Methods

Top Methods Results Discussion Author & Article Info References

Patients

Patients were selected from a consecutive series of 3609 persons who underwent routine clinical exercise testing between 1984 and 1990; 30% of this group had coronary angiography within 3 months of testing and were excluded from the analysis. Also excluded were women (who constituted less than 2% of the sample), patients with significant valvular disease, and those who had previous coronary artery bypass surgery. Most of the remaining 2456 (84%) patients had been referred for testing because of chest pain or for the evaluation of exercise capacity.

Clinical Definitions

Myocardial infarction was defined by the presence of two or more of the following factors: 1) serial electrocardiographic changes; 2) typical chest pain; and 3) myocardial enzyme increase. Congestive heart failure was defined by typical symptoms and signs, plus echocardiographic or radiographic confirmation of cardiomegaly and pulmonary edema. Before treadmill testing, angina pectoris was classified as typical if the patient described substernal pressure, tightness, or pain that was brought on by exertion or emotions, lasted several minutes, and was relieved by nitroglycerin or rest. Angina was considered atypical in the absence of one or more of these features if the pain was thought to be cardiac in origin.

Exercise Testing

The exercise test was done using a standard progressive treadmill protocol [5]. Except for patients undergoing testing before discharge after myocardial infarction, each test was sign or symptom limited using standard recommended criteria for termination [2]; fatigue or chest pain was the reason for termination in most patients. In addition to the maximal systolic blood pressure achieved, the blood pressure response during exercise was coded as a score reflecting exercise-associated changes in systolic blood pressure (0 points = increase > 40 mm Hg; 1 point = 31 to 40 mm Hg; 2 points = 21 to 30 mm Hg; 3 points = 11 to 20 mm Hg; 4 points = 0 to 11 mm Hg; and 5 points = decrease below standing systolic blood pressure taken before testing) [6]. The treadmill was stopped abruptly at the completion of exercise, and the patient was placed in the supine position within 1 minute [7]. Exercise capacity was estimated in multiples of resting oxygen consumption (METs) and was also analyzed as a percentage of normal for age according to an equation derived from a "normal" subset of our referral group [8].

Electrocardiographic Measurements

Left ventricular hypertrophy was coded according to Romhilt and Estes criteria [9]. Patients lacking left ventricular hypertrophy with more than 0.5 mm ST depression in any lead were coded as having "resting ST depression." The exercise electrocardiogram was interpreted as previously described [7].

Measurement of Outcome Variable

Since 1984, the Department of Veterans Affairs Health Care System has developed a series of programs to support Veterans Affairs Medical Center clinical functions as part of the Decentralized Hospital Computer Project (DHCP). Death certificates are routinely completed by Veterans Affairs Medical Center physicians for inpatient and outpatient deaths. Information on care received elsewhere is routinely requested for clinical purposes, and all patients were scheduled for routine appointments at 6-month intervals after testing. Data on hospitalizations and deaths are entered, and retrieval programs are available to obtain dates and information regarding the most recent clinical visit and prescription received as well as those regarding hospitalization or death. To avoid bias, the coding of death certificates and other outcome variables was blinded to the predictor (exposure) variables. Although not designed for research purposes, this administrative and clinical database helped us obtain complete follow-up information.

Data Analysis

All data were entered into R:Base (Microrim, Redmond, Washington) and were analyzed using R:Base, Statgraphics (Statistical Graphics Corporation, Rockville, Maryland), True Epistat (Epistat Services, Richardson, Texas), Confidence Interval Analysis (American College of Physicians, Philadelphia, Pennsylvania), and EGRET (SERC, Seattle, Washington) on a standard 80386-SX-based personal computer (Vectra RS/20C, Hewlett Packard, Palo Alto, California). Survival time in person-days was measured from the time of the exercise test and was censored at the time of noncardiac death, coronary artery bypass surgery, or percutaneous transluminal coronary angioplasty.

