In addition to providing information regarding the diagnosis of coronary artery disease (CAD), noninvasive tests also provide prognostic information useful in risk stratification and clinical management. Accurate risk stratification has become increasingly important to optimize patient outcomes and contain rapidly escalating medical care costs. Tests with high negative predictive values (NPVs) (low event rates after a negative test) are particularly useful because they identify low-risk persons who generally do not need additional tests and interventions. Both stress myocardial perfusion imaging (MPI) and stress echocardiography have prognostic value in patients suspected of having CAD, independent of the information provided by clinical factors and the exercise electrocardiogram (1- 12).

Moreover, exercise imaging tests for the diagnosis of CAD may have variable diagnostic accuracy in women compared with men. The accuracy of both exercise MPI and exercise echocardiography has been reported to be lower in women than in men (13).

We performed a systematic review and meta-analysis to determine the NPV of exercise MPI, including thallium, sestamibi, and tetrofosmin, and exercise echocardiography. Our secondary aim was to assess the NPVs of these imaging modalities in women and men.

A literature search was performed using the PubMed, Cochrane, and DARE databases to identify articles published between January 1990 and May 2005 as part of a larger review of the diagnostic accuracy of exercise MPI and echocardiography in women. The following search terms were used: thallium radioisotopes, radiopharmaceuticals, tomography emission-computed single-photon, technetium TC 99M sestamibi, organotechnetium compounds, SPECT, Cardiolite, Mibi, tetrofosmin, technetium Tc 99m 1,2-bis(bis(2-ethoxyethyl)phosphino)ethane; echocardio*, ultrasound, ultrasonography; exercise, exercise test, exercise tolerance, exercise*, exercising, “stress test”; and diagnosis, diagnoses, diagnostic, diagnosing, predictive values of test. The search was not restricted to English-language literature. Bibliographies of all articles obtained were reviewed to identify additional articles. The date limits were chosen because both exercise echocardiography and exercise MPI using single-photon emission computed tomography (SPECT) with thallium and sestamibi were in widespread use during this period.

Studies were included if they met the following criteria: 1) prospective cohort studies of subjects who underwent exercise SPECT MPI or exercise echocardiography for known or suspected CAD; 2) provided primary data on clinical outcomes of myocardial infarction or cardiac death; 3) follow-up time of 3 months or more; and 4) defined a normal stress MPI or echocardiography test as one without any fixed or reversible perfusion deficit (for MPI) or fixed or inducible wall motion abnormalities (for echocardiography). Studies were excluded if they met the following criteria: 1) noninvasive tests were performed exclusively in patients after myocardial infarction, percutaneous angioplasty, coronary bypass surgery, hospitalization for unstable coronary syndrome, or with documented CAD by angiography; 2) included pharmacologic imaging tests without presenting data separately for pharmacologic and exercise testing; 3) did not include primary outcome data or event rates that could be statistically combined in a meta-analysis; or 4) evaluated planar MPI instead of SPECT. When multiple studies from the same research group were identified, only the largest study was included to avoid potential duplication in patient cohorts.

An initial search identified 3,563 potentially eligible articles. Two investigators reviewed the titles and excluded those that clearly did not provide data on humans or did not address the research question. The abstracts of the remaining articles were reviewed independently by 2 physician investigators to determine eligibility. Disagreements were resolved by consensus.

Two investigators (L.M. and R.H.), blinded to the journal, author, year of publication, and institution, independently reviewed the full text of each potentially eligible study and abstracted data including characteristics of the study, participant characteristics, test characteristics, mean follow-up time, and percentage of the population lost to follow-up. Occurrence of primary outcomes (nonfatal myocardial infarction or cardiac death) and secondary outcomes (revascularization and admission for unstable angina) were recorded. Results, including number of events or event rates based on positive or negative tests, relative risks or hazard ratios for future events, incremental prognostic value based on Cox proportional hazards model, and variables used in multivariate analysis, were abstracted. Results stratified by gender were abstracted separately when provided.

