Exercise Tomographic Thallium-201 Imaging in Patients with Severe Coronary Artery Disease and Normal Electrocardiograms

Methods

We selected the study group from 18 076 consecutive patients who were referred to the Mayo Clinic Nuclear Cardiology Laboratory for exercise thallium studies between November 1986 and June 1992. Of these, 2638 patients had coronary angiography within 6 months of the exercise study and had no intervening cardiac event, valvular heart disease, previous revascularization, or left bundle-branch block. Patients were excluded from the study if they had previous myocardial infarction (n = 1194), digitalis within 48 hours of the exercise thallium-201 study (n = 175), a technically unsatisfactory study (n = 14), or any abnormalities shown on the at-rest electrocardiogram (n = 844) other than sinus bradycardia, as defined by Gibbons and colleagues [13]. These abnormalities included pathologic Q-waves, abnormalities of rhythm, ventricular conduction, QRS voltage, and axis or any ST-T wave changes, including minor nonspecific abnormalities. The study group comprised 411 patients who met all of the entry criteria and none of the exclusion criteria.

We prospectively collected data for clinical, exercise, and thallium-201 variables for each patient at the time of the thallium-201 exercise study, as detailed in previous reports Table 1[7, 18]. Diabetes was classified for each patient as present or absent; when present, the diabetes was classified as that which either required insulin or did not require insulin.

Exercise Protocol

All patients had treadmill exercise using either the Bruce protocol (n = 368) or the Naughton protocol (n = 43). For patients using the Naughton protocol, we applied a conversion factor to equate exercise duration with the Bruce protocol [19]. Heart rate and electrocardiographic rhythm strips (leads II, aVF, and V₅) were obtained continuously; blood pressure by cuff sphygmomanometry and 12-lead electrocardiograms were obtained for each level of exercise. We measured ST-segment displacement 80 ms after the J point and classified it as less than 1, 1 to 1.9, or ≥ 2 mV depression. At peak exercise, 4 mCi of thallium-201 was injected intravenously, and patients (n = 161) exercised an additional minute. Because of a change in protocol in January 1991, 250 patients received 3 mCi of thallium-201 at peak exercise and an additional 1 mCi of thallium-201 3.5 hours after exercise (30 minutes before delayed imaging).

Tomographic Thallium-201 Imaging

Imaging was done while patients were in the supine position 10 to 15 minutes after exercise and again 4 hours later. We initially obtained a single 5-minute anterior planar image to assess cardiac size and the ratio of pulmonary-to-myocardial thallium-201 uptake. Tomographic imaging was then done over a 180-degree arc for 30 images using previously described techniques [7]. Filtered back projection was done using a Ramp-Hanning filter and previously described methods[7].

Visual Analysis of Tomographic Images

Two experienced observers visually assessed the thallium tomographic images. Cardiac enlargement and increased pulmonary uptake were graded as present or absent. In borderline cases, pulmonary uptake was quantified; a ratio of maximal pulmonary-to-myocardial counts of greater than 0.50 was considered elevated. We then divided short-axis images into 14 segments as previously described [20]. We used a 5-point scoring system to assess each segment on both the postexercise and delayed images (4 = normal, 0 = no perfusion). We calculated a global score for each set of images by adding the values assigned to each of the 14 short-axis segments. Therefore, a maximum global score was 4 × 14, or 56, for a normal set of images. An abnormal segment was considered present if perfusion improved one or more grades from the post-stress to the delayed images or if a fixed defect of at least moderate severity was present (≤ 2 grades). We did not consider mild fixed defects (3 grades post-stress and 3 grades delayed) to be abnormal so that potential breast and diaphragmatic attenuation would not be labeled as abnormal. Short-axis segments were also grouped into three coronary distributions as previously described [7].

