There is no perfect study of stress testing. There are no RCTs of stress testing in emergency department patients, so we cannot say with certainty whether they help or hurt. The studies describing test characteristics are also imperfect. Partly, that is because people try to use stress tests to identify a variety of different outcomes. Mostly, however, if is because there are no emergency department studies in which all patients are assessed using the same gold standard, meaning the sensitivity and specificity will be inflated in all the studies discuss below because of partial verification bias and incorporation bias, among others.
- This post is part of a series exploring the practice of stress testing. The appraisal of these articles is based on the conclusions of the other posts, so it’s probably worthwhile reading the entire series.
Does exercise stress testing detect MI or death? (Can stress testing be used for risk stratification?)
When we discharge chest pain patients home from the emergency department, we are worried about missing MIs, with the potential consequence of the patient dying. We order stress tests because we don’t want to miss an MI. This data can get very confusing because so many other outcomes are studied, but it makes sense to start with the outcomes that we (and out patients) care about.
The quick summary of these papers is that stress testing does not pick out patients at risk of MI or death, has a high rate of false positives, and probably leads to unnecessary invasive procedures.
Meyer MC, Mooney RP, Sekera AK. A critical pathway for patients with acute chest pain and low risk for short-term adverse cardiac events: role of outpatient stress testing. Annals of emergency medicine. 2006; 47(5):427-35. PMID: 16631982
- A prospective observational trial of all patients who were scheduled for an outpatient stress test after emergency department evaluation (using a chest pain protocol). Patients had to be over the age of 40 and present to the emergency department with chest pain. After serial troponins, patients are risk stratified into low, intermediate and high risk based on cardiac risk factors. Only the low risk patients were managed as outpatients, and that is what is studied here.
- Exercise stress test was used unless there was a contraindication.
- The primary outcome was death or MI at 6 month follow-up.
- Of 7178 patients over 40 years of age evaluated for chest pain, only 979 (14%) were considered appropriate for outpatient stress testing. Of these 903 (92%) completed a stress test.
- There were no deaths in the follow up period.
- 3 people had MIs after hospital discharge (1 within 1 month). All three had negative stress tests. In other words, the stress test did not find any of the patients at risk for short term MI (sensitivity 0%). Another 5 patients with negative stress tests underwent revascularization within 6 months.
- Most exercise stress tests were normal. Of the 150 patients with an abnormal exercise stress test, 122 (81%) ultimately had confirmatory testing that demonstrated no coronary artery disease. In other words, 81% of the abnormal tests were confirmed to be false positives. 137 underwent a nuclear test after their exercise stress test. 10 patients had PCI performed, but these were patients with normal ECGs and troponins, and stents have been demonstrated not to help in this population. 7 patients had a CABG performed – I don’t have any way to assess the value of those operations.
- In total there were 17 revascularization procedures (of unclear necessity) by 6 months in the patients with positive stress tests and 5 among patients with negative stress tests. The stress test doesn’t seem to do a good job differentiating the “need for revascularization”, and probably causes this difference without demonstrable benefit for patients.
Bottom line: In this cohort, exercise stress test had a sensitivity of 0% for MI, a false positive rate over 80%, and probably resulted in unnecessary invasive procedures.B
Partial verification bias occurs when you don’t use the same gold standard test for everyone studied. These stress testing studies are unblinded observational studies, so are at high risk of bias. The most problematic bias is partial verification bias. Patients with negative stress tests are sent home with no further testing. Patients with positive stress tests are referred on for further testing, often angiogram. Asymptomatic stable coronary artery disease is common in all patients. In patients we negative stress tests, it will never be found. In patients with positive stress tests it will be found, stented, and treated like an important outcome. However, the stress test did not predict the need for revascularization. In fact, it caused the (potentially unnecessary) procedure to be done.
Scheuermeyer F, Innes G, Grafstein E, et al. Safety and Efficiency of a Chest Pain Diagnostic Algorithm With Selective Outpatient Stress Testing for Emergency Department Patients With Potential Ischemic Chest Pain. Annals of Emergency Medicine. 2012;59(4):256-264. PMID: 22221842
- A prospective observational study of emergency department chest pain patients. High risk patients (by history) were immediately consulted to cardiology. Among patients with normal ECGs, negative troponins at 2 and 6 hours, the clinician could decide if they were very low risk (in which case no further testing was done) or “at risk” (in which case an exercise stress test was booked within 48 hours).
