively, in the frequency-domain HRV variables, LF and HF power, normalized LF and HF power, total power and LF/HF ratio. Table 11-3 displays the mean percent difference and the percent LoA of these HRV variables. From Figure 11-8, similar to the plots of SDNN, rMSSD and pNN27, the bias and its standard deviation widened as the magnitude of the frequency-domain HRV variables increased. The mean difference for LF power was -36.0% (LoA= -159.7% to 87.7%), indicating that Nevrokard system gave a value averaging 36% lower than FEXTRACT. The HF power on the other hand, was 101.6% higher (LoA= -8.3% to 211.5%) when measured by the Nevrokard system. The lower LF and higher HF 192 Comparison of novel versus commercial HRV softwares 40 % Difference in m NN % D ifference in fH R 40 20 -20 -40 90 110 130 150 170 190 30 20 10 -10 -20 -30 -40 300 Average fHR (bpm) 600 700 250 % D Ifference in rM S S D % D ifference in S D N N 500 Average mNN (ms) 250 200 150 100 50 -50 -100 -150 400 200 150 100 50 -50 -100 25 50 75 100 125 25 50 75 100 Average rMSSD (ms) Average SDNN (ms) % D ifference in pN N 27 300 200 100 -100 10 20 30 40 50 Average pNN27 (%) Figure 11-7: Bland-Altman plots of percent difference (N-F) against average for timedomain variables (ms). The middle solid line represents mean percent bias, surrounded by upper and lower limits of agreement. N and F= measurements obtained from Nevokard and F-EXTRACT, respectively. 193 Comparison of novel versus commercial HRV softwares ms2 n.u. 1000 -1000 -2000 50 D iffer en ce in L F n o r m D if f e r e n c e in L F p o w e r 2000 500 -50 -100 1000 1500 2000 (ms ) Average LF power 25 50 75 100 (n.u.) Average normalized LF power n.u. ms2 150 D iffer en ce in H F n o r m D if fe r e n c e in H F p o w e r 5000 100 2500 -2500 1000 3000 4000 (ms ) Average HF power ms2 D if f e r e n c e in t o t a l p o w e r D ifference in LF/H F -5 Average LF/HF -50 2000 50 150 (n.u.) Average normalized HF power 100 50 15000 10000 5000 -5000 -10000 2000 4000 6000 8000 10000 Average total power (ms ) Figure 11-8: Bland-Altman plots of absolute difference (N-F) against average for frequency-domain variables. The middle solid line represents mean bias, surrounded by upper and lower limits of agreement. N and F= measurements obtained from Nevokard and F-EXTRACT, respectively. 194 % Difference in LF norm 200 100 -100 -200 1000 500 -300 2000 1500 -50 -100 -150 -200 200 150 100 50 -50 1000 2000 3000 4000 % Difference in total pow er % Difference in LF/HF -50 -100 -150 -200 Average LF/HF 75 100 150 100 50 -50 50 100 150 Average normalized HF power (n.u.) 50 50 200 Average HF power (ms ) -250 25 Average normalized LF power (n.u.) 250 -100 50 Average LF power (ms ) % Difference in HF norm % Difference in HF power % Difference in LF pow er Comparison of novel versus commercial HRV softwares 300 200 100 -100 -200 2000 4000 6000 8000 10000 Average total power (ms ) Figure 11-9: Bland-Altman plots of percent difference (N-F) against average for frequency-domain variables (ms). The middle solid line represents mean percent bias, surrounded by upper and lower limits of agreement. N and F= measurements obtained from Nevokard and F-EXTRACT, respectively. Comparison of novel versus commercial HRV softwares 195 power measured by the Nevrokard system gave rise to low values of LF/HF ratios (131.3% lower than those obtained by F-EXTRACT). The total power, which is a summation of the LF and HF power, was 45.2% higher (LoA= -62.9% to 153.3%) when measured by the Nevrokard system. Normalized units of LF and HF power were 92.9% lower and 61.0% higher, respectively, when measured by the Nevrokard system as compared to those measured by F-EXTRACT. The LoA of normalized LF ranged from -58.8% to 27.0% while that of HF power ranged from from 4.4% to 117.4% (Table 11-3). Discussion In this study, the agreement between two HRV systems was evaluated using the Bland-Altman technique of comparing between methods of measurement. From the Bland-Altman plots, it can be seen that the two HRV softwares agree well (bias =1.3%) for fHR and mNN measurements. This may be because both fHR and mNN are relatively simple measurements that require little or no data processing. The mean difference in mNN of 7.1 ms was relatively similar to that obtained by other studies that compared different HRV systems on adult subjects. Jung J et al. (Jung J et al., 1996) assessed the agreement of different commercially available HRV systems on the same 24-hour Holter tape recording and obtained an inter-system difference ranging from ms to 100 ms in mNN. Sandercock GR et al. (Sandercock GR et al., 2004) used different HRV systems to analyze simultaneously-recorded short-term ECG on 30 healthy adults and found that the bias in mNN measurements between Comparison of novel versus commercial HRV softwares 196 systems ranged from 0.6 ms to 7.0 ms in the supine position. Yi G et al. (Yi G et al., 2000), who studied the differences between different HRV systems on the same 24hour Holter tape in