- Email: [email protected]
Comparative measures of systolic ejection during treadmill exercise by impedance cardiography and Doppler echocardiography
Biological Psychology 36 (1993) 51-61 0 1993 Elsevier Science Publishers B.V. All rights reserved
51 0301.0511/93/$06.00
Comparative measures of systolic ejection during treadmill exercise by impedance cardiography and Doppler echocardiography Paul N. Kizakevich *,a, Steve M. Teague b, Daniel B. Nissman a, Warren J. Jochem a, Romain Niclou b and Mukesh K. Sharma b a Research Triangle Institute, Biomedical Engineering Program Office, Research Triangle
Park,
NC 27709, USA b University of Oklahoma, Cardiovascular Section, Oklahoma City, OK 73190, USA
Measurements of systolic ejection dynamics by impedance cardiography were compared with simultaneous Doppler echocardiography in normal subjects and coronary artery disease patients. Patients with chest pain admitted for elective coronary angiography were monitored by simultaneous impedance cardiography and Doppler echocardiography before, during, and after treadmill exercise. Ensemble-averaged ECG, impedance cardiogram (ICG), the first derivative of ICG (dZ/dt), and Doppler waveforms were analyzed to identify systolic ejection variables. The timing of aortic valve opening was well correlated (r = 0.78) the timing of peak ejection velocity was very well correlated (r = 0.86), and the timing of aortic valve closure was moderately correlated (r = 0.69 and r = 0.73) in these subjects. The thoracic electrical impedance acceleration and normalized impedance acceleration indices were moderately correlated with Doppler model acceleration (r = 0.74, r = 0.79). The impedance cardiogram waveforms are of complex origin and are related to both aortic blood velocity and aortic blood acceleration. Users of dZ/dt timing features for determining aortic valvular events might consider alternative impedance features to improve ejection time accuracy. Keywords: Impedance cardiography; Doppler echocardiography; Systolic time intervals; Aortic blood velocity; Aortic blood acceleration; Preejection period; Electromechanical systole
Introduction
The origin and interpretation of the cardiac electrical impedance waveform have been extensively studied, and yet remain somewhat of a mystery. It is generally held that the cardiac bioimpedance signals arise from a complex
* Corresponding
author.
52
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
mixture of volume conduction effects associated with both pulsatile volume changes in the great arteries and changes in red blood cell alignment during pulsatile blood flow. Even though the relative contribution of these and other aspects is unknown, models relating cardiac bioimpedance signals to stroke volume are often applied in both healthy and unhealthy subject populations. Thoracic electrical impedance (TED measurements are made by applying a high frequency carrier current to the body to establish an electrical field within the thorax. The carrier signal, modulated by interelectrode volume conduction and spacing changes, is picked up by additional electrodes and demodulated to obtain a multicomponent TEI signal. The principal TEI components are the quasistatic low frequency component (Z,[t]) and the higher frequency respiratory and cardiac component (AZ[t]). Using analog, digital, or mixed signal processing, AZ[t] is usually partitioned into three physiological signals: the impedance pneumogram (IPG), the impedance cardiogram (ICG), and the first derivative of the impedance cardiogram (dZ/dr). Using the empirical model developed by Kubicek et al. (1966), stroke volume and cardiac output is calculated from thoracic electrical impedance measurements as follows SV=p(L/Z,)*.dZ/dt.LVET CO = RATE
. (SV/lOOO)
cc L min-’
where p is the blood resistivity (135 a-cm), L is the TEI sense-electrode separation, Z, is the mean TEI, dZ/dt is the maximum value of the rapid systolic excursion of the dZ/dt waveform, LVET is the left ventricular ejection time, and RATE is the mean heart rate. The LVET is often determined by measuring the timing of the opening and closing of the aortic valve according to fiducial features of the dZ/dt waveform (Sheps et al., 1982; Sherwood, Allen, Fahrenberg, Kelsey, Lovallo & Van Dooman, 1990). The application of computer technology to cardiac bioimpedance measurements has resulted in improved signal quality during exercise (Kizakevich, Gollan, McDermott, & Aranda, 1977), but automated analysis and knowledgeable interpretation of waveform measurements are still limited by the origin question. Since cardiac bioimpedance signals are generally related to ejection dynamics (Teague, Kizakevich, Jochem, Petrovick, & Waugh, 1986), we decided to compare measures of systolic ejection dynamics during treadmill exercise using simultaneous impedance cardiography and Doppler echocardiography. Moreover, since this comparison was part of a larger investigation of non-invasive cardiac function measurements in coronary artery disease patients, we decided to carry out the comparisons in both the normal and diseased subjects. By these means, we hope to further unravel the cardiac bioimpedance mystery in a clinically relevant population.