- They include 1140 patients over a 8 month period. This was a relatively high risk cohort with a 10.8% rule in rate for ACS.
- No cases of ACS were missed with their protocol.
- 254 patients received a stress test. It was negative in 204 (76.7%). Of the 50 positive stress tests, 29 were determined to be false positives. All of the 21 “true positives” were unstable angina (none of these patients had ECG changes or positive troponins at any time. None received a final diagnosis of MI). I don’t see any mention of how many of these patients underwent (unnecessary) invasive testing.
- Their definition of unstable angina includes a positive stress test. Considering the inaccuracy of stress testing, this is obviously a bad definition. More importantly, that definition means that it is impossible to determine the accuracy of stress testing for predicting unstable angina. (We are stuck in some crazy circular reasoning. These patients have unstable angina because their stress test is positive. And the stress test is a true positive because they have unstable angina.)
- Not all low risk patients had stress tests, and they relied on clinical follow-up, so it is possible some events were missed.
Bottom line: In this cohort, there were no MIs or deaths in the stress testing group, so sensitivity cannot be calculated. Most stress tests were negative, and the false positive rate was between 58% and 100% (if you exclude the subjective and circular diagnosis of unstable angina).
Incorporation bias is occurs when the gold standard or outcome you use includes the results of the test you are studying. It is a kind of circular reasoning. For example, when the definition of unstable angina includes a positives stress test, the stress test didn’t predict the unstable angina, it simply defined the result. You cannot calculate an accurate sensitivity and specificity in this setting.
Manini AF, McAfee AT, Noble VE, Bohan JS. Prognostic value of the Duke treadmill score for emergency department patients with chest pain. The Journal of Emergency Medicine. 2010;39(2):135-143. PMID: 19062225
- A prospective observational trial looking at a convenience sample of emergency department patients presenting with chest pain or other symptoms suggestive of myocardial ischemia, with non-diagnostic ECGs and one negative troponin. (This is a higher risk group than we usually test, because they didn’t wait for the second troponin). All patients had an exercise stress test using the Duke Treadmill Score.
- Their primary outcome was a composite that combined death, MI, and revascularization.
- They included 196 patients. 20 were lost to follow-up.
- Of the 171 stress tests, 150 (88%) were normal, 16 (9%) were non diagnostic, and 5 (3%) were positive according to cardiology. According the the Duke Treadmill Score, 119 (70%) were low risk, 51 (30%) were intermediate risk, and 1 was high risk.
- Although they conclude that the Duke Treadmill Score is associated with adverse outcomes, a closer look at the data suggests otherwise. (Although, even the numbers they present aren’t great, with a sensitivity of 83%, a specificity of 72%, and a positive predictive value of less than 10%.)
- There were no deaths.
- There were 2 MIs at 30 day follow-up. The standard stress test missed both of these patients (sensitivity 0%). The Duke score caught 1 of 2 (sensitivity 50%). It is unclear from the way the data was presented if this patient would have been caught with a second troponin. However, for that 1 potential catch, there would have been 51 false positives.
- Because stress testing can directly lead to revascularization, you can’t use that as an outcome when calculating sensitivity and specificity. (Not to mention, revascularization is probably an unimportant outcome, or may even represent harm to the patient).
Bottom line: In this cohort, standard stress testing missed both of the patients with MIs in the follow-up period, but the Duke score caught one. For this 50% sensitivity, there were a large number of false positives and presumably unnecessary invasive procedures (although it is hard to tell because this study was done prior to the second troponin).
Amsterdam EA, Kirk JD, Diercks DB, Lewis WR, Turnipseed SD. Immediate exercise testing to evaluate low-risk patients presenting to the emergency department with chest pain. Journal of the American College of Cardiology. 2002; 40(2):251-6. PMID: 12106928
- A prospective observational study looking at a convenience sample (8am-8pm only) of patients who underwent immediate exercise testing if they presented to the ED with nontraumatic chest pain of suspected cardiac etiology but were considered to be at low clinical risk. In addition to a low risk history, this meant a non-specific ECG or an ECG unchanged from previous. Some patients had a troponin performed before the stress test, but some went straight to stress test with no labs.
- They included 1000 patients over a 4.5 year period.
- The stress test was negative in 64% of patients, positive in 13%, and indeterminate in 23%.
