Jeffrey Froning

The Electrocardiographic Exercise Test in a Population with Reduced Workup Bias: Diagnostic Performance, Computerized Interpretation, and Multivariable Prediction

The standard exercise test is still the first step in the evaluation of the stable patient with chest pain that may be due to coronary artery disease. This is because simple ST-segment measurements are as diagnostic as other tests that can be performed by the clinician [1, 2]. Although studies suggest that the discrimination of multivariable equations [3], heart rate adjustment [4], and scores [5] is superior to that of ST-segment measurements, failure to validate this superiority has impeded acceptance of these tools. Even in correlation studies that have appropriately enrolled consecutive patients who have had both exercise testing and coronary angiography, workup bias has been a limitation. Patients in these studies were selected for angiography if a physician judged that the likelihood of coronary disease was high enough to warrant this invasive procedure. This selection process makes patients with abnormal exercise test results more likely to be chosen and excludes patients with normal test results and high exercise capacity; this results in a higher prevalence of disease than would be seen in a clinical population. Prediction equations, scores, and heart rate adjustment algorithms have been derived from population with extensive workup bias and are unlikely to be applicable to patients who present with chest pain [6]. Our study reduced workup bias prospectively by following a protocol that required patients to agree to undergo both exercise testing and coronary angiography. A pilot study that did not avoid workup bias was done in 687 patients at two sites from October 1990 to August 1994. The main study, Quantitative Exercise Testing and Angiography (QUEXTA), enrolled 1274 patients at 12 sites from August 1994 to September 1995. Methods Patients To be included in QUEXTA, patients had to be men 18 years of age or older with probable or definite stable angina. Standard exclusion criteria were used, and patients with previous myocardial infarction or previous abnormal angiograms were excluded. To further minimize workup bias, the study allowed no more than 25% of the patients at any one site to have had a recent treadmill test. The preferred entry point was the clinic, but less than 25% of patients could come from either the exercise or angiography laboratories. Of 1274 consecutive male patients who were enrolled at 12 Veterans Affairs Medical Centers between 22 August 1994 and 15 September 1995, 814 had no myocardial infarction on electrocardiography or history, underwent both coronary angiography and treadmill testing, and had complete data. Institutional review was done centrally and at each study site, and all patients signed a consent form approved for this study. Coronary angiography and treadmill testing had to be done within 30 days of each other. For validation purposes, the 814 patients were divided into a training set of 543 patients (two thirds of the total sample) and a test set of 271 patients (one third of the total sample). Approximately 7000 patients had exercise testing, and 1328 patients were enrolled during the recruitment period. Clinical variables obtained at the initial evaluation were recorded on a standard form. Chest pain was coded as 1 for definite angina, 2 for probable angina, 3 for nonanginal pain, and 4 for no pain. All other clinical variables, except age, body mass index, resting ST-segment depression, hemodynamic variables, and pack-years of cigarette smoking, were coded as present or absent. Exercise Testing All patients had exercise testing done with a ramp treadmill protocol [7]. ST-segment depression was measured at the J junction to the nearest quarter millimeter, and ST slope, measured over the following 60 milliseconds of the ST segment, was classified as upsloping, horizontal, or downsloping. ST slope was coded as 1 for abnormal (horizontal or downsloping and at least 1 mm of depression) or 0 for normal slope. The 12-lead electrocardiograms were read by two cardiologists at each site on separate days, once by using raw signals and once by using the device averages. The cardiologists were blinded to patient identity and test results. Maximal and delta values for hemodynamic variables, along with exercise-induced hypotension, exercise-induced angina, and exercise capacity estimated in metabolic equivalents (METs) from the final treadmill speed and grade, were recorded. Angina during testing was classified according to the Duke Angina Index (2 if angina required that the test be stopped, 1 if angina occurred during or after the test, and 0 if no angina occurred) [8]. No test result was classified as indeterminate [9]. Medications were withheld only on the day of testing, and no maximal heart rate targets were applied [10]. Computer Analysis Electrocardiographic devices were used at all sites to simultaneously record in digital format all 12 electrocardiographic leads through exercise and recovery at 500 samples per second (Mortara