53
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
Methods
We prospectively studied 31 hospitalized male patients with chest pain syndromes admitted for elective coronary angiography. Six had electrocardiographic evidence of prior myocardial infarction, but all were free of valvular heart disease by physical examination. Other conditions for enrolment in the study included informed consent, ambulatory status, and a reasonable expectation of exercising to a moderate workload. Patients underwent routine cardiac catheterization for the determination of left ventricular function and coronary anatomy. Ejection fraction and regional wall motion abnormalities were obtained from the right anterior oblique projection, while end-diastolic pressure was recorded using fluid filled catheters. Preparations for the exercise tolerance tests were as follows: electrodes were applied for 12 lead ECG monitoring in the usual fashion. For thoracic electrical impedance measurements, Kubicek tetrapolar band electrodes were utilized, with pairs spaced 5 cm on center around the neck and just below the xyphoid process. Patients performed a modified symptom-limited Bruce protocol on treadmill with increments in elevation and/or belt speed at 3 min intervals. Endpoints of exercise included the patient’s request, moderate angina, claudication, fatigue or dyspnea, ST segment depression of 4 mm or more, exercise induced hypotension or significant dysrhythmia. Electrocardiograms (V5) and impedance cardiograms (Z,, AZ and dZ/dt) were recorded continuously beginning 4 min before exercise and continuing through 6 min post exercise. Suprasternal Doppler assessments of ascending aortic flow (Doppler or ExerDop ‘) were taken at rest, at peak exercise, and at 2 min into each exercise stage (Teague, 1990). Additional Doppler recordings were taken at 30 s and 2 min into recovery. Each signal was sampled at 400 Hz via a 1Zbit analog-to-digital converter, displayed for quality assurance, and stored on optical disk using a custom waveform processing program (WVSHELL ‘). In addition to analog data passed to the computerized data logging system, the ExerDop system provided on-line analysis of the Doppler echocardiogram. ExerDop measurements included maximal acceleration (MA) and peak modal velocity (VP) of ascending aortic flow. The acquired signals were processed and analyzed off-line according to methods previously described (Kizakevich, Teague, Jochem, Nissman, Niclou & Sharma, 1989). A variety of signal processing procedures were used to prepare the acquired signals for waveform analysis: the ECG was pr,ocessed using a 40 Hz low-pass filter to minimize power-line noise; the ’ ExerDop is a trademark of Quinton Company, Seattle, WA. 2 WVSHELL is a trademark of Research Triangle Institute, Research
Triangle
Park, NC.
54
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
processed ECG was passed to a 4 Hz high-pass filter to generate an ECG TRIGGER signal free of baseline wander and with suppressed P and T waves; the dZ/dt was processed using an 0.5 Hz high-pass filter to eliminate baseline error; the mean Doppler baseline was determined in diastole and the overall Doppler baseline was adjusted; the AZ was tested and corrected for analog rebalance artifacts; and the Z, signal was unmodified. Segments of the processed signals comprising three to five cardiac cycles were displayed to identify the onset of the Doppler echocardiogram series taken during each exercise stage. The first cardiac cycle with a “good” Doppler waveform was then selected as a template for qualifying subsequent cycles during ensemble averaging. During ensemble averaging, successive cardiac cycles were extracted, displayed, and automatically accepted or rejected according to the template parameters. Ensemble averaging proceeded automatically until 32 qualified cycles were averaged, unless the investigator aborted the process before 32 cycles were attained. The primary reason for aborting the ensemble average was the loss of a useful Doppler signal. In such cases, at least 20 cycles had been averaged before terminating the averaging process. Following ensemble averaging, the four averaged waveforms (ECG, dZ/d t, Doppler, and AZ) and the last unaveraged Z, waveform were automatically analyzed for a comprehensive set of waveform features (Figs. 1 and 2, cursors 0, 30, 31,...) and presented for interactive feature validation. Corrective
Fig. 1. Ensemble-averaged ECG, dZ/dt, waveform features (standing at rest).
Doppler,
AZ and Z,
waveforms
showing
identified
55
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
Fig. 2. Ensemble-averaged ECG, dZ/dt, waveform features (treadmill exercise).
Table 1 Doppler echocardiogram
Doppler variable TDavo TDpeak TDavc VP MA
and impedance
Doppler,
cardiogram
AZ, and
Cursor
Description
ms ms ms
20 22 23
Aortic Peak Aortic Peak Peak
dZ/dt variable TZavo TZpeak TZav TZx vTZpeak NvTZpeak AC NAC sTZacce1
a/s l/s fi /s/s l/s/s fi /s/s
AZ variable TZonset TZmax
ms ms
waveforms
measurements
Units
m/s m/s/s
Z,
valve aortic valve aortic aortic
opening velocity closure velocity acceleration
Aortic opening wave (B) Peak dZ/d t waveform Aortic closure wave Aortic closure nadir (X) Peak dZ/dT amplitude Peak dZ/dT amplitude/Z, (vTZpeak)/(TZpeak-TZavo) (vTZpeak)/(Z,*(TZpeak-TZavo)) Peak dZ/dt slope
ms ms ms ms
16 18
Rapid systolic rise Peak AZ waveform
Z, variable Zo
R
Mean thoracic
impedance
with
identified
56
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-6:
action was made on errant feature cursors using full screen display of each waveform, and the resultant measured and calculated variables stored. A composite of the analyzed waveforms was then displayed and a printed copy made for the experiment archive. Selected waveform timing and amplitude measurements were retained for statistical analysis (Table 1). This process of data replay, ensemble averaging, and waveform analysis was repeated for each exercise stage.
Results
A population of five normal and 26 CAD patients was recruited for the study with ages ranging from 36 to 74 years (mean = 59 years). CAD was defined as having at least 70% occlusion of one or more coronary arteries. Fourteen subjects were dropped because of poor signal quality: two with noisy ECGs and 12 with noisy or poor exercise Dopplers. All had adequate impedance signals. The distribution of subjects included in analysis comprised four normals, four subjects with one-vessel disease, one subject with two-vessel disease, six subjects with three-vessel disease, and two subjects with left main CAD.