- There were no deaths identified in this group.
- There were 5 MIs identified in the 30 days follow up period. 4 of these occurred in patients with positive stress tests, but all 4 were identified on a troponin that was drawn before the stress test was done. The only MI that wasn’t diagnosed on day one occured in a patient with a negative stress test.
- Of the 125 patients with positive stress tests, 102 had further testing. Only 33 of these follow up tests were positive, so the vast majority of positive stress tests were false positives.
- 12 patients with positive stress test had a revascularization procedure. Excluding the 4 MIs diagnosed by troponin, the rest were in in patients with negative ECGs and troponins, and were therefore probably unnecessary.
- Of the negative stress tests, in addition to 1 missed MI, there was another missed cardiac event based on follow-up myocardial stress scintigraphy.
- Caveats: Follow up data at 30 days was only available for 85% of patients. Stress testing was done before multiple negative cardiac biomarkers, unlike usual ED practice.
Bottom line: In this cohort, stress testing missed the only patient who had an MI in follow up (sensitivity of 0% for what we care about), had a very poor specificity, and resulted in some revascularization procedures that were probably unnecessary.
Sanchis J, Bodí V, Núñez J, et al. Usefulness of early exercise testing and clinical risk score for prognostic evaluation in chest pain units without preexisting evidence of myocardial ischemia. The American journal of cardiology. 2006; 97(5):633-5. PMID: 16490427
- This is a prospective observational trial looking at 340 patients presenting to the emergency department with chest pain of possible coronary origin.
- To get into the study, patients couldn’t have ST elevation or equivalent, positive troponins (at arrival and 8 or 12 hours), or prior CAD. They also had to be able to exercise.
- Patients were followed for 1 year, with 97% complete follow up.
- 231 patients (68%) had negative stress tests, 54 (16%) were positive, and 55 (16%) were inconclusive.
- The primary outcome was a composite of death and MI. Patients with a positive stress test had a higher incidence of the primary outcome, although it wasn’t statistically significant (7 vs 2%, p=0.06). It also isn’t clear that this is clinically significant. What do I do with this information?
- Mortality at 1 year was 3.7% in the positive stress test group and 2.2% in the negative stress test group.
- An invasive study was recommended for all positive stress tests. 26 patients (48%) were revascularized.
- Despite every patient with a positive stress test getting angiography and possible PCI, their outcomes were still significantly worse. It is possible that their outcomes would have been even worse without this intervention, but given the inaccuracy of stress testing seen in other studies, and that lack of benefit of invasive management in this population (normal ECGs and negative troponins) that seems unlikely. Another possibility is that stress testing was actually causing worse outcomes through iatrogenic harms. We can’t tell from this data.
Bottom line: Stress testing here demonstrated a difference in mortality at one year (although not statistically significant), but the difference was so small it isn’t clear how we are going to act on it. (Are you comfortable with your negative stress test patients having a 2% mortality rate?) Even if we wanted to act on it, it isn’t clear what we should do. Invasive management doesn’t work in this patient group. (Fanning 2016; Stergiopoulos 2012) All these patients had invasive management, and they did worse.
Sandhu AT, Heidenreich PA, Bhattacharya J, Bundorf MK. Cardiovascular Testing and Clinical Outcomes in Emergency Department Patients With Chest Pain. JAMA internal medicine. 2017; 177(8):1175-1182. PMID: 28654959
- A retrospective cohort using an “instrumental variables approach”. The idea behind this methodology is to reduce confounders between groups in observational research. They look for a variable that explains why some people were exposed and others were not, but that should not be related to the outcomes of interest. In this case, we would expect that patients with chest pain will have a similar rate of ACS and bad outcomes whatever day of the week they have their symptoms. However, on the weekend, there is less accessibility to stress testing. Therefore, by comparing weekend chest pain patients to weekday chest pain patients, we hope to get a less confounded sense of the value of stress testing.
- They included emergency department chest pain patients aged 18-64, but excluded patients whose final ED diagnosis suggested ischemia.
- In total, they include 926,633 patients.
- They were correct in their assumption that weekday patients were more likely to get provocative testing (18% vs 12% at 2 days). Otherwise, weekday and weekend patients look very similar in terms of cardiac risk factors.
- Weekday patients were subjected to more angiography (5.5 more per 1000 patients). However, despite the increase in invasive testing in the weekday patients, rates of MI and revascularization on follow up were the same in both groups at 1 year.