Survival Analysis

Analysis was done to predict cardiovascular deaths and infarct-free survival (that is, cardiovascular death and nonfatal myocardial infarction). Kaplan-Meier survival curves were evaluated stratifying one or more variables to explore the data for interactions. The Cox proportional-hazard model was then applied to clinical and resting electrocardiographic variables, hemodynamic variables from treadmill testing, and electrocardiographic changes and angina during the treadmill test. Each variable grouping was also analyzed independently and by combining the strongest or most logical variables. Analysis was also done on the total group, including those who underwent catheterization (588 patients) because they were seen before the decision to catheterize.

Results

Top Methods Results Discussion Author & Article Info References

Follow-up

Computed clinical information was available for all 2546 patients, and follow-up was initiated in February 1991. Of these, 85% were confirmed to be alive by a clinic visit or prescription filled at a minimum of 1 year after their treadmill date, and 187 (7.5%) had died after a mean follow-up period of 45 ± 17 months. Contact either by telephone or letter led to follow-up and verification of vital status in 99%. After review of autopsy, death certificate, or hospital charts, 119 of the deaths (63%) were classified as cardiovascular. Forty-four patients had nonfatal myocardial infarctions, 34 developed congestive heart failure, 46 underwent coronary bypass surgery, and 18 received one or more angioplasties. The average annual cardiac mortality rate was 1.5%.

Clinical Characteristics

Table 1 shows the clinical characteristics of the study cohort grouped by end point. The mean age (±SD) was 59 ± 10 years. One fifth of the patients had typical angina pectoris, and one fifth had a history of previous myocardial infarction or electrocardiograms with diagnostic Q waves. Medications were not changed or withheld before exercise testing; 22% were taking ß-blockers, and 8% were taking digoxin. Statistically significant differences between the "no cardiovascular event" and "cardiovascular death" groups were observed for age, congestive heart failure, myocardial infarction, digoxin use, and most resting electrocardiographic abnormalities (P < 0.01).

Hemodynamic and Electrocardiographic Responses

Group averages for pre-exercise standing heart rate, systolic blood pressure, and double product were 76 beats per minute, 130 mm Hg, and 9800 (heart rate times systolic blood pressure), respectively. Table 2 shows the hemodynamic and electrocardiographic responses during the exercise test. No significant differences were found among end point groups for perceived exertion and occurrence of premature ventricular contractions.

Cox Proportional Hazards Model

The univariate scores and P values for the variables are listed in Appendix Table. No significant interactions were discovered, and thus none are included. Similar results were obtained both when infarct-free survival was considered as an end point (variable order, coefficients, and level of significance) and when the entire cohort was analyzed. The score test statistic listed is the relative weight or importance assigned the variables in the Cox model.

View larger version (15K): [in this window] [in a new window]   Figure 1. Kaplan-Meier survival curves using the clinical-exercise test score to predict cardiovascular death.

In addition, the Duke exercise test score was calculated [15, 21]. Because angina was not coded as the reason for stopping, the treadmill angina index was modified to code angina as 0 for absent and 1 for present. We used METs instead of minutes of exercise (Duke Treadmill Score = METs –5 x [mm ST depression during exercise] –4 x [treadmill angina index]). Figure 2 compares the receiver operating characteristic curves for the Duke score and the Veterans Affairs score predicting cardiovascular deaths in the total group. The area under the Veterans Affairs score curve (0.78) was statistically greater (Z = 2.8, P < 0.01) than the area under the Duke score curve (0.71). The numbers on the curves are the respective cut points for the scores. The Cox hazard model chose the same variables and similar coefficients in both the 2546 patients excluded from catheterization and in the total group of 3134 with the 588 selected for catheterization included.

View larger version (20K): [in this window] [in a new window]   Figure 2. Receiver operating characteristic curves using the Duke treadmill score and the Veterans Affairs clinical-exercise test score to predict cardiovascular deaths (the numbers along the curves are respective cut-points of the scores).