Table 2 Comparison of systolic gram measurements
event timing
cardiogram
and Doppler
echocardio-
Mean difference (Y-X), ms
Percent difference 100* (Y-X)/Y
TDavo TDavo TDavo TDavo
0.78 0.79 0.27 0.79
20 -31 59 6
16 25 49 5
TDpeak TDpeak TDpeak
0.75 0.86 0.87
76 25 62
43 14 35
Peak Doppler acceleration TZAvo TDslope TDslope TZpeak
0.88 0.90
43 -8
30 6
Aortic valve closure TZavc TZX TZmax
0.69 0.73 0.50
14 -21 88
4 6 27
variable
(XI
(Y)
Aortic valve opening TZavo TZpeak TZmin TZonset Peak Doppler velocity TZavo TZpeak TZonset
TDavc TDavc TDavc
variable
impedance
r
Impedance
Doppler
between
57
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61 Table 3 Comparison of systolic ejection gram measurements Impedance
indices between
Doppler
Variable
impedance
cardiogram
and Doppler
echocardio-
Variable
(XI
(Y)
Peak Doppler velocity vTZpeak NTZpeak AC
VP VP VP
0.31 0.29 0.55
Modal Doppler acceleration sTZaccel AC NAC
MA MA MA
0.69 0.74 0.79
Comparisons of impedance cardiogram and Doppler echocardiogram measurements are summarized in Tables 2 and 3. All systolic event timing measurements (Table 2) were made relative to the ECG Q-wave (Figs. 1 and ETT
llwR =.7i:44 s = ,553
(Impedance vs Doppler) RECRElSlON MiRLYSIS
IIC (ohms/s/s)
87 SMPLES 17 SUBJECTS Observeddata
R::
I =
4.84
Fig. 3. Correlation (AC) measurements
RTI M/28/91
_
of aortic blood acceleration during treadmill exercise’.
by Doppler
(MA)
and impedance
cardiogram
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
58
2, cursor 0). For each comparison the impedance variables (X), Doppler variables (Y), and their regression coefficients (r) are given. For systolic event timing the mean paired-difference (Y-X), and percent mean paireddifference (lOO*(Y - X)/Y) are also listed. Each comparison includes 87 observations. The relationship between impedance and Doppler timing measurements was good. The timing of aortic valve opening by TZavo versus TDavo was well correlated (r = 0.78) with TZavo preceding TDavo by an average of 20 ms (P < 0.001). When aortic valve opening was estimated using the onset of the AZ waveform (TZonset), the correlation was slightly better (r = 0.79) and the timing-difference was much smaller (6 ms, P < 0.01). The time-topeak dZ/dt and Doppler velocity waveforms were very well correlated (r = 0.86) with TZpeak preceding TDpeak by an average of 25 ms (P < 0.001). The timing of aortic valve closure using either TZavc or TZx was moderately correlated to TDavc (r = 0.69; r = 0.73). On average, TZavc preceded TDavc by 14 ms (P < 0.001) while TZx followed TDavc by 21 ms (P < 0.001).
NAC (l/s/s)
plj - 7if22 S -16:948 1 4,83
_
Fig. 4. Correlation of aortic blood acceleration by Doppler cardiogram (NAC) measurements during treadmill exercise.
87 SRHPLES
17 SUBJECTS Observed data RTI 1@/28/91 (MA) and normalized
impedance
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
59
The relationship between impedance and Doppler amplitude measurements yielded varied results. The aortic blood velocity variables (vTZpeak and VP) were minimally correlated (r = 0.31). Normalizing vTZpeak with 2, (NvTZpeak) decreased the correlation (r = 0.29). The aortic blood acceleration variables (sTZacce1 and MA) were better correlated (r = 0.69) than their velocity counterparts. When the overall impedance acceleration (AC) was compared with MA, the correlation was improved (r = 0.74) (Fig. 3) and normalizing AC with Z, (NAC) further improved the correlation (1. = 0.79) (Fig. 4).
