Author + information
- Received July 6, 2000
- Revision received January 12, 2001
- Accepted January 29, 2001
- Published online May 1, 2001.
- Lazaro A Diaz, MD∗,
- Richard C Brunken, MD, FACC§,
- Eugene H Blackstone, MD, FACC†,‡,
- Claire E Snader, MA∗ and
- Michael S Lauer, MD, FACC∗,* ()
- ↵*Reprint requests and correspondence:
Dr. Michael S. Lauer, Director of Exercise Laboratory and Clinical Research, Desk F25, Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, Ohio 44195
The goal of this study was to determine the value of thallium201single photon emission computed tomography (SPECT) imaging for prediction of all-cause mortality when considered along with functional capacity and heart rate recovery.
Myocardial perfusion defects identified by thallium201SPECT imaging are predictive of cardiac events. Functional capacity and heart rate recovery are exercise measures that also have prognostic implications.
We followed 7,163 consecutive adults referred for symptom-limited exercise thallium SPECT (mean age 60 ± 10, 25% women) for 6.7 years. Using information theory, we identified a probable best model relating nuclear findings to outcome to calculate a prognostic nuclear score.
There were 855 deaths. Intermediate- and high-risk prognostic nuclear scores were noted in 28% and 10% of patients. Compared with those with low-risk scans, patients with an intermediate-risk score were at increased risk for death (14% vs. 9%, hazard ratio: 1.67, 95% confidence interval [CI]: 1.44 to 1.95, p < 0.0001), while those with high-risk scores were at greater risk (24%, hazard ratio: 2.98, 95% CI: 2.49 to 3.56, p < 0.0001). In multivariable analyses that adjusted for clinical characteristics, functional capacity and heart rate recovery, an intermediate-risk nuclear score remained predictive of death (adjusted hazard ratio: 1.50, 95% CI: 1.28 to 1.76, p < 0.0001), as did a high-risk score (adjusted hazard ratio: 2.76, 95% CI: 2.13 to 2.56, p < 0.0001). Impaired functional capacity and decreased heart rate recovery provided additional prognostic information.
Myocardial perfusion defects detected by thallium SPECT imaging are independently predictive of long-term all-cause death, even after accounting for exercise capacity, heart rate recovery and other potential confounders.
Myocardial perfusion imaging is used for diagnosing coronary artery disease and for risk stratification in patients with known or suspected disease (1–5). Although imaging results are often considered alongside results of exercise electrocardiography for prediction of cardiac events, recent studies have highlighted the ability of nonelectrocardiographic exercise measures, such as functional capacity (6,7)and heart rate recovery (8,9), to predict mortality. Relatively little is known about how to best integrate these exercise measures with nuclear perfusion abnormalities for optimal risk stratification. In this study, we sought to assess the independent long-term importance of thallium201single photon emission computed tomography (SPECT) perfusion abnormalities relative to functional capacity and heart rate recovery for prediction of all-cause mortality in a large cohort of patients. All-cause mortality was studied as an objective and unbiased end point (10,11).
Consecutive adults (age ≥30 years) referred for thallium SPECT imaging in conjunction with symptom-limited exercise testing between September 1990 and December 1993 at the Cleveland Clinic Foundation were studied. Patients with known heart failure or left ventricular dysfunction (ejection fraction ≤40%), valvular disease and a history of pacemaker implantation were excluded. Because we used the Social Security Death Index (12)for assessing outcome, foreign nationals were also excluded. The study protocol was approved by the local institutional review board.
The methods used for prospective study of patients undergoing stress testing at our institution have been described in detail elsewhere (6,13). Since 1990, all patients undergoing stress testing undergo a structured interview and chart review. Data collected prospectively include existing medical diagnoses, symptoms, risk factors, medications and prior procedures. All data are entered on-line before exercise testing; thus, clinical data collected are not influenced by stress or nuclear study results.
Exercise testing, functional capacity and heart rate recovery
The methods of exercise testing and prospective, computerized data collection used in our laboratory have been described in detail elsewhere (13,14). Estimated functional capacity in metabolic equivalents (METS) was classified as high, good, average, fair or poor for age and gender based on a classification system (6)that has been validated. Patients were unfit if their age- and gender-based functional capacity was fair or poor. Heart rate recovery was defined as the difference between heart rate at peak exercise and 1 min into a recovery cool-down period; a value of ≤12 beats/min was abnormal (8). We chose not to study chronotropic response to exercise (13,15)because patients taking beta-adrenergic blocking agents were not excluded. We did not analyze ST-segment responses or Duke treadmill scores (16,17)because of the presence of factors such as digoxin use or marked baseline electrocardiographic abnormalities that would limit their interpretation.