A quality assessment was performed by the 2 blinded investigators based on presence of the following parameters: 1) complete follow-up for 90% or more of the baseline cohort; 2) outcome data were obtained by investigators blinded to the test results; and 3) whether hospital records and death certificates were used to corroborate outcomes. Studies were defined as good quality if they fulfilled the criterion of 90% or greater follow-up and at least 1 of the other 2 criteria. Fair-quality studies fulfilled only 1 of the criteria and poor-quality studies fulfilled none of the criteria.

The primary analysis determined the summary estimate of the rate of myocardial infarction or cardiac death after a normal noninvasive test. Secondary analysis determined summary estimates of the rate of revascularization or unstable angina after a normal test. Pooled summary estimates and 95% confidence intervals for event rates after a negative noninvasive test were calculated from the primary data with each study result weighted by sample size using STATA software (version 8.0, STATA Corp., College Station, Texas). These analyses were repeated in subgroups of studies of MPI based on the criteria for defining a positive test, the radioisotope used, gender, and follow-up time. Findings were assessed for heterogeneity using a Q statistic, with a p value <0.10 considered statistically significant.

Estimated annual event rates were calculated with the caveat that the occurrence of events may not be linear. Estimated annualized event rates for each study were calculated as averages over the lengths of follow-up, and pooled summary annualized event rates were calculated by weighting study estimates by sample size.

A multivariate analysis was performed using STATA software. We used weighted multiple linear regression models to assess the dependence of the event rate after a normal test on characteristics of each study, including average age of participants, percent of participants with prior myocardial infarction, percent women, and length of follow-up. Analyses were weighted in proportion to the number of subjects in each study.

A sensitivity analysis based on study quality was performed by stratifying the studies into 2 groups, good quality and fair or poor quality, and calculating the summary estimates in each group. Publication bias was assessed by calculating the correlation coefficient, Kendall’s tau, for the association of sample size and event rate after a negative test (14).

The initial literature search yielded 3,563 study titles. Among these, 148 full-text articles evaluating the prognostic utility of exercise MPI or exercise echocardiography were reviewed, and 20 studies were found to be eligible for the systematic review. Of the 148 full-text articles reviewed, 17 did not address the research question (15- 31), 18 were excluded because they used pharmacologic rather than exercise stress (32- 49), 27 did not include primary data on event rates that could be combined in a meta-analysis (47,50- 75), 16 did not address a population with known or suspected CAD (76- 91), 5 evaluated only hospitalized patients or patients evaluated for chest pain syndromes in the emergency department (92- 96), 16 included only patients with angiographically proven CAD or positive tests (97- 112), 21 were potentially overlapping patient populations of other studies (53,113- 132), 5 evaluated planar MPI (133- 137), and 3 did not define a negative test as one without any abnormalities (138- 140).

Of the 17 studies included that evaluated MPI (141- 156), 5 used thallium (143,147,153- 154,157), 8 used sestamibi (141- 142,146,148- 150,152,156), 2 used both thallium and sestamibi (145,155), and 2 used tetrofosmin (144,151). Ten of the 17 MPI studies measured rates of unstable angina and revascularization in addition to primary events (141- 145,147- 148,152,154,156), and 3 stratified the data by gender (53,120,128). Of the 4 studies of exercise echocardiography (115,154,158- 159), 2 included data on rates of unstable angina and revascularization in addition to myocardial infarction and cardiac death (154,158) and 2 stratified data by gender (140,158).

The 17 exercise MPI studies included 8,008 subjects with a mean age of 54 years, of whom 34% were women (Table 1). A normal MPI test was defined as the absence of any fixed or reversible defect in each of the studies. The risk of myocardial infarction or cardiac death after a normal test was 1.2% with a NPV of 98.8% over a mean follow-up of 36 months, corresponding to an estimated annualized event rate of 0.45% per year if a presumption of linear event rates is made (Table 1). The test of heterogeneity demonstrated that the MPI study results are homogeneous (Q = 13.06, p = 0.7321).