Coronary Angiography

All patients had coronary angiography within 6 months of thallium-201 exercise testing. A review of patients with normal at-rest electrocardiograms studied in our laboratory between 1986 and 1992 showed that the odds for being referred for coronary angiography was 4:1 for patients with an abnormal thallium scan compared with those with a normal scan. We visually estimated coronary artery narrowing and expressed it as percent luminal diameter stenosis. A 70% narrowing of the internal diameter of the left-anterior descending, left circumflex, and right coronary arteries or their major branches and a 50% narrowing of the left main coronary artery were considered significant [21].

Follow-up

All patients were contacted by letter or telephone. Events were determined by physician contact or hospital records. We defined significant cardiac events as nonfatal myocardial infarction or cardiac death. For the primary analysis, patients who had coronary artery bypass grafting or coronary artery angioplasty were not censored at the time of revascularization. A secondary analysis was done with patients censored (that is, not included in the survival analysis) at the time of revascularization.

Cost Analysis

We determined all costs for tests using 1992 reimbursement fees by Medicare in the State of Minnesota. The cost of exercise thallium scintigraphy includes the cost of a standard exercise treadmill.

Statistical Analysis

We used an unpaired t-test or Pearson chi-square test to identify variables significantly associated with three-vessel or left main coronary artery disease. Logistic regression analysis was used to develop multivariable models for predicting the presence or absence of three-vessel or left main coronary artery disease [22]. Variables were considered for stepwise inclusion in the model until a simultaneous test of all variables not entered was not significant (that is, P > 0.05). We used this strict criteria for model entry to prevent the detection of spurious associations. First, to construct a clinical model, we considered clinical variables alone. Exercise variables and, subsequently, thallium-201 variables were then considered for inclusion in the model to determine whether they added independent predictive value. No thallium-201 variables met the strict entry criteria once the clinical and exercise variables were considered. Because thallium-201 variables have previously been shown to be highly predictive of severe coronary disease [6, 7], the inclusion criteria were relaxed and thallium-201 variables were entered into the model based on their individual P value rather than on the simultaneous test of residual variables. We then used each logistic regression model to estimate the probability of developing three-vessel or left main coronary artery disease for each patient by classifying them as having low (<15%), intermediate (15% to 35%), or high (>35%) probability. We chose these probability groupings because they corresponded to those of previous studies [7, 18]. Patients were considered to be correctly classified if their predicted probability was low and they did not have three-vessel or left main coronary artery disease or if they were predicted to be at high risk and had severe disease. Incorrect classification was the reverse case: patients classified as having low probability who were ultimately found to have three-vessel or left main disease or those classified as having high probability who had no such disease. Patients with intermediate probability were not considered as correctly or incorrectly classified. We repeated the analysis with a second set of probability cutoff values (low, <10%; intermediate, 10% to 25%; high, >25%) to assess the sensitivity of the results to the cutoff values.

We did a cross-validation multivariate analysis to remove any bias in estimates of correct classification rates. In this analysis, we randomly divided the patients into 10 groups. Using 9 of the 10 groups, we developed a clinical model, a clinical plus exercise model, and a clinical plus exercise plus thallium-201 model and applied them to the remaining group to obtain a predicted probability of developing three-vessel or left main disease for that decile of patients. The process was repeated for 10 analyses so that each patient was classified by a predicted probability of developing severe disease based on a model derived from the remaining 90% of the study group. We determined a correct change to be an increase in the predicted probability of severe disease in patients who had severe disease or a decrease in probability in patients without severe disease. The number of patients with a correct change by this jack-knife analysis was generated for the sequential addition of exercise and then thallium variables to the clinical model.

We used the net increase in correct classifications minus the net decrease in incorrect classifications as a measure of improvement between successive models. This can be expressed as a weighted combination of the number of patients who change probability categories (with weights corresponding to the number of categories changed) with a positive or negative sign according to whether the change represents an improvement or worsening. The standard error for this net improvement measure was based on a five-category multinomial distribution.

