1. Introduction

The exercise ECG lead systems commonly applied are bipolar, the Mason-Likar 12-lead and the three-dimensional vectorcardiographic. The most widely used is the Mason-Likar modification of the standard 12-lead system [Mason and Likar, 1966], where the conventional wrist and ankle electrodes are placed at the base of the limbs. The 12-lead system comprises six limb (I, II, III, aVR, aVL, and aVF) and six chest leads (V1 to V6). The chest leads are unipolar, the reference for them being the so-called Wilson central terminal (average of the potentials at the right and left arms and left leg). Three of the limb leads are bipolar, measuring the potential difference between two points, and another three are augmented unipolar leads, when the reference for the measurement electrode is the average of two other limb leads. Arising from these differences in the measurement sensitivity of the individual leads, the diagnostic properties the leads are also dissimilar. The objective of this study to compare the diagnostic properties of the standard exercise ECG leads in the detection of CAD using the ST/HR hysteresis, ST/HR index and ST-segment value at the end of exercise.

2. Material and Methods

Study Population
The study material comprised 317 patients who had undergone the computerized exercise ECG test in Tampere University Hospital (Tampere, Finland) [Viik et al., 1997]. All patients had been referred for routine clinical exercise ECG testing. There were 128 patients with significant CAD proved by coronary angiography and 189 patients with a low likelihood of CAD. None of the patients had left/right bundle branch block pattern in resting ECG or recent myocardial infarction (< 8 weeks) and everyone had ECG recording of at least 3 minutes during the recovery phase. Selective coronary angiography was performed using the Judkins technique. CAD was considered significant when ³50% luminal narrowing in the major coronary arteries was present. Forty-nine patients had significant stenosis in all three major coronary arteries or in the left main coronary artery, and 45 and 34 patients with significant stenosis in a single vessel and two vessels, respectively. The reference group was selected on the basis of previous clinical history. All the reference patients had no history of any cardiac disease, had a normal resting ECG, and had no anginal type chest pain or cardiac medication. The major characteristics of the study population are presented in Table 1.

	CAD group	Reference group
	(n = 128)	(n = 189)
Age (years)	55±8	47±12
Sex (male/female)	101/27	100/89
β-blockers	105	0
Calcium antagonist	47	0
Nitrates	87	0
Maximum heart rate (bpm)	126±21	164±19
Anginal type chest pain	61	0

Figure 1. Standard deviations (SD), standard errors (SE) and means (x) for the ST-segment depression at the end of exercise, ST/HR index and ST/HR hystresis lead by lead. Filled symbols illustrate CAD group and non-filled symbols represent patients with a low likelihood of the disease.
CAD = Coronary artery disease

Exercise Electrocardiography
All exercise tests were performed on a bicycle ergometer using a computerized recording system. The ECG recordings were made with a SYSTEM II EXES recorder (Siemens-Elema, Solna, Sweden). The lead system used was the Mason-Likar modification of the standard 12-lead system [Mason and Likar, 1966]. The graded protocol followed a standard clinical routine with an initial workload of 40W for women and 50W for men and an increment of 40W and 50W every 4 minutes for women and men, respectively. The exercise tests were sign- and symptom-limited maximal tests using the recommended criteria for termination [Arstila et al., 1984]; fatigue or chest pain were the reasons for termination in most cases.

ST-segment amplitude, heart rate and workload data were automatically determined by commercial analyzer at intervals of 60 seconds throughout the exercise test. The ST-segment amplitudes were measured with an accuracy of 10 mV. Computer-determined ST-segment amplitudes were defined at 60 ms after the J-junction [Lehtinen, et al., 1997; Okin et al., 1991], considering the end of PR-segment as the isoelectric line, for each of the 12 leads from the beginning of the exercise test up to the first three consecutive minutes of post-exercise recovery. ST-segment amplitude, heart rate and workload data were stored digitally for further processing and analysis.