Discussion For over 20 years (Lababidi, Ehmke, Durnin, Leaverton & Lauer, 1970), many investigators have studied the relationship of the thoracic electrical impedance signals to the timing of cardiac events (Sheps, Petrovick, Kizakevich, Wolf & Craige, 1982; Sherwood, Allen, Fahrenberg, Kelsey, Love110 & Van Dooman, 1990). Only a few have reported on the comparison of impedance cardiograms with M-mode echocardiograms (Schieken, Patel, Falsetti, Barnes & Lauer, 1978; Petrovick, Kizakevich, Stacy & Haak, 1980) or with Doppler echocardiograms (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986). This study represents the first trial of impedance cardiography as compared with Doppler echocardiography in resting and exercising subjects with coronary artery disease. A surprising outcome was that TDavo was more closely associated with TZonset (AZ) than any dZ/dt feature. On average, TZavo (Labadidi’s B point) preceded TDavo by 20 ms, a finding inconsistent with other studies (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986). Furthermore, TDpeak followed TZpeak by an average of 25 ms, indicating a possible phase delay in the Doppler signal. Since TDavc occurred midway between the TZavc and TZx waveform features, adjusting these results for a Doppler phase delay would move TDavc nearer TZavc. Considering the results of our previous work in exercising normals and resting cardiac patients (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986), the poor correlation of dZ/dt amplitude (vTZpeak) with Doppler velocity (VP) found in this study was disappointing. It is unclear whether the lack of correlation arose from subject variability, basic methodology, instrumentation, or signal processing. Many subjects were dropped because of unstable baselines in the analog Doppler velocity signal. The instability appeared related to ultrasound transducer interference caused by the impedance cardiograph current source. Perhaps those subjects with “acceptable” Doppler baselines were actually unacceptable, thus contributing to errors in the determination of VP. In contrast, the correlation of acceleration
60
P. N. Kizakeuich et al. / Biological Psychology, 36 (1993) 51-61
measurements was fairly good. It is likely that the problems that occurred with the Doppler velocity baseline may have had less effect on maximal acceleration, a differential measurement. A variety of other systematic and random factors may have influenced the results. Doppler echocardiograms represent ejection dynamics as sampled within the aortic arch. Impedance cardiograms, however, are less specific and represent the integral of ejection dynamics throughout the thorax. This results in a distribution of sampled volume and summation of impedance changes across time, as the pulse propagates through the arterial tree. The Doppler echocardiogram was processed internally by the ExerDop system for automatic measurement of MA and VP, and to produce a continuous analog signal representing aortic blood velocity. MA and VP measurements were determined on sequential cardiac cycles and their average listed on the ExerDop printer. The analog signal was sampled simultaneously with the impedance signals for ensemble averaging and systolic timing analysis. Accuracy of the amplitude and phase of the analog Doppler velocity signal was unspecified and could not be easily determined. The ensemble averaging and the feature analysis procedures may also have contributed to measurement uncertainty (De Souza & Panerai, 1981). Although individual cycles were qualified for averaging, the averaging process accepted cycles with as much as 20% variability in heart rate and 25% variability in Doppler velocity amplitude. Thus, while random disturbances in signal amplitude were reduced by averaging, systematic variability in systolic timing as occurs with respiration may have affected the clarity of waveform features. Finally, since feature extraction included both automated and human-verified waveform analyses, systematic and operator-dependent errors may have contributed to measurement uncertainty. For example, alternative measures of left ventricular ejection time using the same set of Doppler velocity signals may differ by as much as 5% (Koito & Spodick, 19891. Equivalent differences between impedance and Doppler measurements as found in our study might therefore be considered reasonable. The impedance cardiogram waveforms (AZ and dZ/dt) are of complex origin and are correlated with both the aortic blood velocity and aortic blood acceleration. Users of dZ/dt timing features for determining aortic valvular events might consider alternative impedance features to improve ejection time accuracy.
References De Souza, W.M., & Panerai, R.B., (1981) Variability of thoracic impedance cardiograms in man. Medical and Biological Engineering and Computing, 19, 411-415. Kizakevich, P.N., Gollan, F., McDermott, .I., & Aranda, J. (1977). Continuous non-invasive monitoring of myocardial function: Its application in exercise stress testing. Proceedings of the
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
61
Symposium on Computer Applications in Medical Care (pp. 3251, Washington, DC: IEEE Computer Society Press, Catalog Number 77CH1270-8 C. Kizakevich, P.N., Teague, S.M., Jochem, W.J., Nissman, D.B., Niclou, R., & Sharma, M.K. (1989). Detection of Ischemic Responses During Treadmill Exercise by Computer-Aided Impedance Cardiography. Proceedings of the Second Annual IEEE Symposium on ComputerBased Medical Systems (pp. lo-15), Washington, DC: IEEE Computer Society Press, Catalog Number 89CH2755-7. Koito, H., & Spodick, D.H. (1989) Optimal Doppler measurement of left ventricular ejection time. The American Journal of Cardiology, 63, 257-259. Kubicek, W.G., Karnegis, J.R., Patterson, R.P., Witsoe, D.A., & Mattson, R.H. (1966). Development and evaluation of an impedance cardiac output system. Aerospace Medicine, 37(12), 1208-1212. Lababidi, Z., Ehmke, D.A., Durnin, R.E., Leaverton, P.E., & Lauer, R.M. (1970) The first derivative thoracic impedance cardiogram. Circulation, 41, 651-658. Petrovick, M.L., Kizakevich, P.N., Stacy, R.W., & Haak, E.D. (1980) A comprehensive cardiac exercise stress processor for environmental health effect studies. Journal of Medical Systems, 4(2), 137-150. Schieken, R.M., Patel, M.R., Falsetti, H.L., Barnes, R.W., & Lauer, R.M. (1978) Effect of aortic valvular regurgiation upon the impedance cardiogram. British Heart Journal, 40, 958-963. Sheps, D.S., Petrovick, M.L., Kizakevich, P.N., Wolf, C., & Craige, E. (1982). Continuous noninvasive monitoring of left ventricular function during exercise by impedance cardiography. American Heart Journal, 103(4), 519-524. Sherwood, A., Allen, M.T., Fahrenberg, J., Kelsey, R.M., Lovallo, W.R., & van Dooman, L.J.P. (1990). Methodological guidelines for impedance cardiography. Psychophysiology, 27, l-23. Teague, S.M., Kizakevich, P.N., Jochem, W.J., Petrovick, M.L., & Waugh, R. (1986). An echocardiographic dissection of the impedance cardiogram. Auiation, Space and Enuironmental Medicine, 57(5), 506 (Abstract). Teague, S.M. (1990). Stress Doppler Echcardiography. Netherlands: Kluwer Academic Publishers.