Instrument, Milwaukee, Wisconsin) on optical disks [11]. Optical disk recordings were processed off-line by using a microcomputer at the exercise electrocardiography core laboratory. After the raw data were averaged, QRS measurement landmarks were determined by using software developed by Sunny-side Biomedical (Vista, California) [12]. Coronary Angiography Coronary angiography was done with standard techniques after administration of nitroglycerin. Trained observers at each site made blinded quantitative measurements. All stenoses with visual percentage narrowing greater than 30% were measured. Raw measurements were sent to the core angiography laboratory in Seattle, where they were converted to true diameters after correction for distortion. In a randomized selection, the mean difference per stenosis between the measurement in the core laboratory and measurements at participating sites was 0.9%, with a mean absolute value of 11.4%. Patients were categorized as having significant coronary artery disease if at least one stenosis with narrowing of 50% or more by quantitative measurement was present in any artery or branch with a reference diameter of at least 1 mm. Statistical Analysis The summary statistics were examined, and several variables were eliminated from the model building because of their low prevalence or low variance. Examination of the distribution of the visual ST-segment measurements and their relation to the angiographic results led to the choice to use raw visual interpretation of the maximal abnormal ST-segment depression in either exercise or recovery. On the basis of these results, 20 variables were chosen for multivariate analyses (Table 1 and Table 2). The training set for diagnosis of any coronary artery disease was divided into two groups, one with and one without significant angiographic coronary artery disease. After a logistic regression Equation was developed for predicting pre-exercise test probability for coronary artery disease, the exercise test hemodynamic and nonelectrocardiographic variables were added to the pre-exercise test variables as candidates. This allowed variable selection for three additional models to predict post-exercise test probability for coronary artery disease and to compare the discriminating power of computerized and visual measurements. Table 1. Clinical Characteristics of Patients in the QUEXTA Study (n = 814)* Table 2. Hemodynamic and Visual Electrocardiographic Characteristics of Patients in the QUEXTA Study* On the basis of the protocol and previous publications, the computer variables considered for the equations were 1) ST/HR [heart rate] index calculated at ST0 and ST60, 2) the Hollenberg score, 3) depression at ST60 [ST amplitude 60 milliseconds after J junction] in V5 at a heart rate of 100 beats/min, 4) ST integral in V5 at 3.5 minutes of recovery, 5) ST slope in V5 at maximal exercise, 6) ST slope in V5 at 3.5 minutes of recovery, 7) ST amplitude at J junction with a horizontal ST slope in V5 at maximal exercise, 8) ST amplitude at J junction with a horizontal ST slope in V5 at 3.5 minutes of recovery, 9) ST60 in V5 at 3.5 minutes of recovery, and 10) ST60 in II at 3.5 minutes of recovery. The most ST60 depression and the sum of ST60 depression in the three major perpendicular leads (II, V2, and V5) at maximal exercise and 3.5 minutes of recovery, as well as ST60, ST0, and ST integral in V2 and II, were considered individually. Comparisons were based on the area under the receiver-operating characteristic (ROC) curve and on sensitivity at the fixed specificity for visual ST-segment depression. Table 3 shows the results obtained by comparing visual ST-segment depression separately with every other model. Table 3. Results from the QUEXTA Test Set Obtained by Using Unsimplified Multivariable Equations* Comparison with the pilot population (Appendix Table 2), which had a prevalence of disease similar to that of the study population and was tested by using the same methods, showed how effective the protocol was in reducing workup bias. Appendix Table 2. Values in Pilot Study Group Derived by Using Unsimplified Equations at the Specificity Matching Visual Analysis (69%) in the Pilot Group and the Specificity Matching Visual Analysis in the QUEXTA Test Set (85%)* Results Clinical and Resting Electrocardiographic Variables Table 1 shows summary statistics for clinical variables in the full diagnostic group. According to the angiographic criteria, 276 patients in the training set and 135 patients in the test set had coronary disease. We noted that in our patients, all of whom had stable chest pain, the probability of coronary artery disease was almost halved if the pain ever occurred at rest. Thus, pain at rest was included as a candidate variable. The pre-exercise test variables chosen by the logistic model for the pre-exercise test Equation included age (explaining 60% of total variance), chest pain type (explaining 30% of total variance), diabetes, and pack-years of smoking. Exerc

Usefulness of exercise-induced ST-segment depression in the inferior leads during exercise testing as a marker for coronary artery disease.