We have published scintigraphic methods used in our laboratory in the early 1990s (18,19). The heart was divided into twelve segments and, to avoid any inherent prognostic limitation related to a 12-segment model (20), each segment was weighted according to its relative contribution to total left ventricular mass: anterior (weighted by 3 mU), anterior-lateral (3 mU), anterior-septal (3 mU), septal (2 mU), apical (1 mU), inferior (2 mU), inferior-lateral (2 mU), inferior-septal (2 mU), lateral (3 mU), posterior (1 mU), posterior-lateral (1 mU) and posterior-septal (1 mU). Each segment was coded as normal, fixed (scar) or reversible (ischemia). The justification for mass weightings has been reported (20). In cases of suspected artifact or attenuation, the segment was considered normal.
As previously described (18), images were displayed and interpreted using quantitative color scales. Myocardial counts were referenced to peak myocardial activity on the resting images; thus, counts were expressed as a percent of maximal myocardial activity on the redistribution images. Segments were considered normal if the count variation within the segment was ≤20% compared with the segment with the highest count rate. For the apex and diaphragmatic segments, a range of 30% to 40% was allowed. Ischemia was considered present if the counting rate increased by ≥20% on redistribution images; if the segment activity increased by <20%, scar was considered present. Segments were only considered abnormal if the suspected defect was seen on two or more consecutive slices and could be verified in an orthogonal plane. Although only one reader evaluated each study, that reader was blinded to clinical history, results of prior coronary arteriograms, results of the exercise test (except for peak heart rate) and the hypothesis of this study.
The primary end point was all-cause mortality as assessed by the Social Security Death Index (12). We have described the high specificity and sensitivity of this measure in detail elsewhere (21). The mean follow-up was 6.7 years, with a minimum follow-up among survivors of 4.5 years.
Clinical, exercise and scintigraphic abnormalities were related to time-to-death using the Cox proportional hazards model (22). The proportional hazards assumption was verified by inspection of log (-log [survival function]) curves. Unadjusted survival estimates were calculated using the Kaplan-Meier product-limit technique (23), while adjusted survival functions and confidence limits were calculated and plotted based on a likelihood-based approach estimation of the baseline survivorship function (24).
A number of potential nuclear abnormalities have been considered as prognostically important, including the quantity and distribution of scar and ischemic defects (1). To determine how best to characterize perfusion defects for prediction of risk, we used information theory (25), first identifying an optimal model and then determining the most likely parameter estimates (coefficients). Seven potential models were chosen prospectively: 1) the presence of any defect; 2) the presence of any reversible defect; 3) the presence of defects in more than one vascular territory; 4) the total mass of left ventricular myocardium with perfusion defects of any type (on a scale of 0 to 24), in effect, a modified summed stress score (1); 5) the total mass of scar; 6) the total mass of ischemia; and 7) a composite of scar and ischemia mass, with each type of abnormality considered and weighted differently.
To determine which model was best, Akaike Information Criterion (25,26)values and weights were calculated for each model. As shown in Table 1, the best model was the composite of the masses of scar and ischemia, with a 92% probability of best fit. This was verified by performing a series of 100 bootstrap resamplings and Cox modeling for variable selection; the composite model was chosen as best 94% of the time. The bootstrap resamplings approach has been shown to be more efficient for model validation than the simple identification of separate “test” and “validation” datasets (27,28).
To determine parameter coefficients, 500 bootstrap resamplings were performed, and a composite prognostic nuclear score was derived from the resulting median values of each beta coefficient from the Cox regression model. The score was calculated as: score = scar mass + 0.5 (ischemia mass). We prospectively divided subjects into patients with normal scans (score = 0), patients with intermediate risk scans (abnormal scans but with scores below the 90th percentile, i.e., score >0 and <8) and patients with high-risk scans (scores above the 90th percentile, i.e., score ≥8).
The associations of intermediate- and high-risk composite prognostic nuclear scores with mortality were assessed in several prespecified subgroups, and potential interactions were explored. Particular attention was paid a priori to subgroups with impaired physical fitness, abnormal heart rate recovery or both, along with whether or not prior revascularization had been performed.
Multivariable models were used to assess the independent association of intermediate- and high-risk nuclear scores. The results were validated by two series of bootstrapping resamplings (27,29,30), with consideration of variable selection first and determination of parameter coefficients for variables that entered at least 50% of models second. A p value of <0.05 was required for variable retention. No variables were forced. We only considered variables available to clinicians at the time of the stress test. All analyses were performed using the SAS 6.12 system (SAS, Inc., Cary, North Carolina).