- In an analysis adjusted for observable risk factors, the numbers look even worse, with 92 extra angiograms and 12 extra revascularizations per 1000 patients, but still no change in the 1 year MI rate.
Bottom line: This is observational data only, but it is the best controlled data we have. What it tells us it that if you show up during the week, you are more likely to get a stress test and (probably as a result of that) you are more likely to get an angiogram and invasive management. However, the invasive management was not associated with any benefit, because the 1 year outcomes were the same for both groups.
These are just a handful of the many studies that have looked at stress testing in emergency department patients. I don’t think it is necessary to go into the same depth on every paper, because they all pain the same picture.
Khare (2008) report a retrospective study that looked at all the patients who had a stress test (of any kind) performed in their chest pain observation unit. There were 1194 total patients, and 1085 (91%) had negative stress tests. 48 (4%) had indeterminate tests, of whom only 1 had coronary artery disease on angiogram. There were 62 patients (5%) with positive tests, but among the patients who had an angiogram most (64%) were negative. In other words, the stress test was low yield and had far more false positives than true positives. (And it isn’t clear that the true positives result in management changes that actually help patients). The costs for the patients with positive stress tests were 5 times higher than those with negative stress tests.
Hermann (2013) retrospectively looked at 4181 chest pain patients who underwent stress testing in a prospectively collected data base. 470 (11%) had positives tests. Of the patients with confirmatory invasive testing, 63 (51%) had obstructive disease, but less than half of these patients met guidelines for revascularization. The other 59 patients had no obstructive disease. There was no follow up for the negative tests, so we can’t comment on sensitivity.
Paoloni (2013) retrospectively looked at 657 emergency department chest pain patients who were booked for an outpatient exercise stress test (73% actually had the test). 73% of tests were negative, and 14% were positive. 41 patients ultimately underwent angiography, and only 4 had significant coronary artery disease. They don’t present any data that allows calculations of sensitivity or specificity, but once again stress testing leads to invasive testing, with a very low yield, and without any clear benefit to patients.
Cortalan (2014) looked at 459 patients who had noninvasive testing in a chest pain observation unit (396 stress test and 63 coronary CT). Noninvasive test results did not predict adverse outcomes. If you look at what they define as a low risk group (atypical chest pain, normal ECG, no diabetes, and no know CAD), 187 patients had noninvasive testing. The test was negative in 176, but there were 3 adverse outcomes (1 MI, 1 death, and 1 revascularization) in this group. Among the 11 with positive tests, there were no bad outcomes, and the one angiogram performed was negative. In other words, stress testing had a sensitivity of 0% and a specificity of 0%.
Napoli (2014) performed a prospective observational study looking at 3543 chest pain patients admitted to an observation unit. I didn’t include it as a detailed write up because there were no bad outcomes (MIs or deaths) in this entire cohort. About half of the patients had a stress test after negative biomarkers, and only 20/1754 (1%) were considered true positives (based only on angiogram anatomy). There were 9 false positives. In other words, in the real world, the yield of stress testing is incredibly low.
Aldous (2015) performed a planned secondary analysis of patients in the ASPECT and ADAPT trials. They look at 1483 chest pain patients with negative ED workups. 755 of these patients underwent stress testing. There were zero cardiac deaths or MIs in the 1 year follow up period, so sensitivity and specificity for these outcomes cannot be calculated for these important outcomes. There were 32 positive stress tests (4.3%), 25 of whom underwent angiography, and 19 were revascularized. However, considering these were patients with negative ECGs and troponins, these were likely unnecessary revascularizations.
Foy (2015) looked at a large insurance database covering 421,774 emergency department chest pain patients. There was no difference in the rate of missed MI between the patients who had noninvasive testing and those who did not (although the groups are not randomized, and confounders certainly exist).
Reinhart (2017) did a secondary analysis of the ROMICAT-II RCT. When they compared patients who received noninvasive testing to those who had a clinical evaluation only, noninvasive testing was associated with a longer length of stay, more downstream testing, more radiation, and greater cost, without any improvement in clinical outcomes. This is observational data, however, and therefore may be unrecognized confounders. (Reinhart 2017)
Exercise stress test summary
This is a bit of a mixed bag methodologically speaking, but I think some things are clear from this data. (Although this is the exercise stress test section, I should note that a number of these studies mixed in other stress test modalities). These studies make it clear that exercise stress testing is very low yield, which makes sense considering that the patients we send for stress testing are very low risk. The stress test doesn’t seem to catch the patients we care about, with almost all MIs occurring in patients with negative stress tests. Stress tests will lead to more invasive testing, and more revascularization. This increase probably represents partial verification bias rather than any predictive capacity of the test. Furthermore, as will be discussed more in Part 4, we know that revascularization doesn’t help in this population, so the increase in invasive management probably represents harm to our patients rather than benefit.