Multiple lead systems are shown to have a higher sensitivity than that of single leads for detecting coronary artery disease (CAD) during exercise testing, but the value of ST-segment depression isolated to the inferior leads is questionable. To ascertain the diagnostic accuracy of inferior limb lead II compared with that of precordial lead V5, a retrospective analysis of 173 men was performed (108 in a training population and 65 in a validation cohort). All patients had a standard exercise test and underwent diagnostic coronary angiography within 15 days of the exercise test (range 1 to 65). Sixty-three patients had greater than or equal to 1 coronary stenoses greater than or equal to 70%, or left main lesion greater than or equal to 50%, whereas 45 patients in the training population did not. Exclusion criteria were female sex, left ventricular hypertrophy, left bundle branch block or resting ST-segment depression on the baseline electrocardiogram, previous myocardial infarction or revascularization procedures, and any significant valvular or congenital heart disease. Lead V5 had a better combination of sensitivity (65%) and specificity (84%) (chi-square = 24.11; p less than 0.001) than that of lead II (sensitivity 71%, specificity 44%) (chi-square = 2.25; p = 0.13) at a single cut point, and this improved specificity was substantial (95% confidence interval for observed difference 22 to 58%). Receiver-operating characteristic curve analysis also revealed that lead V5 (area = 0.759) was markedly superior to lead II (area = 0.582) over multiple cut points (z = 3.032; 2p = 0.002).(ABSTRACT TRUNCATED AT 250 WORDS)

Comparison of computer ST criteria for diagnosis of severe coronary artery disease

To determine which computer ST criteria are superior for predicting patterns and severity of coronary artery disease during exercise testing, 230 male veterans were studied who had both coronary angiography and a treadmill exercise test. Significant (p < or = 0.05) differences in computer-scored ST criteria were observed among patients with progressively increasing disease severity. Three-vessel/left main disease produced responses significantly different from 1- and 2-vessel disease or those with < 70% occlusion. Discriminant function analysis revealed that horizontal or downsloping ST depression measured at the J junction during exercise or recovery, or both, was the most powerful predictor of severe disease. With use of a cut point of 0.075 mV ST depression, horizontal or downsloping ST depression alone yielded a sensitivity of 50% (95% confidence interval = 35 to 65%) and specificity of 71% for prediction of severe disease; the only additional variable that added significantly to the prediction was exercise capacity, which improved sensitivity to 57% (95% confidence interval = 41 to 72%) with no change in specificity. Measurements of ST amplitude at the J junction and at 60 ms after the J point without slope considered and other scores, including the Treadmill Exercise Score, ST Integral, and ST/heart rate index, had a lower but comparable predictive accuracy when compared with horizontal or downsloping ST depression. Prediction of coronary artery disease severity can be achieved using computerized electrocardiographic measurements obtained during exercise testing. The most powerful marker for severe coronary artery disease is the amount of horizontal or downsloping ST-segment depression during exercise or recovery, or both, a measurement that stimulates the traditional visual approach.

Problems and limitations of ECG baseline estimation and removal using a cubic spline technique during exercise ECG testing: Recommendations for proper implementation

One common variety of exercise-induced artifact is baseline wander resulting from movement, respiration, and poor electrode contact. Although filters can be designed to remove much of this baseline variation, they will distort the low-frequency components of the ECG complex, such as the TP-segment, the PR-segment, and, most problematically, the ST-segment. The ST-segment is the most diagnostically relevant measure of the ECG taken during exercise. While linear baseline interpolation and removal may be adequate at lower heart rates, they also will introduce significant distortions. This is particularly evident when excessive nonlinear wander is present, as seen at higher heart rates and respiration rates. A nonlinear, third-order, polynomial estimator of baseline wander, known as the cubic spline, has been used for nearly 15 years. It is a very robust technique applied to exercise ECG recordings. Since the cubic spline is not a filter and use an a priori knowledge of the shape of the ECG signal, it estimates the true baseline and avoids distortion better. The more common implementations of this technique use relatively short ECG recordings. With the advent of increasing power in computerized ECG systems, the implementation of the cubic spline algorithm for removing baseline wander in continuous, longer-duration ECG records and in real-time processing is being attempted. However, the correct application of the cubic spline to continuous recordings is not straightforward and involves a number of previously unforeseen difficulties. The accuracy and resolution of both floating point and integer operations is critical during long-term application of the cubic spline function.(ABSTRACT TRUNCATED AT 250 WORDS)