Baseline and exercise characteristics
The number of patients who were eligible was 7,163 (age 60 ± 10, 25% women). Thallium SPECT abnormalities were present in 2,817 patients (39%); reversible defects were noted in 1,201 (17%). Based on their composite nuclear prognostic score, patients were classified as low-risk, intermediate-risk and high-risk in 62%, 28% and 10% of cases. Fair or poor physical fitness was found in 2,674 patients (37%); an abnormal heart rate recovery of ≤12 beats/min was present in 2,612 patients (37%).
Baseline clinical and exercise characteristics according to the composite prognostic nuclear score are summarized in Table 2. Compared with patients with low-risk or intermediate-risk composite prognostic nuclear scores, patients with high-risk scores were older, more likely to be men and more likely to have diabetes, Q-waves, a history of prior coronary bypass surgery, more likely to be taking digoxin or beta-blockers and were more likely to have impaired physical fitness or an abnormal heart rate recovery.
Thallium SPECT perfusion defects and mortality
During 47,717 person-years of follow-up, there were 855 deaths. Compared with patients with normal nuclear scans who had a seven-year death rate of 9%, patients with intermediate- and high-risk nuclear scores had increased death rates of 14% and 24% (hazard ratios for intermediate- and high-risk scans 1.67, 95% confidence interval [CI]: 1.44 to 1.95 and 2.98, 95% CI: 2.49 to 3.56). Hazard ratios according to nuclear score in a number of prespecified subgroups were similar. An interaction was noted whereby high-risk scans were less predictive in patients taking beta-blockers (hazard ratios: 1.87 vs. 3.45, p for interaction: 0.004).
Figure 1shows seven-year mortality according to the number of exercise abnormalities and nuclear scores among patients who had not undergone prior revascularization. Patients who were physically fit and who had normal heart rate recovery were at low risk (<1% per year), irrespective of nuclear score. Conversely, patients who were both physically unfit and who had an abnormal heart rate recovery were at very high risk for death (>3% per year), even with normal nuclear findings. Patients who had either impaired physical fitness or an abnormal heart rate recovery, but not both, had intermediate to high risks for death, depending on the nuclear score. Figures 2 and 3⇓⇓show analogous seven-year mortality among patients who had undergone prior coronary artery bypass grafting or prior percutaneous intervention (but without prior bypass grafting). Within each of the three exercise groups, abnormal nuclear scans were associated with increased risk.
After adjusting for age; gender; resting blood pressure and heart rate; hypertension; diabetes; smoking; chronic lung disease; ECG findings including Q waves, left ventricular hypertrophy, left and right bundle branch block and atrial fibrillation; prior coronary bypass grafting; prior percutaneous revascularization; use of medications including aspirin, lipid-lowering drugs, beta-blockers, nondihydropyridine and dihydropyridines calcium channel blockers, angiotensin-converting enzyme inhibitors, vasodilators, diuretics and digoxin; functional capacity; heart rate recovery and interaction terms, an intermediate risk nuclear score remained associated with increased risk of death, as was a severe-risk score (Table 3). Other independent predictors of death included fair or poor physical fitness and an abnormal heart rate recovery. Covariate-adjusted survival functions according to nuclear scan results are shown in Figure 4.
To validate these results, a series of bootstrap resamplings with repetitive Cox proportional hazards modeling were performed. First, 100 resamplings for variable selection found that variables that entered models at least 90% of the time included high-risk nuclear score (100%), age (100%), abnormal heart rate recovery (100%), smoking (100%), use of digoxin (99%), intermediate nuclear score (99%) and diabetes (94%). A second set of 500 bootstrap resamplings that adjusted for these variables along with others that entered at least 50% of models demonstrated virtually identical adjusted hazard ratios (for intermediate-risk score adjusted hazard ratio: 1.50, 95% CI: 1.25 to 1.78; for high-risk score adjusted hazard ratio: 2.91, 95% CI: 2.24 to 3.72).
In this large cohort of patients referred for symptom-limited exercise treadmill thallium201SPECT testing, myocardial perfusion defects were independently predictive of long-term all-cause mortality, even after accounting for clinical differences, exercise capacity and heart rate recovery. Using information theory methods (25), we identified a composite model that incorporated both evidence of scar and ischemia as being best for prediction of death.
Because of the long period of follow-up (mean of 6.7 years with a minimum of 4.5 years for survivors) and the large number of outcome events (855 deaths), it was possible to carefully assess the relative associations of nuclear abnormalities and exercise test findings as risk predictors. Among patients who had not undergone prior revascularization (Fig. 1), exercise test findings could identify low-risk (<1% per year) and high-risk (>3% per year) groups with little additional prognostic information provided by nuclear findings. Those patients who had either impaired functional capacity or an abnormal heart rate recovery, but not both, represented an intermediate-risk group for whom the nuclear test was particularly useful in stratifying risk, consistent with the Bayesian theory (31).