survivors of acute myocardial infarction (MI), found a maximum mean difference of 1.8 ms in mNN measures between systems. Besides fHR and mNN, all other HRV measures show poor agreement between the two softwares (bias >30% and wide LoA). The discrepancy between the two softwares widened with increasing values of HRV measurement. This is not surprising since the values of most HRV variables were approximately doubled when measured by Nevrokard as compared to F-EXTRACT. Increasing values of x, 3x, and 5x measured by F-EXTARCT will be measured as 2x, 6x and 10x, respectively, by Nevrokard. And their respective inter-system differences will be x (2x-x), 3x (6x-3x) and 5x (10x-5x), which increase as the measured HRV values increase. The mean difference in SDNN measured by the two HRV systems was found to be 21 ms in this study. The maximum mean inter-system differences in SDNN were observed to be 36 ms, 9.9 ms and 4.4 ms by Jung J et al. (Jung J et al., 1996), Sandercock GR et al. (Sandercock GR et al., 2004) and Yi G et al. (Yi G et al., 2000), respectively. The mean difference in rMSSD in this study was high (30 ms) relative to the maximum mean inter-system differences in rMSSD obtained by Jung (12 ms), Sandercock (2.2 ms) and Yi et al. (11.1 ms). Our mean difference in pNN27 (13%) was also higher than the maximum mean difference in pNN50 between HRV systems used by Jung (3%). Such wide variations in differences may be explained by Comparison of novel versus commercial HRV softwares 197 dissimilar subjects and durations of ECG sampled. Jung and Yi examined 24-hour HRV whereas Sandercock examined short-term HRV of 300 s. Adult HRV were measured in the other studies whereas foetal HRV was measured in this study. Sandercock’s subjects consisted of 30 healthy adults, Yi’s subjects were 26 post-MI patients, whilst Jung’s subjects included patients with aortic valve disease and healthy volunteers. Generally, discrepancies in HRV measurements obtained from different ECG recordings and analysis systems can arise due to differences in ECG recording and/or differences in analogue-to-digital (A/D) conversion of the ECG signal. Since the ECG recording and A/D conversion had been performed by the same equipment (FEMO), and the same strip of recording was used for HRV analysis, the reason for the incomparable results from the Nevrokard and F-EXTRACT could be due to differences in the HRV analysis algorithms and/or differences in the processing of erroneous beats. Due to the fact that both HRV systems applied the same mathematical formulae for time-domain analysis, as well as the same FFT technique for frequency-domain analysis, the disparity in the HRV results generated by Nevrokard and F-EXTRACT is most likely due to the different system-dependent processing of erroneous beats. The Nevrokard system is manufactured for analysis of HRV in adults, whose RR-intervals contain considerably less errors as compared to foetal RR-intervals. Errors in RR-interval data occur when R peaks are misidentified. This can arise when Comparison of novel versus commercial HRV softwares 198 the recording system either fails to mark an R peak or mistakenly marks a nonexistent R peak. Interference of skin surface conductivity and muscle action potentials can further introduce errors. Abdominal foetal ECG recording may encounter the problem of low voltages of foetal ECG signals due to the great distance of foetal heart from the recording electrodes, as well as reduced conductivity (due to the vernix layer). Foetal movements, which are beyond the control of the mother or investigator, may also influence the ECG signals. These problems are not encountered during short-term HRV recording in adults. As such, the algorithm for artifact-correction in softwares such as the Nevrokard system, which are used for analyzing adult HRV, may not be as robust as required for the analysis of foetal HRV. The F-EXTRACT system was specially developed for foetal HRV analysis. It has an automated correction algorithm that filtered beats that deviated from the previous qualified interval by more than 30% and replaced them by linearly-interpolated beats. In addition to this, a filter with upper and lower threshold limits of fHRs of 170 and 110 bpm was incorporated to eliminate extreme values of foetal RR-interval durations. On the contrary, the Nevrokard system does not contain any algorithms for automatic correction of ectopic and artifact beats, although it does allow the operator to edit extreme RR-interval values by a click-and-drag motion to a rough estimated value (not by linear interpolation). It has been demonstrated that uncorrected RR-interval data exhibited much higher variability than corrected ones, as shown in the greater magnitude of all HRV parameters (Marchant-Forde RM et al., 2004). Thus, in the F-EXTRACT system, Comparison of novel versus commercial HRV softwares 199 stricter criteria in artifact-processing resulted in a higher elimination rate of RRintervals. This may explain the lower