Ten studies evaluated the predictive value of exercise MPI for cardiac revascularization or unstable angina; risk for these events after a normal test was 3.4% with an NPV of 96.6% over a mean follow-up of 36 months, corresponding to an annualized event rate of 1.25% per year (Tables 2, 3).

The NPV for the risk of myocardial infarction or cardiac death was approximately 97% over a mean follow-up of 45 months for thallium, approximately 99% after a normal sestamibi test over a mean follow-up of 32 months, over 99% when both isotopes were used over a mean follow-up of 23 months, and 99% after a normal tetrofosmin test over a mean follow-up of 43 months. The corresponding annualized event rates were approximately 0.7% for thallium, 0.3% for sestamibi, 0.5% when both isotopes were used, and 0.4% for tetrofosmin (Table 3).

Two studies focused on exercise MPI studies in women as part of larger databases included in the aforementioned summary estimates (120,128), and 1 study stratified by gender (53). The NPVs of myocardial infarction or cardiac death after a normal test were approximately 99% over an average follow-up of 32 months in women and 99% over 20 months in men. The annualized event rates were approximately 0.3% for women and 0.8% for men (Table 4).

The mean duration of follow-up was between 1 and 3 years in 9 studies (141- 143,145,148- 150,152,155), including 6,352 subjects; between 3 and 5 years in 6 studies (144,146,151,153- 154,157), including 1,050 subjects; and 6 to 8 years in 2 studies (147,156), including 606 subjects. The rate of myocardial infarction and cardiac death was 0.7% over a mean follow-up of 20 months in the 1- to 3-year subgroup, corresponding to an annualized event rate of 0.4%. The event rate was 2.7% over a mean follow-up of 46 months in the 3- to 5-year subgroup, corresponding to an annualized event rate of 0.7%. The 6- to 8-year follow-up subgroup had an event rate of 4% over a mean follow-up of 81 months, or a 0.6% annualized event rate.

The 4 eligible studies of the predictive value of exercise echocardiography included 3,021 subjects with a mean age of 56 and 46% women (Table 5). A normal test was defined as the absence of any wall motion abnormalities with stress or rest in all studies. The rate of myocardial infarction or cardiac death after a normal test was approximately 1.6% with an NPV of 98% over the mean follow-up of 33 months, corresponding to an estimated annualized event rate of 0.54% (Tables 3, 5). The test of heterogeneity demonstrated that the exercise echocardiography study results are homogeneous (Q = 0.46, p = 0.7952).

Two of the studies, including 380 subjects, included data on rates of unstable angina and revascularization. The NPV for these events was approximately 97% over a mean follow-up of 32 months, corresponding to an annualized event rate of 0.95% (Tables 3, 6).

Only 2 of the exercise echocardiography studies provided primary data stratified by gender, both of which found high NPVs for both men and women (140,158). The NPV was 98% in women over 38 months of follow-up, and 96% in men over 38 months of follow-up, corresponding to annualized event rates of 0.75% in women and 1.24% in men (Table 4).

The mean duration of follow-up was between 1 and 3 years in 2 studies (158- 159), including 263 subjects, and between 3 and 4 years in 2 studies (115,154), including 2,758 subjects. The rate of myocardial infarction and cardiac death was approximately 2% over a mean follow-up of 26 months in the 1- to 3-year subgroup, corresponding to an annualized event rate of 0.9%. The event rate was 1.5% over a mean follow-up of 40 months in the 3- to 4-year subgroup, corresponding to an annualized event rate of 0.5%.

Separate multivariate analyses were performed to determine the effect of cohort characteristics on the NPV of the MPI and echocardiography. For MPI, the rate of myocardial infarction and cardiac death after a normal test was about 0.2 percentage points higher for each 10 percentage point increase in proportion of subjects with a prior myocardial infarction. There was no significant difference in event rate after MPI related to percentage of women or mean age of the cohort. For stress echocardiography, the percentage of women in the study was the only variable that was significantly associated with the rates of myocardial infarction and cardiac death after a negative test. For each 10 percentage point increase in the proportion of women in the study, the event rate after a normal test was approximately 1 percentage point lower.