A number of groups have demonstrated prognostic associations between nuclear perfusion abnormalities and cardiac events (1,4). This study expounds upon these in several respects. First, we found that nuclear findings predict all-cause death. As has been discussed elsewhere (10,11), the end point of “cardiac death” is inherently suspect due to poor quality, biased recording of death certificates and complexities of human disease (32). All-cause death is a relevant end point in a cardiac population, as shown in clinical trials in which an intervention designed to treat cardiac disease resulted in reductions in all-cause mortality (33).
Second, instead of comparing the nuclear scan with the exercise electrocardiogram or the Duke treadmill score (16,17), we compared it against functional capacity and heart rate recovery; exercise findings are among the strongest predictors of risk and are available in nearly all patients, regardless of medication use or baseline electrocardiographic abnormalities. Third, because of the large size of the cohort and number of events, we could do robust subgroup analyses of the prognostic usefulness of nuclear imaging among patients with and without exercise test abnormalities.
The scintigraphic methods used in our laboratory in the early 1990s (18,19)are not representative of the most modern techniques available. Newer isotopes, better acquisition protocols (34), gated SPECT imaging (35)and attenuation correction (36)may mean that nuclear images obtained now may have better prognostic value. By studying patients who were imaged in the early 1990s, however, we were able to analyze the impact of nuclear findings on long-term outcome.
Defect severity for each segment was not examined in the analysis. Whether defect severity might have had an impact on our study is uncertain. It is possible that measures of defect extent and severity might have both contributed to the prognostic value of imaging. There are certain inherent limitations with SPECT imaging techniques that fundamentally limit the ability to ascertain actual perfusion defect severity (37).
Segments thought to represent attenuation were considered normal. This was the most conservative approach as the appearance of artifacts may be surrogates for comorbid conditions and elevate event rates in patients classified as normal. Body mass index was not available. Exactly how consideration of attenuation might have affected the prognostic model is not entirely clear, as attenuation defects may also be associated with leanness, which may itself be associated with reduced mortality (38).
Other limitations are that our study was performed at a single tertiary care center, leading to possible biases of patient selection. We used bootstrapping to provide relatively unbiased low-variance estimates of prediction without having to sacrifice data for model development (28). Nonetheless, confirmation of our results on entirely independent external datasets is essential. We were unable to perform detailed analyses of left ventricular dilation and increased lung uptake of thallium. Formal measures of left ventricular function were not available. We did exclude patients with known heart failure or left ventricular systolic dysfunction, though, and we accounted for use of digoxin and diuretics in our multivariable analyses.
Despite these limitations, we found that myocardial perfusion abnormalities as assessed by thallium201SPECT imaging are powerful and independent predictors of long-term all-cause mortality. The value of exercise nuclear imaging for risk stratification is particularly enhanced when exercise capacity and heart rate recovery are also considered.
☆ Dr. Lauer is a recipient of an Established Investigator Award of the American Heart Association.
- confidence interval
- metabolic equivalents
- single photon emission computed tomography
- Received July 6, 2000.
- Revision received January 12, 2001.
- Accepted January 29, 2001.
- American College of Cardiology
- Hachamovitch R,
- Berman D.S,
- Shaw L.J,
- et al.
- Kaul S,
- Lilly D.R,
- Gascho J.A,
- et al.
- Machecourt J,
- Longere P,
- Fagret D,
- et al.
- Snader C.E,
- Marwick T.H,
- Pashkow F.J,
- Harvey S.A,
- Thomas J.D,
- Lauer M.S
- Lauer M.S,
- Blackstone E.H,
- Young J.B,
- Topol E.J
- Curb J.D,
- Ford C.E,
- Pressel S,
- Palmer M,
- Babcock C,
- Hawkins C.M
- Lauer M.S,
- Mehta R,
- Pashkow F.J,
- Okin P.M,
- Lee K,
- Marwick T.H
- Lauer M.S,
- Okin P.M,
- Larson M.G,
- Evans J.C,
- Levy D
- Go R.T,
- Marwick T.H,
- MacIntyre W.J,
- et al.
- Go R.T,
- MacIntyre W.J,
- Cook S.A,
- et al.
- Cox D
- Lawless J.F
- Burnham K.P,
- Anderson D.R
- Berman D.S,
- Kiat H,
- Friedman J.D,
- et al.
- Smanio P.E,
- Watson D.D,
- Segalla D.L,
- Vinson E.L,
- Smith W.H,
- Beller G.A