ST/HR Hysteresis and ST/HR Index Calculation
The pairs of ST depression and heart rate were measured before commencement of exercise, at the end of each minute of exercise, at the end of exercise and at the end of the first three consecutive minutes of the recovery phase. ST-segment changes during the exercise phase and up to three minutes of recovery were plotted as a function of heart rate, termed here the ST/HR diagram. ST/HR hysteresis (mV) was calculated, as described by Lehtinen et al. [Lehtinen, et al., 1996a], by integrating the difference in ST depression between the exercise and recovery phases over the heart rate from the minimum heart rate during recovery to the maximum heart rate in the exercise test. The integral was divided by the heart rate difference over the integration interval in order to normalize the ST/HR hysteresis with respect to the recovery heart rate decrement. This variable represents the average difference in ST depressions between the exercise and recovery phases at an identical heart rate up to three minutes of recovery.

Calculation of the ST/HR index was made as suggested by Detrano and associates [Detrano et al., 1986]: The overall ST-segment deviation at end of exercise was divided by the exercise-induced change in heart rate (mV/bpm). Thus, both the ST depression and the ST elevation are included in the beginning and in the end of the exercise phase.

The calculation of the ST/HR hysteresis and ST/HR index was made with a computer program developed for comprehensive ST/HR analysis [Lehtinen, et al., 1996b].

Data Analysis and Statistical Methods
The principal statistical method for comparison of the discriminative capacities of exercise ECG variables was receiver operating characteristic (ROC) analysis. Quantitative variables were analyzed using two-tailed Student's t-test.
ROC analysis was used because it allows comparison of continuous diagnostic variables without any partition value (i.e. operating point). In ROC analysis the sensitivity and specificity values are plotted in the ROC space over the range of test measurement partition values. The area under the ROC curve represents overall diagnostic performance, i.e. the probability that a random pair of patients with and without CAD will be correctly diagnosed [Hanley and McNeil, 1982]. Statistical differences between the areas under two ROC curves were compared using nonparametric analysis of correlated ROC curves [DeLong et al., 1988] with a routine written by Vida (version 2.5) [Vida, 1993].

3. Results

The mean values of ST/HR hysteresis, ST/HR index and STend between the patient and reference groups were significantly different in almost every lead (Fig. 1). Only lead aVL did not evince significant differences at a level of p < 0.001 in any of the variables used and lead V1 attained a significant difference only in the case of ST/HR hysteresis. Despite the good discriminative capacity of the individual leads, there were important differences between the leads. The areas under the ROC curves for ST/HR hysteresis, ST/HR index and STend in each individual standard lead as are presented in Fig. 2. In each variable the highest areas under the ROC curves were in chest leads V4, V5 and V6, and in limb leads I and -aVR. The most deficient areas under the ROC curves were distinctly those in chest lead V1 and in limb lead aVL in all variables (p<0.0001 vs. V5 and I in each variable).

The sensitivities as well as partition values at fixed specificity varied between the individual leads. This behavior was discernible in all variables. Fig. 3 presents the ROC curves of leads V5, I, V1 and aVL with the partition values yielding nearest to 90% specificity.

Figure 2. The areas under the ROC curves in standard leads for the ST/HR hysteresis, ST/HR index, and ST_end shown on scales (0% to 100%) in direction of lead. The horizontal view presents the results for chest leads and the frontal view results for limb leads. Values are percentages of total ROC space. HR = heart rate; ROC = receiver operating characteristic; ST_end = ST-segment value at the end of exercise.

4. Discussion

The number of ECG leads has been a difficult topic over the decades in the matter of detecting CAD by exercise ECG. It has been stated that most ischemic responses can be seen in lead V5. As far back as the 1970s, however, several researchers [Baron et al., 1980; Chaitman et al., 1978; Robertson et al., 1976; Tucker, et al., 1976] demonstrated that the sensitivity of the exercise test could be improved by using multiple leads. Subsequently other researchers [Carlens et al., 1985; Fox et al., 1984; Miller et al., 1987; Moussa et al., 1992; Simoons and Block, 1981] suggested that the use of 12 leads does not significantly improve the sensitivity or diagnostic accuracy of the exercise ECG in the detection of CAD over lead V5. However, the diagnostic criterion of ST depression is generally applied to the ECG lead with the deepest ST depression. Using this kind of approach the sensitivity of the ECG test can be enhanced [Viik et al., 1998; Viik et al. 1999]. However, the number of false-positive responses increases concomitantly and the specificity of the test is thus reduced. In view of this problem, the exercise standards [Fletcher et al., 1995], guidelines [Gibbons et al., 1997] and a textbook [Froelicher and Myers, 2000] recommend use of V5 with some bipolar or inferior lead patterns for patients with normal resting ECG. However, the diagnostic properties of each individual 12 leads are not studied or compared simultaneously.