We generated receiver-operator characteristic curves for the three incremental models using either the standard logistic regression or jack-knife method and compared them by paired rankings of the predicted probability of developing severe disease for each patient by model (that is, the probability of the clinical model compared with that of the clinical plus exercise model). We used the Kaplan-Meier method to estimate event-free survival curves for subgroups of patients identified as having low, intermediate, or high probability of developing three-vessel or left main coronary artery disease on the basis of the clinical and exercise model and the clinical, exercise, and thallium-201 model. We used a log-rank test to compare the curves.

Results

Of the 411 patients in the study group, 77 (19%) had significant three-vessel or left main coronary artery disease; of these, 33 (43%) had significant left main disease. One hundred thirteen patients (27%) had no significant coronary artery disease, 117 patients (28%) had single-vessel disease, and 113 patients (27%) had two-vessel coronary artery disease but not left main disease.

Univariate Analysis

We observed multiple significant differences between patients with and those without three-vessel or left main coronary artery disease (Table 2). More frequently, patients with disease were older, male, and diabetic and had a history of typical angina. Patients with severe disease exercised for a shorter time and to a lower peak heart rate and peak heart rate x blood pressure product, developed chest pain more frequently, and had a greater magnitude of ST-segment depression. Patients with three-vessel or left main coronary artery disease had a lower thallium global left ventricular score after exercise (indicating a larger stress perfusion defect) and greater change in global score (indicating greater redistribution) than patients without such disease. We noted no significant difference between the groups regarding increased pulmonary-to-myocardial uptake ratio of thallium-201, which was infrequent.

Multivariable Analysis

A summary of the multivariable analysis is shown in Table 3. When we considered only clinical variables Table 1, diabetes mellitus, history of typical angina, sex, and age were independently associated with the presence of three-vessel or left main coronary artery disease. When we considered exercise variables Table 1, the magnitude of ST depression with exercise (P < 0.001) and the heart rate-blood pressure product at peak exercise (P < 0.001) added independent information to the four significant clinical variables, and the model chi-square more than doubled to 65.0. When the relaxed entry criterion was used, only the global thallium-201 score redistribution Table 1 added independent information (P = 0.02) to the six clinical and exercise variables. The overall chi-square increased from 65.0 for the clinical and exercise model to 70.0 for the clinical, exercise, and thallium-201 model. When we considered all variables simultaneously for step-wise inclusion Table 1, as in standard logistic regression analysis, we selected the following variables as independent predictors of developing three-vessel or left main disease in the following order: peak heart rate-blood pressure product, magnitude of ST depression, redistribution thallium score, diabetes mellitus, sex, and history of typical angina.

Prediction of Probability Groups

Standard Logistic Regression

We classified patients into probability groups (low, <0.15; intermediate, 0.15 to 0.35; high, >0.35) by each of the three logistic regression models (Figure 1). When the clinical model was used, 189 patients (46%) were correctly classified (low probability with absence of severe coronary disease or high probability with presence of severe coronary disease), and 37 patients (9%) were incorrectly classified (low probability with presence of severe coronary disease or high probability with absence of severe coronary disease). The clinical and exercise model increased the number of patients correctly classified by 37 but at the expense of 13 additional patients incorrectly classified, for a net of 24 additional correct classifications (6% of the study group; SE = 3%). The clinical, exercise, and thallium-201 model led to 12 additional correct classifications and 2 fewer incorrect classifications, for a net increase of 14 correct classifications (3% of the study group; SE = 2%). All of these reclassified patients moved from the intermediate- to low-probability group.

The results were similar when the second set of probability cutoff values were used (low probability, <0.10; high probability, >0.25). The exercise and clinical model led to the correct classification of 188 patients (46%). The clinical, exercise, and thallium-201 model resulted in 3 more correct classifications (<1%) and 2 more incorrect classifications, for a net of 1 additional correct classification over clinical and exercise variables.