HRV values (approximately half) obtained by F-EXTRACT as compared to those obtained by the Nevrokard system. Although variability of time-domain indices measured by the Nevrokard system was approximately doubled, the frequency-domain indices did not all show equal increases in spectral power. The HF power obtained by the Nevrokard system was 102% higher while LF power was 36% lower than that obtained by the FEXTRACT system. HRV frequency-domain analysis is known to be much more susceptible than time-domain analysis to artifacts in the RR-interval data. Although such artifacts yield a broadband increase in spectral power, this increase in spectral power may not distribute equally across the frequency bands (Task Force, 1996). Similarly, other HRV studies that performed frequency-domain analysis have observed greater variability and disparity in frequency-domain indices than in timedomain parameters amongst different HRV algorithms used. Unacceptable LoA between different HRV systems have been reported for all frequency-domain indices especially in HF power and LF/HF ratio (Sandercock GR et al., 2004; Jung J et al., 1996). This may be because the HF power was most affected by the inclusion of artifacts in the RR-interval data (Marchant-Forde RM et al., 2004), where even a single artifact was found to considerably increase the HF variability (Berntson GG and Stowell JR, 1998). Another study reported that an artifact-correction algorithm whereby non-sinus beats were deleted and replaced with beats calculated by linear Comparison of novel versus commercial HRV softwares 200 interpolation markedly reduced HF power by 402% (Storck N et al., 2001). Since this is similar to the artifact-correction algorithm used by F-EXTRACT, it may explain the much-lowered HF values generated by F-EXTRACT as compared to Nevrokard. Of all the HRV spectral indices examined in this study, LF/HF ratio measured by the two HRV systems demonstrated the greatest disparity. This may be because the LF/HF ratio is dependent on measurement of both LF and HF power. Small variations in both of these values may be summated to affect the LF/HF ratio. The much higher HF power and slightly lower LF power obtained by the Nevrokard resulted in the calculation of a very much-lowered LF/HF value by the Nevrokard system. Although the HRV values obtained by the Nevrokard system were higher than and incomparable to those obtained by the F-EXTRACT system, they showed a fairly similar trend of change with gestational age. For example, the values of LF/HF obtained by both HRV systems similarly decreased with foetal age. It was observed that in the foetal HRV spectrograms processed from abdominal foetal ECG recordings of 36 to 40 week-old foetuses, the HF power was approximately 25% that of total power (Groome LJ et al., 1994a; Ferrazzi E et al., 1989). The values obtained by the F-EXTRACT system concurred with this observation. On the other hand, as measured by the Nevrokard system, the HF power was >50% of the total power. In addition to this, the integration of proper and Comparison of novel versus commercial HRV softwares 201 systematic artifact-correction algorithm in the F-EXTRACT system may help to reduce errors in the RR-intervals data extracted from foetal ECG recordings. To date, there are no established standards in artifact processing and spectral analysis (e.g. FFT or autoregressive methods) for the proper assessment of HRV. Consequently, different manufacturers of HRV systems utilize their own algorithms in artifact processing and for HRV analysis, which results in different values of HRV indices generated by different systems. Future guidelines for accurate HRV evaluation should preferably include recommendations for standardized algorithms for optimal artifact processing as well as for HRV analysis. Summary This chapter evaluated the agreement of F-EXTRACT and Nevrokard HRV systems using Bland-Altman analysis. With the exception of fHR and mNN, which show negligible difference between the two systems, the measurements of all other HRV parameters show poor agreement between them. The Nevrokard system generally produced HRV measurements that were approximately twice the magnitude of those measured by the F-EXTRACT system. This resulted in higher values of time- and frequency-domain HRV indices when the Nevrokard system was used. The disparity in the HRV values may be due to the different levels of accuracy in removal of non-sinus beats. This demonstrates the importance of the processing of artifact beats, especially in foetal RR-intervals where they may occur quite frequently. There is currently no accepted “gold standard” in the technique of determination of foetal Comparison of novel versus commercial HRV softwares 202 HRV. Commercial HRV systems are manufactured for processing adult RR-intervals and analyzing adult HRV. Thus, their correction algorithms for RR-intervals artifacts may not be sufficient for the high level of artifacts observed in the RR-intervals from abdominal fECG recordings. F-EXTRACT incorporated strict artifact-correction algorithms that eliminated most errors in the foetal RR-interval data, and thus it may be more suitable for the analysis of foetal HRV as compared to commercial HRV softwares, which not take into account the above-mentioned considerations. [...]