Discussion

Top Methods Results Discussion Author & Article Info References

Since the pioneering studies from the University of Alabama [11, 12], many investigators have used clinical, exercise test, and catheterization data to predict prognosis in patients with coronary artery disease. Implicit in these studies has been the issue of which variables are predictive and whether exercise testing and coronary angiography offer sufficient independent predictive power to justify their use for this purpose. A careful literature search yielded nine studies [13-20], including our first report, that applied multivariate survival analysis techniques to this problem. Some investigators combined variables, others did not consider key variables, and still others excluded patients with certain clinical features (such as congestive heart failure and digoxin use). Of these nine studies, three found an association between death and a history of congestive heart failure; three, between death and exercise systolic blood pressure; and two, between death and resting ST depression. In contrast to our first study, three found exercise-induced ST depression to be predictive of death, and six showed an association between low exercise capacity and mortality rate. Only the study by Marks and colleagues [15] did not consider maximal systolic blood pressure. One explanation for these divergent results is the possibility that "work-up" bias selected these study samples for cardiac catheterization.

The first study from Duke [15] used only hospitalized patients, all of whom had a catheterization, whereas a more recent report only included outpatients evaluated before the decision to do cardiac catheterization [21]. Their score, based on treadmill time, exercise-induced ST depression, and an angina score during the test, performed as well for prognosis as it did in the first report. Therefore, "work-up" bias did not affect their prognostication model. We attempted the same type of validation in this study. In contrast to their strictly exercise test-based score, we included exercise systolic blood pressure and clinical data in our model. Although history of congestive heart failure or digoxin use was the most powerful variable in both of our studies, surprisingly, different exercise test variables were chosen. The model from our first Veterans Affairs study in patients selected for catheterization only included exertional hypotension, whereas the model from this second Veterans Affairs study (only noninvasive clinical evaluation) found exercise-induced ST depression, exercise systolic blood pressure, and exercise capacity to have predictive power. Because our score considered a wider range of variables and was developed in this Veterans Affairs cohort, it outperformed the score developed at Duke. More importantly, both scores functioned well.

The "work-up" bias [22] inherent in choosing patients for cardiac catheterization in our first study [5] resulted in a sicker, older, more disabled group with a higher annual cardiac mortality rate (2.6% compared with 1.5%). In contrast, this second study included a cohort with a near-normal, age-adjusted exercise capacity. Age is not chosen as a predictive variable by most of the studies, including ours, because of the narrow age range for patients referred for evaluation of possible coronary disease and its relation to other variables. All studies appropriately dealt with interventions that alter the natural history, but each censored on them as we did, except for the earlier Veterans Affairs Coronary Artery Bypass Study [19]. The failure of ST depression to predict prognosis in our first study and in five of the other eight studies was probably due to the fact that the clinical process of selecting patients for cardiac catheterization was highly effective in selecting high-risk patients associated with exercise-induced ST depression, which variably affected its impact.

Enthusiasm for cardiac catheterization may have led to an acceptance of invasive measurements as superior to clinical variables for prognosis in patients with coronary artery disease. Although clinical variables were mentioned in early studies, key variables were often not considered, nor were they considered together or defined as accurately as they are today. Rapid technologic advancement has led to the assumption that laboratory methods and images are more accurate and precise than simple clinical data. In addition, studies have suggested that patients are selected for cardiac catheterization for reasons other than associated high risk [23, 24].

An important limitation of this type of study is the assumption of the Cox model that censoring is a random event. Interventions on which we censor are not random, however, and are instead linked to exercise-induced ST depression as well as to other variables. Nevertheless, all of the other studies had this same limitation, and because a true natural history study is not ethically possible, results obtained in this manner will probably continue to provide the best possible guidelines for practice. Also, because only male veterans were considered in our study, the results cannot be generalized to other groups or to the general population until they are validated.

Conclusions

Physicians need to guide their practice based on studies of patients seen in clinical practice. Patients selected for cardiac catheterization may not represent such a sample. Based on clinical and exercise test data, however, patients with signs and symptoms of coronary heart disease can be classified into low- and high-risk categories. This simple stratification could have a large effect on the appropriate use of cardiac catheterization. A common clinical problem lies in justifying intervention to improve survival for patients whose symptoms respond satisfactorily to medical management. Many studies have shown that simple clinical indicators can stratify these patients with stable coronary artery disease into high- or low-risk groups. Cardiac catheterization is not needed to stratify most of these patients. In the Veterans Affairs patient population, a history of congestive heart failure or digoxin use and three exercise test responses (systolic blood pressure, exercise capacity, and ST segment depression) are the most important predictors of cardiovascular death.