Cross-Validation (Jack-knife) Method

The cross-validation multivariate analysis showed that 25 patients correctly changed probability classifications when we added exercise variables to the clinical variables (6% of the study group; SE = 3%). The addition of thallium-201 variables to clinical and exercise variables resulted in only two additional net correct classifications (0.5% of the study group; SE = 2%). The cross-validation analysis for the probability cut-off values of less than 0.10, 0.10 to 0.25, and greater than 0.25 resulted in 72 patients correctly changing probability groups when the exercise variables were added; however, 6 patients incorrectly changed groups when we added thallium variables to the clinical and exercise variables.

The receiver-operator characteristic curves for all three models derived by standard logistic regression analysis are shown in Figure 2. The curves of the clinical model and the clinical exercise model significantly differed (P = 0.05), but we noted no significant difference between the curves for the clinical plus exercise model and the clinical plus exercise plus thallium-201 model. The curves were similar when we used the cross-validation analysis. Figure 3 is a graph for predicting the probability of severe coronary artery disease for any individual patient on the basis of clinical and exercise variables derived from the multivariate model (Table 3).

Cost Analysis

When we used only clinical variables, we correctly predicted the presence or absence of three-vessel or left main coronary artery disease in 189 patients (46%). When we used the Minnesota Medicare reimbursement fee for exercise treadmill testing of $89, the cost per additional correct classification with exercise testing was $1524 per patient correctly classified (411 x $89 per 24 additional correct classifications). These values for exercise variables were essentially identical when we used the unbiased classifications derived from the jack-knife analysis. When we added thallium scintigraphy to exercise and clinical data using the standard regression method, 14 additional patients were correctly classified. When we used the Minnesota Medicare reimbursement for an exercise thallium-201 study of $700 (includes treadmill charge of $89), the cost was $20 550 per additional correct classification ([411 x $700]/14). The cost for thallium scintigraphy was considerably higher when we used the unbiased classifications from the cross-validation analysis: $143 880 per additional correct classification ([411 x $700]/2).

Follow-up Data

Early revascularization (3 or fewer months after thallium exercise testing) was done in 144 (35%) of patients (86 patients had coronary bypass surgery, and 58 patients had coronary angioplasty). An additional 54 patients (13%) had late revascularization. Of the patients having early revascularization, 49 (34%) had three-vessel or left main coronary artery disease. By design, no patient had a revascularization procedure between the angiogram and the exercise study. Early revascularization was done in 36 of 61 patients (59%) who had been classified as having a high probability of developing severe disease on the basis of the clinical, exercise, and thallium-201 model. However, we found a nearly identical rate of early revascularization in patients who had been classified as high risk on the basis of clinical and exercise variables alone (40 of 64 patients [63%]). Conversely, early revascularization was done in 51 of 218 patients (23%) who had been classified as low risk by the clinical, exercise, and thallium-201 model.

The mean follow-up for the study group was 2.8 ±1.0 years (median, 2.8 years; 25th to 75th percentiles, 2.0 to 3.4 years). Follow-up was 98% complete, with 10 patients lost to follow-up (7 of whom lived outside of North America). Twenty-one cardiac events occurred: 14 nonfatal myocardial infarctions and 7 cardiac deaths. Overall, the myocardial infarction-free survival rate was excellent (95%) after 4 years whether or not patients were censored (that is, excluded from the survival analysis) at the time of revascularization. Event-free survival (range, 94% to 97%) did not differ between patients classified as having a low probability and those classified as having a high probability of severe disease on the basis of the clinical and exercise model or the clinical, exercise, and thallium-201 model. We obtained similar results for patients who were censored and those who were uncensored at the time of revascularization.

Discussion

In clinical practice, test results should not be interpreted in isolation, although studies evaluating a particular diagnostic technique have been designed in this manner. An incremental approach to a multivariate model for predicting the presence of coronary artery disease using exercise thallium imaging has been validated in a multicenter study [23]. Such an approach allows the evaluation of cardiac imaging modalities in addition to what can be learned from clinical and exercise measurements alone. This is essential as the pretest probability changes (and the diagnostic certainty increases) as clinical and exercise data become known.