51 0301.0511/93/$06.00
Comparative measures of systolic ejection during treadmill exercise by impedance cardiography and Doppler echocardiography Paul N. Kizakevich *,a, Steve M. Teague b, Daniel B. Nissman a, Warren J. Jochem a, Romain Niclou b and Mukesh K. Sharma b a Research Triangle Institute, Biomedical Engineering Program Office, Research Triangle
Park,
NC 27709, USA b University of Oklahoma, Cardiovascular Section, Oklahoma City, OK 73190, USA
Measurements of systolic ejection dynamics by impedance cardiography were compared with simultaneous Doppler echocardiography in normal subjects and coronary artery disease patients. Patients with chest pain admitted for elective coronary angiography were monitored by simultaneous impedance cardiography and Doppler echocardiography before, during, and after treadmill exercise. Ensemble-averaged ECG, impedance cardiogram (ICG), the first derivative of ICG (dZ/dt), and Doppler waveforms were analyzed to identify systolic ejection variables. The timing of aortic valve opening was well correlated (r = 0.78) the timing of peak ejection velocity was very well correlated (r = 0.86), and the timing of aortic valve closure was moderately correlated (r = 0.69 and r = 0.73) in these subjects. The thoracic electrical impedance acceleration and normalized impedance acceleration indices were moderately correlated with Doppler model acceleration (r = 0.74, r = 0.79). The impedance cardiogram waveforms are of complex origin and are related to both aortic blood velocity and aortic blood acceleration. Users of dZ/dt timing features for determining aortic valvular events might consider alternative impedance features to improve ejection time accuracy. Keywords: Impedance cardiography; Doppler echocardiography; Systolic time intervals; Aortic blood velocity; Aortic blood acceleration; Preejection period; Electromechanical systole
Introduction
The origin and interpretation of the cardiac electrical impedance waveform have been extensively studied, and yet remain somewhat of a mystery. It is generally held that the cardiac bioimpedance signals arise from a complex
* Corresponding
author.
52
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
mixture of volume conduction effects associated with both pulsatile volume changes in the great arteries and changes in red blood cell alignment during pulsatile blood flow. Even though the relative contribution of these and other aspects is unknown, models relating cardiac bioimpedance signals to stroke volume are often applied in both healthy and unhealthy subject populations. Thoracic electrical impedance (TED measurements are made by applying a high frequency carrier current to the body to establish an electrical field within the thorax. The carrier signal, modulated by interelectrode volume conduction and spacing changes, is picked up by additional electrodes and demodulated to obtain a multicomponent TEI signal. The principal TEI components are the quasistatic low frequency component (Z,[t]) and the higher frequency respiratory and cardiac component (AZ[t]). Using analog, digital, or mixed signal processing, AZ[t] is usually partitioned into three physiological signals: the impedance pneumogram (IPG), the impedance cardiogram (ICG), and the first derivative of the impedance cardiogram (dZ/dr). Using the empirical model developed by Kubicek et al. (1966), stroke volume and cardiac output is calculated from thoracic electrical impedance measurements as follows SV=p(L/Z,)*.dZ/dt.LVET CO = RATE
. (SV/lOOO)
cc L min-’
where p is the blood resistivity (135 a-cm), L is the TEI sense-electrode separation, Z, is the mean TEI, dZ/dt is the maximum value of the rapid systolic excursion of the dZ/dt waveform, LVET is the left ventricular ejection time, and RATE is the mean heart rate. The LVET is often determined by measuring the timing of the opening and closing of the aortic valve according to fiducial features of the dZ/dt waveform (Sheps et al., 1982; Sherwood, Allen, Fahrenberg, Kelsey, Lovallo & Van Dooman, 1990). The application of computer technology to cardiac bioimpedance measurements has resulted in improved signal quality during exercise (Kizakevich, Gollan, McDermott, & Aranda, 1977), but automated analysis and knowledgeable interpretation of waveform measurements are still limited by the origin question. Since cardiac bioimpedance signals are generally related to ejection dynamics (Teague, Kizakevich, Jochem, Petrovick, & Waugh, 1986), we decided to compare measures of systolic ejection dynamics during treadmill exercise using simultaneous impedance cardiography and Doppler echocardiography. Moreover, since this comparison was part of a larger investigation of non-invasive cardiac function measurements in coronary artery disease patients, we decided to carry out the comparisons in both the normal and diseased subjects. By these means, we hope to further unravel the cardiac bioimpedance mystery in a clinically relevant population.
