Assessment and quantification of foetal electrocardiography and heart rate variability of normal foetuses from early to late gestational periods 4

3 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

CHAPTER 11 COMPARISON OF NOVEL VERSUS COMMERCIAL

HRV SOFTWARES

1 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 4 different commercial HRV systems on the same Holter ECG recordings were significantly different

2 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

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 RR-intervals 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

Figure 11-1: A screen on Nevrokard HRV software showing the graph of foetal RR-interval versus time

(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

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 frequency-domain 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, domain 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,

time-LF and HF power, and the time-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 EXTRACT In total, 374 foetal ECG (fECG) recordings performed on 18 to 41 week-

F-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)

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

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 frequency-domain parameters (absolute LF power, normalized LF power, absolute HF power, normalized HF power, LF/HF ratio and total power) The Nevrokard and F-EXTRACT systems were compared using the HRV data determined from their algorithms

3 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 (X-axis) In this study, HRV analyses were performed using Nevrokard and F-EXTRACT 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

(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

2

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 log-transformation (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

value, independent of the HRV variable being compared This is useful considering that there are many different HRV variables that vary substantially in their mean values

4 Results

4.1 Mean measurements obtained by Nevrokard and F-EXTRACT

In Chapter 10, the mean values of HRV parameters obtained by EXTRACT 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 two times higher when measured by the Nevrokard system

F-In Figure 11-5, it can be observed that the LF power obtained by the two HRV systems did not differ as much as the HF power, which was almost 5 times higher when measured by the Nevrokard system Similar to most time-domain variables, the total power measured by the Nevrokard system was approximately twice of that measured by F-EXTRACT When the effect of the total power was removed, the normalized LF and HF power measured by the Nevrokard system were lower (approximately half) and higher (almost twice), respectively, than those

Figure 11-4: Bar charts of mean time-domain parameters at various gestational ages

as measured by Nevokard and F-EXTRACT

Figure 11-5: Bar charts of mean frequency-domain parameters at various

gestational ages as measured by Nevokard and F-EXTRACT

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 EXTRACT 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 F-EXTRACT 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)

F-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

Table 11-2: Bland-Altman analysis (mean difference) of time- and frequency-

domain variables

HRV parameter Mean

difference (N-F)

SD of difference

Lower limit

of agreement

Upper limit

of agreement

F- measurements obtained from F-EXTRACT

Table 11-3: Bland-Altman analysis (mean percent difference) of time- and

frequency-domain variables

HRV parameter Mean %

difference (N-F)

SD of difference

Lower limit

of agreement

Upper limit

of agreement

Mean difference, SD of bias and limits of agreements are in %

N- measurements obtained from Nevrokard

F- measurements obtained from F-EXTRACT

measurements obtained from Nevokard and F-EXTRACT, respectively

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 F-EXTRACT 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

Average normalized LF power

Average normalized HF power

Average total power

measurements obtained from Nevokard and F-EXTRACT, respectively

500 1000 1500 2000 -300

Average normalized LF power

Average normalized HF power

Average total power

N and F= measurements obtained from Nevokard and F-EXTRACT, respectively

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)

5 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 4 different commercially available HRV systems on the same 24-hour Holter tape recording and obtained an inter-system difference ranging from 7 ms to 100 ms in mNN Sandercock GR et al (Sandercock GR et al., 2004) used 3 different HRV systems to analyze simultaneously-recorded short-term ECG on 30 healthy adults and found that the bias in mNN measurements between