Previous reports have shown the value of thallium-201 scintigraphy in detecting three-vessel coronary artery disease, but patients with normal electrocardiograms in these studies were not separated [6, 7]. A previous report from our laboratory showed that 95% of patients with chest pain and a normal at-rest electrocardiogram who were referred for radionuclide ventriculography had a normal at-rest ejection fraction [24]. A population-based study showed that the prognosis for patients with normal resting electrocardiograms who had a diagnosis of angina pectoris was the same as that for patients without coronary artery disease [25]. In contrast, patients who presented with an abnormal electrocardiogram had a 10-year survival that was 72% of that expected [15]. Patients with previous myocardial infarction have a greater prevalence of severe coronary artery disease and usually will have abnormal at-rest electrocardiograms [26]. For these reasons, it can be expected that the utility of exercise thallium scintigraphy for detecting three-vessel or left main coronary artery disease will differ for the patients with normal electrocardiograms.

Our study confirms that thallium-201 scintigraphy adds limited incremental information to clinical and exercise variables when the resting electrocardiogram is normal. Forty-six percent of the study group could be correctly classified by clinical variables only (diabetes, sex, age, and history of typical angina) with no additional noninvasive testing and thus no additional cost. Exercise testing added additional information to clinical variables in these patients with normal at-rest electrocardiograms. Six percent of the study group were reclassified correctly by the addition of the heart rate-blood pressure product at peak exercise and the magnitude of ST depression from treadmill testing at a cost of more than $1500 per additional correct classification. The addition of the change in global thallium-201 score (a measure of thallium-201 redistribution) increased the chi-square of the clinical and exercise model. However, the number of patients actually reclassified composed only 3% of the study group. This net change cost more than $20 000 per additional correct classification; all of the patients moved from the intermediate- to low-risk group. The jack-knife analysis, which provides a less biased interpretation of the data, suggested that the incremental value of thallium imaging was even less.

The cost-effectiveness analysis of this study is of particular interest. It is difficult to assess how many clinicians on a national level refer patients with a normal at-rest electrocardiogram for exercise thallium-201 testing to screen for severe coronary artery disease. In Olmsted County, Minnesota, 59% of patients presenting with chest pain have normal at-rest electrocardiograms [25]. If the primary goal of a clinician is to identify patients with severe coronary artery disease, a careful history and standard treadmill exercise test can achieve results similar to those achieved with exercise thallium-201 scintigraphy at a much lower cost. Consequently, the potential savings in health care expenditure is potentially great if such an approach is routinely followed.

The power of clinical variables to predict severe coronary disease has been reported by Hubbard and colleagues [27]. In that study, age, history of typical angina, diabetes, sex, and electrocardiographic evidence of previous infarction led to the correct classification of 31% of patients for the presence or absence of three-vessel or left main coronary artery disease. Our study group differs from that of Hubbard and colleagues: Half of the patients in that study had previous myocardial infarction, and a higher percentage had three-vessel or left main coronary artery disease. In other studies that examined the ability of clinical variables to predict severe coronary disease, investigators also found a high prevalence of previous myocardial infarction [26, 28]. It appears from our results that clinical variables can be used to better discriminate between the presence and absence of severe disease when patients have no history of previous infarction.

Although more patients with three-vessel or left main coronary artery disease had an increased pulmonary uptake than those without, this was an infrequent finding. This probably reflects a high prevalence of normal resting left ventricular function in our study group. Previous studies of patients with abnormal at-rest electrocardiograms have shown elevated pulmonary uptake to be more prevalent and have found this variable to be an important determinant of severe coronary artery disease [16, 29].