Nine studies were considered good quality (141- 142,149- 152,154- 156) and 10 fair or poor quality (143- 148,153,157- 159). For both exercise MPI and echocardiography, event rates after a normal test were similar when fair- and poor-quality studies were excluded.

There was no evidence of significant publication bias for either MPI (Kendall’s tau = −0.29, p = 0.06) or exercise echocardiography studies (Kendall’s tau = −0.33, p = 0.25).

Our systematic review and meta-analysis evaluating the prognostic utility of exercise MPI and exercise echocardiography demonstrated that the NPVs of both exercise imaging modalities are high. Although formal statistical comparison of the tests cannot be performed due to differing length of follow-up, possible nonlinearity of events, and inability to adjust fully for differences in factors that may affect the risk of cardiac events, the event rates with a normal test were low for both exercise MPI and echocardiography based on estimated annualized event rates. The estimated annualized rate of myocardial infarction or cardiac death was 0.45% per year after a negative MPI test, and 0.54% per year after a normal echocardiography test. These annualized event rates are both similar to a normal age-matched population, who carry a rate of <1% per year (160). Thus, both noninvasive imaging modalities accurately identify low-risk patients.

The development of myocardial ischemia begins with coronary stenoses, which lead initially to hypoperfusion, followed by wall motion abnormalities, a temporal sequence known as the ischemic cascade. The later development of wall motion abnormalities in this sequence suggests that stress MPI may be more sensitive in detecting CAD, and therefore more useful for prognosis than stress echocardiography. However, in our analysis, we found that normal echocardiography and MPI are both associated with low event rates similar to the general population.

The estimated annualized event rates for thallium, sestamibi, combined thallium/sestamibi, and tetrofosmin were all <1%. Therefore, each of the nuclear isotopes or combination of isotopes has a high NPV and is useful in identification of low-risk patients.

Some studies evaluating the diagnostic accuracy of stress electrocardiography, MPI, and echocardiography have found differing tests characteristics in women compared with men, and there is debate as to whether any particular modality may be more accurate in women (13,161). In our systematic review, both men and women with normal MPI tests had annualized event rates of <1%. Among the 2 exercise echocardiography studies that provided data stratified by gender, the annualized event rate after a normal test was <1% in women, and about 1% in men. It is important to note that we did not have high power to detect a difference in event rates between gender subgroups. However, the available data suggest that both noninvasive tests are useful to identify low-risk patients in both men and women.

Subgroup analyses based on follow-up time showed a slight increase in absolute event rate with increasing follow-up time for MPI, with an estimated annualized event rate <1% for each of the subgroups. Among the echocardiographic studies, longer duration of follow-up was not associated with increased event rates. However, the longest follow-up averaged only 3 to 4 years, and the subgroup with 1 to 3 years of follow-up included only 263 subjects compared with 2,758 in the group with longer follow-up, limiting the power to detect a true difference between the groups.

Although events such as myocardial infarction and cardiac death are almost universally evaluated in studies of prognosis and are considered least susceptible to bias, admissions for unstable angina or heart failure, revascularization, and anginal symptoms are associated with high costs and resource utilization, as well as substantial patient risk. In addition, as technology and medical treatments improve, rates of myocardial infarction and cardiac death are decreasing, and other outcomes, such as angina and revascularization, are becoming more common (162). In our study, the NPVs for revascularization and admissions for unstable angina were high for both exercise MPI and echocardiography. Estimated annualized event rates for secondary events were approximately 1% per year for both MPI and echocardiography.