Does exercise stress testing detect coronary artery disease?
When stress tests are ordered in the emergency department, clinicians are generally concerned about short term outcomes. The above studies illustrate that stress testing does not predict important short term outcomes like death and MI, but it could still play another role. Although not usually the domain of the emergency physician, we might want to know if a patient has stable coronary artery disease to help guide long term management. Therefore, it is reasonable to ask whether exercise stress testing detects coronary artery disease.
Although there are a large number of studies that attempt to answer this question, most use different gold standards for patients depending on whether the stress test is positive or negative. This introduces partial verification bias, which artificially increases both sensitivity and specificity. The only study I am aware of in which every patient received both a stress test and an angiogram as the gold standard is:
Froelicher VF, Lehmann KG, Thomas R, et al. The electrocardiographic exercise test in a population with reduced workup bias: diagnostic performance, computerized interpretation, and multivariable prediction. Veterans Affairs Cooperative Study in Health Services #016 (QUEXTA) Study Group. Quantitative Exercise Testing and Angiography. Ann Intern Med. 1998;128:(12 Pt 1)965-74. PMID: 9625682
- 814 consecutive patients with angina (so exactly the kind of patient we would discharge from the ED) underwent both an exercise stress test and angiography.
- Using angiography as a gold standard, exercise stress testing had a 45% sensitivity and an 85% specificity.
Bottom line: Stress testing is very poor at identifying anatomical disease. These numbers are especially bad when you try to apply the test to a group with a very low pretest probability of disease. (Using a sensitivity of 45% and a specificity of 85%, the positive likelihood ratio of stress testing is 3, while the negative likelihood ratio is 0.65).
Therefore, stress testing is unreliable even if your target is coronary artery disease, and certainly should not be used routinely in a low risk population, as the chance of both false positives and false negatives is high.
What is the accuracy of other types of stress testing?
The stress testing literature is complicated by the fact that there are many different types of stress test. Without performing the same deep dive on every modality, I will briefly outline some numbers for comparison. Unfortunately, most of these studies suffer from the same problems discussed above (incorporation bias, partial verification bias, and a focus on non-patient oriented outcomes). The brief summary is that although some of the numbers are a little bit better than exercise stress tests, none of the stress test modalities is accurate enough to use in the very low risk population for whom we routinely order these tests.
Colon PJ, Guarisco JS, Murgo J, Cheirif J. Utility of stress echocardiography in the triage of patients with atypical chest pain from the emergency department. The American journal of cardiology. 1998; 82(10):1282-4, A10. PMID: 9832109
- This study is often quoted as illustrating a excellent sensitivity and specificity for stress echocardiography, probably because of their conclusion that “this study demonstrates that a normal stress echocardiography study predicts an excellent prognosis for cardiac event-free survival over the ensuing year…” However, the actual results pain a different picture.
- This is a prospective observational study looking at 108 emergency department chest pain patients who had completed 4 hours of observation, with negative ECGs and biomarkers.
- All patients had both an echocardiogram and ECG based stress test (70% exercise, 30% dobutamine).
- Over 1 year follow up, there were no deaths and no MIs, so again it is impossible to say what the test characteristics are for those important outcomes.
- They used clinical follow up only, meaning we can’t get an accurate representation of the sensitivity and specificity for coronary artery disease.
- Stress testing was positive in 10 patients (9%) based on ECG criteria, and 8 patients (7%) based on echo criteria. The two tests only agreed on 4 patients. (So there is a very poor inter-test reliability with stress testing).
- 4 patients underwent angiograms that were perfectly normal (false positives).
- There were 6 patients who had revascularizations, with very limited details. However, once again, it is likely that the stress test result caused the revascularization rather than predicting the need for it. All 6 of these patients had normal troponins and normal ECGs, so the need for revascularization is questionable.