The prognosis in patients with normal electrocardiograms was excellent, regardless of the predicted probability of having severe disease, a finding similar to that of Connolly and colleagues [25]. The limited incremental value of thallium-201 scintigraphy in patients with normal electrocardiograms is in agreement with three previous studies. Gibbons and coworkers [13] used the same probability groupings that we used in our study and found that only 3% of patients were correctly reclassified regarding the likelihood of severe disease when radionuclide angiography was added to the clinical and exercise variables. Simari and colleagues [14] showed a limited effect of radionuclide angiography on clinical and exercise variables for predicting prognosis in patients with normal at-rest electrocardiograms. A study from Cedars-Sinai Medical Center [15] showed that most prognostic information in patients with normal electrocardiograms (72%) was provided by the clinical variables alone. The addition of exercise testing and thallium scintigraphy only increased the prognostic value of the model by 5%. The event rate in their study group with normal electrocardiograms was similarly low at 4%.

Conversely, Pollock and coworkers [16] showed that thallium-201 scintigraphy added prognostic information to clinical and exercise measurements in a population that included patients with previous infarction and abnormal electrocardiograms. Ladenheim and associates [15] showed that in patients with abnormal electrocardiograms, clinical variables provided 58% of the prognostic information. The addition of exercise followed by thallium variables increased the prognostic power 14% in each instance [15]. However, both studies included patients with previous myocardial infarction, and the prevalence of elevated pulmonary uptake was markedly greater. It is evident from these studies that the use of thallium-201 scintigraphy for detecting severe disease and prognosis is best restricted to patients with abnormal at-rest electrocardiograms.

Limitations

The major limitation to our study is the influence of referral bias; the results of the thallium-201 exercise study probably influenced the decision to proceed to coronary angiography. This limitation is unavoidable in studies that use coronary anatomy as an end point unless all patients referred for exercise testing have coronary angiography regardless of the test results. The effect of referral bias on the sensitivity and specificity of a noninvasive exercise test has been well described [11, 30]. At the Mayo Clinic, the odds of being referred for coronary angiography for patients who have normal at-rest electrocardiograms and an abnormal thallium-201 scan is 4:1 for angiography. The prognostic analysis is limited by the number of patients referred for early revascularization (35%), particularly the high percentage with severe coronary artery disease. The rate of early revascularization was not influenced by the addition of thallium-201 variables to the clinical and exercise variables. However, it must be emphasized that we focused on the noninvasive identification of patients with severe coronary artery disease. Clearly, thallium scintigraphy was not necessary in this regard, nor were the scintigraphic findings better for prognostication than the combination of clinical and exercise variables.

Our study was done in a tertiary referral center. This patient population may thus not be representative of centers in which primary care predominates. Furthermore, qualitative interpretation of thallium-201 tomographic images varies among institutions, and therefore these results may not be readily applicable to other institutions. However, the relatively weak effect of thallium-201 scintigraphy in our study lessens the importance of this limitation.

The outcome of interest chosen in our study was left main or three-vessel coronary artery disease. It has been suggested that patients with proximal left-anterior descending artery disease and a second proximal stenosis may benefit from revascularization [31]. We did not classify this patient group as having severe coronary disease. However, only 13 patients in the study group had this particular disorder. Consequently, the decision to place them in the nonsevere group probably had little effect on the results.

Finally, we did not quantify the thallium tomographic images. A previous large study from Mahmarian and colleagues [32] showed no significant difference in the detection of multivessel coronary artery disease when qualitative or quantitative assessment of thallium-201 tomographic images was used. All images in our study were assessed by two experienced observers. For these reasons, it is unlikely that quantitative interpretation would have changed the results significantly. We did not evaluate transient ischemic dilatation of the left ventricle, a marker of severe disease [33], although we did evaluate postexercise cardiac size by planar imaging. We did not systematically quantify pulmonary uptake of thallium-201 on all patients but did subjectively assess it in all patients. Reinjection imaging was done in more than half the study group because of a change in laboratory protocol. It is possible that thallium-201 redistribution would have been a stronger variable if all patients had had reinjection imaging.

Conclusions

Despite these limitations, our results suggest that most patients with normal at-rest electrocardiograms can be classified accurately for the presence or absence of three-vessel or left main coronary artery disease using clinical and exercise variables. Thallium scintigraphy had little additional effect; thus, the routine use of sophisticated imaging techniques does not appear to be justified for this purpose.