Stress imaging tests can be used for risk stratification (1). Bayesian analysis indicates that stress tests are most useful for patients with intermediate pre-test probability of disease by moving them into a higher- or lower-risk group, thereby informing the choice of additional diagnostic tests, interventions, and medical management, which are costly and carry significant risks. By identifying a low-risk group of patients, defined by an annual event rate of <1%, additional interventions can be avoided in most cases (160). Our study indicates that both exercise MPI and exercise echocardiography are useful in identifying such low-risk patients over a spectrum of pre-test probabilities. Even in study cohorts or subsets of patients with relatively high pre-test probability, such as those with a higher percentage of prior myocardial infarction or positive exercise treadmill tests, the primary event rate with normal exercise MPI or echocardiography is relatively low (129,136,144,153). In multivariate analyses, markers of pre-test probability had modest effects on the event rates with a normal test, although there was a trend toward increased event rates in studies having higher percentages of prior myocardial infarction for MPI, but not for echocardiography.

The main limitation of our meta-analysis is that our summary estimates of event rates are unadjusted for some factors that may affect risk of cardiac events. Few of the studies included in the analysis provided adjusted relative risks of events. Multivariate models were used in many of the studies, but these models commonly assessed incremental value, which do not provide statistics that can be combined across studies.

Several studies of large cohorts undergoing exercise MPI or echocardiography were not included in this review due to lack of primary data that could be combined in meta-analysis or inclusion of pharmacologic tests. These studies had findings consistent with our meta-analysis (22,38,40,54,57,60- 61).

Assessments of the prognostic value of a test are susceptible to referral bias or the preferential referral to further testing or intervention in subjects with a positive test. In studies of prognosis, referral bias could result in a relatively higher event rate after a negative test. Despite potential referral bias, the NPVs of both imaging tests were high in our study.

We did not include analyses of positive predictive values in our systematic review. These values are particularly subject to bias due to differing patient risk factor profiles, and the effect of positive tests on subsequent revascularization and medical management. Although many studies included in our meta-analysis attempted to deal with referral bias by excluding subjects with early revascularization, this bias persists in unadjusted results.

Spectrum bias, or variation in test performance among persons at higher or lower risk of disease, may also affect the results of prognostic studies (163). Our systematic review includes studies of cohorts with varying pre-test risk of disease, as manifested by a broad range of percentage of subjects with prior myocardial infarction, advanced age, and male gender. Estimated annualized event rates were low, around 1% or less, for each of the studies despite varying pre-test risk of disease. Our multivariate analysis showed that event rates after a negative test increased minimally with increasing percentages of prior myocardial infarction for MPI, and did not increase for echocardiography. However, the studies included in our review did not include patient cohorts at higher risk of events, such as those who have recently had a myocardial infarction or undergone revascularization, and those undergoing pharmacologic imaging tests, and should, therefore, not be generalized to these populations. Although our study demonstrated the low annualized event rate generally associated with negative stress imaging tests, a negative test may not be associated with a low risk of events in certain populations, including diabetics (39,81- 82).

In addition to the prognostic utility of perfusion defects and wall motion abnormalities, other exercise imaging parameters have also been shown to carry prognostic value, although they were outside the scope of this meta-analysis. In particular, Duke treadmill score (11,164- 165), heart rate recovery (166- 167), ischemic left ventricular dilation (168), and change in end-systolic volume (169) have been shown to have independent prognostic value.

Our systematic review demonstrates that both exercise MPI, including thallium, sestamibi, and combination thallium/sestamibi, and exercise echocardiography have high NPVs for primary and secondary cardiac events. The event rates after normal tests with each modality are low, suggesting that the use of any of these noninvasive tests is appropriate, depending on experience and cost at particular institutions. In addition, the prognostic utility of both modalities is generally similar for both men and women. The NPVs of exercise MPI and echocardiography are useful in clinical practice to identify low-risk patients, thereby avoiding unnecessary tests and interventions.

The authors would like to thank statisticians Dan Moore and Eric Vittinghoff of the UCSF Women’s Health Clinical Research Center for providing statistical analysis.