53
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
Methods
We prospectively studied 31 hospitalized male patients with chest pain syndromes admitted for elective coronary angiography. Six had electrocardiographic evidence of prior myocardial infarction, but all were free of valvular heart disease by physical examination. Other conditions for enrolment in the study included informed consent, ambulatory status, and a reasonable expectation of exercising to a moderate workload. Patients underwent routine cardiac catheterization for the determination of left ventricular function and coronary anatomy. Ejection fraction and regional wall motion abnormalities were obtained from the right anterior oblique projection, while end-diastolic pressure was recorded using fluid filled catheters. Preparations for the exercise tolerance tests were as follows: electrodes were applied for 12 lead ECG monitoring in the usual fashion. For thoracic electrical impedance measurements, Kubicek tetrapolar band electrodes were utilized, with pairs spaced 5 cm on center around the neck and just below the xyphoid process. Patients performed a modified symptom-limited Bruce protocol on treadmill with increments in elevation and/or belt speed at 3 min intervals. Endpoints of exercise included the patient’s request, moderate angina, claudication, fatigue or dyspnea, ST segment depression of 4 mm or more, exercise induced hypotension or significant dysrhythmia. Electrocardiograms (V5) and impedance cardiograms (Z,, AZ and dZ/dt) were recorded continuously beginning 4 min before exercise and continuing through 6 min post exercise. Suprasternal Doppler assessments of ascending aortic flow (Doppler or ExerDop ‘) were taken at rest, at peak exercise, and at 2 min into each exercise stage (Teague, 1990). Additional Doppler recordings were taken at 30 s and 2 min into recovery. Each signal was sampled at 400 Hz via a 1Zbit analog-to-digital converter, displayed for quality assurance, and stored on optical disk using a custom waveform processing program (WVSHELL ‘). In addition to analog data passed to the computerized data logging system, the ExerDop system provided on-line analysis of the Doppler echocardiogram. ExerDop measurements included maximal acceleration (MA) and peak modal velocity (VP) of ascending aortic flow. The acquired signals were processed and analyzed off-line according to methods previously described (Kizakevich, Teague, Jochem, Nissman, Niclou & Sharma, 1989). A variety of signal processing procedures were used to prepare the acquired signals for waveform analysis: the ECG was pr,ocessed using a 40 Hz low-pass filter to minimize power-line noise; the ’ ExerDop is a trademark of Quinton Company, Seattle, WA. 2 WVSHELL is a trademark of Research Triangle Institute, Research
Triangle
Park, NC.
54
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
processed ECG was passed to a 4 Hz high-pass filter to generate an ECG TRIGGER signal free of baseline wander and with suppressed P and T waves; the dZ/dt was processed using an 0.5 Hz high-pass filter to eliminate baseline error; the mean Doppler baseline was determined in diastole and the overall Doppler baseline was adjusted; the AZ was tested and corrected for analog rebalance artifacts; and the Z, signal was unmodified. Segments of the processed signals comprising three to five cardiac cycles were displayed to identify the onset of the Doppler echocardiogram series taken during each exercise stage. The first cardiac cycle with a “good” Doppler waveform was then selected as a template for qualifying subsequent cycles during ensemble averaging. During ensemble averaging, successive cardiac cycles were extracted, displayed, and automatically accepted or rejected according to the template parameters. Ensemble averaging proceeded automatically until 32 qualified cycles were averaged, unless the investigator aborted the process before 32 cycles were attained. The primary reason for aborting the ensemble average was the loss of a useful Doppler signal. In such cases, at least 20 cycles had been averaged before terminating the averaging process. Following ensemble averaging, the four averaged waveforms (ECG, dZ/d t, Doppler, and AZ) and the last unaveraged Z, waveform were automatically analyzed for a comprehensive set of waveform features (Figs. 1 and 2, cursors 0, 30, 31,...) and presented for interactive feature validation. Corrective
Fig. 1. Ensemble-averaged ECG, dZ/dt, waveform features (standing at rest).
Doppler,
AZ and Z,
waveforms
showing
identified
55
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
Fig. 2. Ensemble-averaged ECG, dZ/dt, waveform features (treadmill exercise).
Table 1 Doppler echocardiogram
Doppler variable TDavo TDpeak TDavc VP MA
and impedance
Doppler,
cardiogram
AZ, and
Cursor
Description
ms ms ms
20 22 23
Aortic Peak Aortic Peak Peak
dZ/dt variable TZavo TZpeak TZav TZx vTZpeak NvTZpeak AC NAC sTZacce1
a/s l/s fi /s/s l/s/s fi /s/s
AZ variable TZonset TZmax
ms ms
waveforms
measurements
Units
m/s m/s/s
Z,
valve aortic valve aortic aortic
opening velocity closure velocity acceleration
Aortic opening wave (B) Peak dZ/d t waveform Aortic closure wave Aortic closure nadir (X) Peak dZ/dT amplitude Peak dZ/dT amplitude/Z, (vTZpeak)/(TZpeak-TZavo) (vTZpeak)/(Z,*(TZpeak-TZavo)) Peak dZ/dt slope
ms ms ms ms
16 18
Rapid systolic rise Peak AZ waveform
Z, variable Zo
R
Mean thoracic
impedance
with
identified
56
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-6:
action was made on errant feature cursors using full screen display of each waveform, and the resultant measured and calculated variables stored. A composite of the analyzed waveforms was then displayed and a printed copy made for the experiment archive. Selected waveform timing and amplitude measurements were retained for statistical analysis (Table 1). This process of data replay, ensemble averaging, and waveform analysis was repeated for each exercise stage.
Results
A population of five normal and 26 CAD patients was recruited for the study with ages ranging from 36 to 74 years (mean = 59 years). CAD was defined as having at least 70% occlusion of one or more coronary arteries. Fourteen subjects were dropped because of poor signal quality: two with noisy ECGs and 12 with noisy or poor exercise Dopplers. All had adequate impedance signals. The distribution of subjects included in analysis comprised four normals, four subjects with one-vessel disease, one subject with two-vessel disease, six subjects with three-vessel disease, and two subjects with left main CAD.