Comparison of Computerized and Standard Visual Criteria of Exercise ECG for Diagnosis of Coronary Artery Disease

Objective: To review past and current studies of computerized exercise ECG criteria in order to establish which, if any, are superior to standard visual analysis for the diagnosis of coronary artery disease (CAD).

A real-time data-logger system using an optical disk WORM for archiving continuous 12-lead ECG data during exercise testing

An exercise ECG analysis program was developed over 15 years on a number of mainframes, minicomputers and, most recently, microcomputer-based systems. It has been rehosted into both Motorola MC68000 and Intel 80286 microprocessor-based development systems and is currently used with a removable 200 Mbyte optical disk (Write-Once-Read-Many, WORM) based data-logger system that can record and store all 12 leads simultaneously and continuously for an entire exercise test (up to 38 minutes). Data is acquired with 12-bit A/D resolution at 500 samples/sec. All ECG data and patient information are archived on the optical disk for later off-line recall and analysis on a PC or real-time replay through a D/A converter. Recorded ECG signals are at patient levels so they can be replayed through the patient cable box on any commercial system. Current development includes both simultaneous on-line processing and storage of 12-lead ECG data and off-line processing and development performed on the long-term, continuous ECG data being archived on optical disk. Patient medical histories and clinical information are separately entered into an applications database, where ECG measures and test results are later included. This new optical disk based exercise ECG database contains more than 600 complete exercise tests and is projected to increase to nearly 3,000 within 2 years.

Classification of CAD severity using fusion neural net analysis of multiple exercise ECG waveform representations

A hierarchical fuzzy neural-net approach has been developed to classify averaged serial ECG waveforms gathered during exercise testing in order to determine the severity of coronary artery disease (CAD). The ST-T complex of each ECG was first transformed on a sample-by-sample basis to form multiple representations (e.g. raw amplitudes, delta-baseline, slope through J-junction) to highlight the salient features of the signal. These representations were then initially classified by individual neural-nets and the resultant CAD-group outputs used as inputs into a secondary fusion-net for final classification. Also at the fusion-net level, additional parameters (e.g., HR and phase) were added to the fusion-nets input dimension space. Using only one lead, the processing gives nearly perfectly discrimination between angiographic normals and patients with severe 3-vessel disease. The use of FMM neural-nets is particularly relevant for this type of medical application since they allow the user to determine why and where the network decided on its results.<<ETX>>

Computerized exercise ECG testing and measurement for optimizing the prediction of coronary artery disease

A series of studies has been undertaken to evaluate computer processing of the exercise electrocardiogram (ECG) for predicting the presence and severity of angiographic coronary artery disease (CAD). Over 400 patients at the Long Beach VA Medical Center have undergone both exercise treadmill testing and cardiac catheterization with exercise ECG being stored on optical disks. Preliminary analyses have focused on visual-versus-computer measures, the contribution of traditional ST measures, and proposed exercise ECG scores. A summary of key findings is reported. Both discriminant function analyses and receiver operating characteristic curves have been used to make comparisons. The strongest predictors of CAD have been found to be ST changes during recovery.<<ETX>>

Exercise ECG analysis and measurement using an expert system approach

The difficulties of computerized analysis and measurement of the ECG (electrocardiogram) during exercise testing are compounded by the increases in noise and signal complexity at higher workloads and heart rates. By incorporating a hybrid expert-system structure in the analysis approach, information from the serial exercise ECG records are submitted to statistical and heuristic decision-logic to improve the accuracy of detection and measurement algorithms. Thus, in a manner similar to an expert cardiologist, the ECG analysis utilizes findings from previous records to aid in its determinations during each current record as the exercise test progresses.<<ETX>>