Overall Diagnostic Performances of Individual Leads
For each method the highest diagnostic performances according to ROC analysis were achieved in chest leads V4, V5, V6 and limb leads I, -aVR. The excellent diagnostic performances of chest leads V5 and V6 are apparently maintained irrespective of the analysis method employed. The results support those obtained in previous studies, where the lateral precordial leads have been found to detect most ST depressions [Fletcher et al., 1995; Froelicher et al., 1976; Miller et al., 1987; Miranda et al., 1992; Simoons and Block, 1981; Tavel and Shaar, 1999]. Contrary to the current conception, leads I and -aVR achieved an overall diagnostic performance comparable to that with lateral precordial leads in each variable. By reason of the low overall measurement sensitivity, these leads are commonly underestimated in conventional CAD diagnosis. One further indubitable reason for the underestimation of lead aVR is that it is not usually used inverted.

Figure 3. The ROC curves for chest leads V5 and V1 and limb leads I and aVL in each study variable. Partition values presented in curves indicate variable values yielding a specificity of 90% (in millivolts for ST/HR hysteresis, ST-segment depressions and in microvolts per minute for ST/HR index). Numbers after marking of lead express area under the ROC curves as percentages. Differences between the areas under the curves in leads V5 or I and V1 or aVL were highly significant in each method. No statistically significant differences were observed between leads V5 and I with any of these variables.
HR = heart rate; ROC = receiver operating characteristic; STend = ST-segment value at the end of exercise.

Leads aVL and V1 were the most unreliable in the discrimination of patients with CAD and patients with a low likelihood of the disease. The areas under the ROC curves were the smallest and highly significantly smaller than those obtained with lead V5 in each variable. Consistent with this observation, leads V1 and aVL have been excluded in many studies employing the standard 12 leads during exercise tests [Kligfield et al., 1989; Morise and Duval, 1995; Sievänen et al., 1991]. This poor overall diagnostic performance is understandable considering the measurement orientation (lead direction) of those leads. The directions in leads aVL and V1 are perpendicular to that of the main injury current arising from subendocardial ischemia in the left ventricle. Thus, the majority of ischemic responses shown in these leads are modest and might be observed as either ST depression or ST elevation, which tends to complicate analysis of the ECG. Moreover, the low diagnostic performance of leads aVL and V1 might derive from the sensitivity of the leads to interindividual differences in position and rotation of the heart [Hoekema et al., 1999; Huiskamp and van Oosterom, 1992; Hyttinen, 1994]. On the other hand, these leads might be useful when detecting severe myocardial ischemia with ST elevation [Dunn et al., 1981a; Dunn et al., 1981b; Longhurst and Kraus, 1979].

Partition Values for the ST and ST/HR Variables
Guidelines and standards for the detection of ischemia by conventional ST-segment depression analysis recommend use of the same fixed partition value (0.10 mV) for every lead [Fletcher et al., 1995; Gibbons et al., 1997; Heikkilä, 1991; Jain and Murray, 1995]. Furthermore, many studies [Fox et al., 1984; London et al., 1988; Miller et al., 1987; Tucker et al., 1976] have shown lead V5 to be capable of detecting the majority of ischemic responses when a positive test criterion of ³0.10 mV ST-segment depression is used. The results here indicate that larger partition values are most suitable for the lateral precordial leads (V4-V6). Since the different sensitivity distributions of the individual ECG leads mean that a fixed global partition value does not treat individual leads equally, it would be quite natural use dissimilar partition values for different ECG leads, especially for computerized analysis. The results support those in previous studies [Froelicher et al., 1976; Hyttinen et al., 1997; Miranda et al., 1992; Viik et al., 1995; Viik et al., 1998] in which it has been suggested that more detailed, lead-specific criteria should be defined and applied for the ST depression.

Acknowledgements

This work has been financially supported by the Academy of Finland, the Ella and Georg Ehrnrooths Foundation, the Ida Montin Foundation, the Finnish Cultural Foundation (Pirkanmaa Fund), the Ragnar Granit Foundation, and the Wihuri Foundation.

References