Table 2 Comparison of systolic gram measurements
event timing
cardiogram
and Doppler
echocardio-
Mean difference (Y-X), ms
Percent difference 100* (Y-X)/Y
TDavo TDavo TDavo TDavo
0.78 0.79 0.27 0.79
20 -31 59 6
16 25 49 5
TDpeak TDpeak TDpeak
0.75 0.86 0.87
76 25 62
43 14 35
Peak Doppler acceleration TZAvo TDslope TDslope TZpeak
0.88 0.90
43 -8
30 6
Aortic valve closure TZavc TZX TZmax
0.69 0.73 0.50
14 -21 88
4 6 27
variable
(XI
(Y)
Aortic valve opening TZavo TZpeak TZmin TZonset Peak Doppler velocity TZavo TZpeak TZonset
TDavc TDavc TDavc
variable
impedance
r
Impedance
Doppler
between
57
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61 Table 3 Comparison of systolic ejection gram measurements Impedance
indices between
Doppler
Variable
impedance
cardiogram
and Doppler
echocardio-
Variable
(XI
(Y)
Peak Doppler velocity vTZpeak NTZpeak AC
VP VP VP
0.31 0.29 0.55
Modal Doppler acceleration sTZaccel AC NAC
MA MA MA
0.69 0.74 0.79
Comparisons of impedance cardiogram and Doppler echocardiogram measurements are summarized in Tables 2 and 3. All systolic event timing measurements (Table 2) were made relative to the ECG Q-wave (Figs. 1 and ETT
llwR =.7i:44 s = ,553
(Impedance vs Doppler) RECRElSlON MiRLYSIS
IIC (ohms/s/s)
87 SMPLES 17 SUBJECTS Observeddata
R::
I =
4.84
Fig. 3. Correlation (AC) measurements
RTI M/28/91
_
of aortic blood acceleration during treadmill exercise’.
by Doppler
(MA)
and impedance
cardiogram
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
58
2, cursor 0). For each comparison the impedance variables (X), Doppler variables (Y), and their regression coefficients (r) are given. For systolic event timing the mean paired-difference (Y-X), and percent mean paireddifference (lOO*(Y - X)/Y) are also listed. Each comparison includes 87 observations. The relationship between impedance and Doppler timing measurements was good. The timing of aortic valve opening by TZavo versus TDavo was well correlated (r = 0.78) with TZavo preceding TDavo by an average of 20 ms (P < 0.001). When aortic valve opening was estimated using the onset of the AZ waveform (TZonset), the correlation was slightly better (r = 0.79) and the timing-difference was much smaller (6 ms, P < 0.01). The time-topeak dZ/dt and Doppler velocity waveforms were very well correlated (r = 0.86) with TZpeak preceding TDpeak by an average of 25 ms (P < 0.001). The timing of aortic valve closure using either TZavc or TZx was moderately correlated to TDavc (r = 0.69; r = 0.73). On average, TZavc preceded TDavc by 14 ms (P < 0.001) while TZx followed TDavc by 21 ms (P < 0.001).
NAC (l/s/s)
plj - 7if22 S -16:948 1 4,83
_
Fig. 4. Correlation of aortic blood acceleration by Doppler cardiogram (NAC) measurements during treadmill exercise.
87 SRHPLES
17 SUBJECTS Observed data RTI 1@/28/91 (MA) and normalized
impedance
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
59
The relationship between impedance and Doppler amplitude measurements yielded varied results. The aortic blood velocity variables (vTZpeak and VP) were minimally correlated (r = 0.31). Normalizing vTZpeak with 2, (NvTZpeak) decreased the correlation (r = 0.29). The aortic blood acceleration variables (sTZacce1 and MA) were better correlated (r = 0.69) than their velocity counterparts. When the overall impedance acceleration (AC) was compared with MA, the correlation was improved (r = 0.74) (Fig. 3) and normalizing AC with Z, (NAC) further improved the correlation (1. = 0.79) (Fig. 4).