Bottom line: This study tells us nothing about the accuracy of stress testing. There were no patient important outcomes (MI or death). Because the decision to perform invasive procedures was based on the stress test, it is also impossible to make any comment on its accuracy there.
Forgetting about patient important outcomes, what is the accuracy of stress echocardiogram for coronary artery disease?
Gurunathan S, Young G, Parsons G, et al. 132 Diagnostic Accuracy of Stress Echocardiography Compared with Invasive Coronary Angiography with Fractional Flow Reserve for The Diagnosis of Haemodynamically Significant Cad in Patients with Known or Suspected CAD Heart. 2016; 102(Suppl 6):A94-A95.
- Available only in abstract form.
- They looked at 184 patients who had a stress echo and an angiogram within 6 months of each other. (About half the stress tests were exercise based and half were dobutamine based).
- The stress test was positive in 59% of patients. (This is a much higher risk population that we see in the emergency department, or that has been included in other studies).
- However, only 21% of the angiograms were positive.
- They do their calculations “at the vessel level”, which isn’t explained so I can’t check their results. However, they report a sensitivity of 70% and a specificity of 77%.
Bottom line: In a study where every patient gets the same gold standard, the accuracy of stress test is poor, with sensitivity and specificity both less than 80%.
de Jong MC, Genders TS, van Geuns RJ, Moelker A, Hunink MG. Diagnostic performance of stress myocardial perfusion imaging for coronary artery disease: a systematic review and meta-analysis. European radiology. 2012; 22(9):1881-95. PMID: 22527375
- This is a systematic review and meta-analysis looking at all types of perfusion imaging.
- They were looking at how accurate the test was in predicting coronary artery disease (rather than the MI or death that we are looking for in the ED).
- They only included trials where all patients received the same gold standard (angiography) to avoid partial verification bias.
- There were 795 patients in 10 studies looking at stress echocardiography. The sensitivity was 87% (95% CI 81-91%) and specificity of 72% (95% CI 56-83%). This translates to a positive likelihood ratio of 3.08 and a negative likelihood ratio of 0.18. These are the best numbers I can find anywhere for stress testing.
Bottom line: Stress echo has a moderate accuracy for coronary artery disease. It is not a good enough test to rule in or rule out the disease, and it isn’t clear how the results of this test will be used to benefit patients.
Nuclear stress test
de Jong MC, Genders TS, van Geuns RJ, Moelker A, Hunink MG. Diagnostic performance of stress myocardial perfusion imaging for coronary artery disease: a systematic review and meta-analysis. European radiology. 2012; 22(9):1881-95. PMID: 22527375
- This is the same systematic review and meta-analysis discussed in the stress echo section. The focus is coronary artery disease, not MI or death.
- There were 1323 patients from 13 studies looking at nuclear stress test (SPECT). The sensitivity was 83% (95% CI 73-89%) and the specificity was 77% (95% CI 64-86%). The positive likelihood ratio is 3.56 and the negative likelihood ratio 0.22, although with reasonably wide confidence intervals.
Bottom line: Nuclear stress testing has a moderate accuracy for coronary artery disease. It is not a good enough test to rule in or rule out the disease, and it isn’t clear how the result of this test will be used to benefit patients.
There is also one RCT looking at nuclear stress testing in a non emergency department population, which is discussed in the next section.
We don’t have the same kind of studies looking at outcomes in emergency department patients with these stress test modalities (although both were used to varying degrees in the studies discussed above). Stress echocardiography and nuclear perfusion scanning are both marginally better than exercise ECG stress testing if you are looking for stable coronary artery disease. However, neither is accurate enough to rule in or rule out disease. More importantly, as is discussed in part 4, it isn’t clear that we can use this information to provide better care for our patients.
Randomized controlled trials: Does stress testing help improve patient outcomes?
The accuracy of stress testing is poor. Furthermore, as is discussed in the next post, because invasive interventions don’t help patients with stable coronary artery disease, it isn’t clear exactly what we should do with the results of a positive test. Therefore, it is very unlikely that stress testing could improve patient outcomes, and there is some data that supports that hypothesis.
Lim SH, Anantharaman V, Sundram F, et al. Stress myocardial perfusion imaging for the evaluation and triage of chest pain in the emergency department: A randomized controlled trial. J. Nucl. Cardiol.. 2013; 20(6):1002-1012.