... Normalized units of LF and HF power were 92.9% lower and 61.0% higher, respectively, when measured by the Nevrokard system as compared to those measured by F-EXTRACT The LoA of normalized LF ranged from -58.8% to 27.0% while that of HF power ranged from from 4. 4% to 117 .4% (Table 11-3) 5 Discussion In this study, the agreement between two HRV systems was evaluated using the Bland-Altman technique of. .. Bland-Altman plots of absolute and percent differences, respectively, in the frequency-domain HRV variables, LF and HF power, normalized LF and HF power, total power and LF/HF ratio Table 11-3 displays the mean percent difference and the percent LoA of these HRV variables From Figure 11-8, similar to the plots of SDNN, rMSSD and pNN27, the bias and its standard deviation widened as the magnitude of. .. addition to this, a filter with upper and lower threshold limits of fHRs of 170 and 110 bpm was incorporated to eliminate extreme values of foetal RR-interval durations On the contrary, the Nevrokard system does not contain any algorithms for automatic correction of ectopic and artifact beats, although it does allow the operator to edit extreme RR-interval values by a click -and- drag motion to a rough... Interference of skin surface conductivity and muscle action potentials can further introduce errors Abdominal foetal ECG recording may encounter the problem of low voltages of foetal ECG signals due to the great distance of foetal heart from the recording electrodes, as well as reduced conductivity (due to the vernix layer) Foetal movements, which are beyond the control of the mother or investigator, may... data points) with increasing values of SDNN, rMSSD and pNN27 As recommended, Bland-Altman plots of percent difference were plotted when the criteria of constant standard deviation is not met (Figure 11-7) For SDNN, the mean percent difference was 47 .9% and the LoA ranged from -44 .3% to 140 .1% This indicates that the values of SDNN measured by the Nevrokard system were nearly 50% higher than those measured... than and incomparable to those obtained by the F-EXTRACT system, they showed a fairly similar trend of change with gestational age For example, the values of LF/HF obtained by both HRV systems similarly decreased with foetal age It was observed that in the foetal HRV spectrograms processed from abdominal foetal ECG recordings of 36 to 40 week-old foetuses, the HF power was approximately 25% that of total... 1.3 47 .9 87.9 79.0 Frequency -domain LF power (ms2) LF norm (n.u.) HF power (ms2) HF norm (n.u.) LF/HF ratio Total power (ms2) -36.0 -92.9 101.6 61.0 -131.3 45 .2 HRV parameter 7.3 7.3 47 .0 45 .8 67.6 Lower limit of agreement -15.6 -13.0 -44 .3 -1.8 -53.6 Upper limit of agreement 12.9 15.7 140 .1 177.6 211.6 63.1 33.6 56.1 28.9 39 .4 55.2 -159.7 -158.8 -8.3 4. 4 -208.5 -62.9 87.7 -27.0 211.5 117.6 - 54. 1... surrounded by upper and lower limits of agreement N and F= measurements obtained from Nevokard and F-EXTRACT, respectively Comparison of novel versus commercial HRV softwares 195 power measured by the Nevrokard system gave rise to low values of LF/HF ratios (131.3% lower than those obtained by F-EXTRACT) The total power, which is a summation of the LF and HF power, was 45 .2% higher (LoA= -62.9% to 153.3%)... 34. 8 25.7 19.0 11 .4 Lower limit of agreement -20.2 -61.1 -29.7 -6.9 -9.1 Upper limit of agreement 17.2 75.3 70.9 67 .4 35.5 310.7 15 .4 869.0 20.0 1.2 2199.7 -662.5 -63.2 -793.1 -1.8 -3.6 -3075.0 555.7 -2.9 2613.5 76 .4 1.1 5 547 .8 SD of difference 189 Comparison of novel versus commercial HRV softwares Table 11-3: Bland-Altman analysis (mean percent difference) of time- and frequency-domain variables Timedomain... were presented In Figures 11 -4 and 11-5, the bar charts display the means of time- and frequency-domain HRV parameters obtained by the two systems and in relation to foetal gestational age It can be seen in Figure 11 -4 that the foetal heart rate (fHR) and mNN measurements obtained by the two HRV systems were very similar The values of the other time-domain variables (SDNN, rMSSD, pNN27) however, were approximately . means of time- and frequency-domain HRV parameters obtained by the two systems and in relation to foetal gestational age. It can be seen in Figure 11 -4 that the foetal heart rate (fHR) and mNN. after selection of the corresponding 256 s epoch and elimination of ectopic and artifact beats, foetal HRV was evaluated by both the Nevrokard and F- EXTRACT. In total, 3 74 foetal ECG (fECG). systems for the computation and derivation of HRV, and for rejection and replacement of ectopic beats and other artifacts, may lead to variations in the HRV data generated from the same ECG recordings.