Discussion For over 20 years (Lababidi, Ehmke, Durnin, Leaverton & Lauer, 1970), many investigators have studied the relationship of the thoracic electrical impedance signals to the timing of cardiac events (Sheps, Petrovick, Kizakevich, Wolf & Craige, 1982; Sherwood, Allen, Fahrenberg, Kelsey, Love110 & Van Dooman, 1990). Only a few have reported on the comparison of impedance cardiograms with M-mode echocardiograms (Schieken, Patel, Falsetti, Barnes & Lauer, 1978; Petrovick, Kizakevich, Stacy & Haak, 1980) or with Doppler echocardiograms (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986). This study represents the first trial of impedance cardiography as compared with Doppler echocardiography in resting and exercising subjects with coronary artery disease. A surprising outcome was that TDavo was more closely associated with TZonset (AZ) than any dZ/dt feature. On average, TZavo (Labadidi’s B point) preceded TDavo by 20 ms, a finding inconsistent with other studies (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986). Furthermore, TDpeak followed TZpeak by an average of 25 ms, indicating a possible phase delay in the Doppler signal. Since TDavc occurred midway between the TZavc and TZx waveform features, adjusting these results for a Doppler phase delay would move TDavc nearer TZavc. Considering the results of our previous work in exercising normals and resting cardiac patients (Teague, Kizakevich, Jochem, Petrovick & Waugh, 1986), the poor correlation of dZ/dt amplitude (vTZpeak) with Doppler velocity (VP) found in this study was disappointing. It is unclear whether the lack of correlation arose from subject variability, basic methodology, instrumentation, or signal processing. Many subjects were dropped because of unstable baselines in the analog Doppler velocity signal. The instability appeared related to ultrasound transducer interference caused by the impedance cardiograph current source. Perhaps those subjects with “acceptable” Doppler baselines were actually unacceptable, thus contributing to errors in the determination of VP. In contrast, the correlation of acceleration
60
P. N. Kizakeuich et al. / Biological Psychology, 36 (1993) 51-61
measurements was fairly good. It is likely that the problems that occurred with the Doppler velocity baseline may have had less effect on maximal acceleration, a differential measurement. A variety of other systematic and random factors may have influenced the results. Doppler echocardiograms represent ejection dynamics as sampled within the aortic arch. Impedance cardiograms, however, are less specific and represent the integral of ejection dynamics throughout the thorax. This results in a distribution of sampled volume and summation of impedance changes across time, as the pulse propagates through the arterial tree. The Doppler echocardiogram was processed internally by the ExerDop system for automatic measurement of MA and VP, and to produce a continuous analog signal representing aortic blood velocity. MA and VP measurements were determined on sequential cardiac cycles and their average listed on the ExerDop printer. The analog signal was sampled simultaneously with the impedance signals for ensemble averaging and systolic timing analysis. Accuracy of the amplitude and phase of the analog Doppler velocity signal was unspecified and could not be easily determined. The ensemble averaging and the feature analysis procedures may also have contributed to measurement uncertainty (De Souza & Panerai, 1981). Although individual cycles were qualified for averaging, the averaging process accepted cycles with as much as 20% variability in heart rate and 25% variability in Doppler velocity amplitude. Thus, while random disturbances in signal amplitude were reduced by averaging, systematic variability in systolic timing as occurs with respiration may have affected the clarity of waveform features. Finally, since feature extraction included both automated and human-verified waveform analyses, systematic and operator-dependent errors may have contributed to measurement uncertainty. For example, alternative measures of left ventricular ejection time using the same set of Doppler velocity signals may differ by as much as 5% (Koito & Spodick, 19891. Equivalent differences between impedance and Doppler measurements as found in our study might therefore be considered reasonable. The impedance cardiogram waveforms (AZ and dZ/dt) are of complex origin and are correlated with both the aortic blood velocity and aortic blood acceleration. Users of dZ/dt timing features for determining aortic valvular events might consider alternative impedance features to improve ejection time accuracy.
References De Souza, W.M., & Panerai, R.B., (1981) Variability of thoracic impedance cardiograms in man. Medical and Biological Engineering and Computing, 19, 411-415. Kizakevich, P.N., Gollan, F., McDermott, .I., & Aranda, J. (1977). Continuous non-invasive monitoring of myocardial function: Its application in exercise stress testing. Proceedings of the
P.N. Kizakevich et al. / Biological Psychology, 36 (1993) 51-61
61
Symposium on Computer Applications in Medical Care (pp. 3251, Washington, DC: IEEE Computer Society Press, Catalog Number 77CH1270-8 C. Kizakevich, P.N., Teague, S.M., Jochem, W.J., Nissman, D.B., Niclou, R., & Sharma, M.K. (1989). Detection of Ischemic Responses During Treadmill Exercise by Computer-Aided Impedance Cardiography. Proceedings of the Second Annual IEEE Symposium on ComputerBased Medical Systems (pp. lo-15), Washington, DC: IEEE Computer Society Press, Catalog Number 89CH2755-7. Koito, H., & Spodick, D.H. (1989) Optimal Doppler measurement of left ventricular ejection time. The American Journal of Cardiology, 63, 257-259. Kubicek, W.G., Karnegis, J.R., Patterson, R.P., Witsoe, D.A., & Mattson, R.H. (1966). Development and evaluation of an impedance cardiac output system. Aerospace Medicine, 37(12), 1208-1212. Lababidi, Z., Ehmke, D.A., Durnin, R.E., Leaverton, P.E., & Lauer, R.M. (1970) The first derivative thoracic impedance cardiogram. Circulation, 41, 651-658. Petrovick, M.L., Kizakevich, P.N., Stacy, R.W., & Haak, E.D. (1980) A comprehensive cardiac exercise stress processor for environmental health effect studies. Journal of Medical Systems, 4(2), 137-150. Schieken, R.M., Patel, M.R., Falsetti, H.L., Barnes, R.W., & Lauer, R.M. (1978) Effect of aortic valvular regurgiation upon the impedance cardiogram. British Heart Journal, 40, 958-963. Sheps, D.S., Petrovick, M.L., Kizakevich, P.N., Wolf, C., & Craige, E. (1982). Continuous noninvasive monitoring of left ventricular function during exercise by impedance cardiography. American Heart Journal, 103(4), 519-524. Sherwood, A., Allen, M.T., Fahrenberg, J., Kelsey, R.M., Lovallo, W.R., & van Dooman, L.J.P. (1990). Methodological guidelines for impedance cardiography. Psychophysiology, 27, l-23. Teague, S.M., Kizakevich, P.N., Jochem, W.J., Petrovick, M.L., & Waugh, R. (1986). An echocardiographic dissection of the impedance cardiogram. Auiation, Space and Enuironmental Medicine, 57(5), 506 (Abstract). Teague, S.M. (1990). Stress Doppler Echcardiography. Netherlands: Kluwer Academic Publishers.