- 1508 emergency department patients with chest pain, normal ECGs, and negative biomarkers are 0,3, and 6 hours. (Single center in Singapore).
- They were randomized to either a nuclear (Tc-99m tetrofosmin) myocardial perfusion scan or clinical assessment based on risk factors and emergency department ECG.
- Primary outcome: There was no difference in the rate cardiac events at 30 days (0.4% vs 0.8%) and 1 year (0.7% vs 1%). **Like we saw in part 1, risk is very low in patients with negative ED workups.**
- There were no cardiac deaths at 30 days. At 1 year, there were 3 cardiac deaths in the stress testing group as compared to 0 in the clinical assessment group (not statistically significant).
- Admission rate was lower in the stress testing group (10% vs 18%), but this requires you to have the ability to get the stress test in the ED. In most settings, this protocol would certainly result in much higher admission rates in the stress testing group, just to get the test.
- Unlike every other study, in this group the stress testing group actually had fewer angiograms performed. (22% vs 39%; p=0.03). This is probably explained by the incredibly high rate of angiography in the control group (keeping in mind that these were all patients with negative ED workups).
Frisoli TM, Nowak R, Evans KL, et al. Henry Ford HEART Score Randomized Trial. Circ Cardiovasc Qual Outcomes. 2017; 10(10).
- 105 patients with 2 negative troponins (at least 3 hours apart) and a modified HEART score less than or equal to 3 (could not have a positive troponin) who were referred to a chest pain observation unit.
- They were randomized to either immediate discharge or admission with stress testing.
- At 30 days, there were no deaths, MIs, or revascularizations in either group.
- There was only 1 abnormal stress test, and it resulted in medical management.
- Not surprisingly, those randomized to immediate discharge had a significantly shorter hospital stay (6 vs 26 hours; p<0.001).
- Hospital charges were also significantly lower ($3,000 vs $10,000; p<0.01).
Young LH, Wackers FJ, Chyun DA, et al. Cardiac outcomes after screening for asymptomatic coronary artery disease in patients with type 2 diabetes: the DIAD study: a randomized controlled trial. JAMA. 2009;301:(15)1547-55. PMID: 19366774 [free full text]
- A randomized controlled trial of 1123 outpatients with type 2 diabetes (as a marker of high cardiac risk).
- Patients were randomized to either stress radionuclide myocardial perfusion or no screening.
- Patients were followed for 5 years. There were 7 nonfatal MIs and 8 cardiac deaths (2.7%) in the nuclear stress test group and 10 nonfatal MIs and 7 cardiac deaths (3.0%) in the untested group.
- There were no changes in long term outcome whether or not patients were tested
- Caveats: These are not emergency department patients (but they look a lot like the kind of patient that we send for stress testing). Clinicians were not blinded, which could have led to other, unseen changes in management.
- Bottom line: Testing stable patients with a low to moderate risk of CAD with cardiac stress imaging does not prevent deaths or MI.
The DECREASE-II study asked a similar question in a higher risk patient group. (Poldermans 2006) They randomized 770 patients undergoing major vascular surgery to either have a stress test or not prior to surgery. Stress testing didn’t help. (There was no difference in 30 day rates of cardiac death or MI). However, stress testing did lead to more invasive procedures, which given the lack of overall benefit, should probably be counted as harms.
Although not directly related to stress testing, another RCT that gives us a little information is Mahler (2015). This is an RCT of the HEART pathway, where 282 patients 21 and older with symptoms suggestive of ACS were randomized to either the HEART pathway or usual care. There was a 12% absolute decrease in objective cardiac testing (primarily stress testing) in the HEART pathway group, with no difference in 30 day MACE. This tells us that we are certainly over using stress tests, as decreasing their use made no difference to patient outcomes. However, given everything we know about stress testing so far, my prediction is that we could continue decreasing stress testing all the way to 0% and it would still make no difference in patient outcomes.
There is not one simple answer to the question: how accurate are stress tests? There are different stress test modalities, and stress testing is used to look for different outcomes. From an emergency medicine perspective, where we are primarily concerned about catching patients with missed MI, the literature suggests that stress testing is basically useless, if not harmful. The sensitivity for MI and death was close to 0% in these studies, but despite not identifying patients at risk, there were a large number of false positives, with patients subjected to unnecessary invasive testing.
The numbers were somewhat better for the identification of stable coronary artery disease, although it isn’t clear to me that this is an important function for emergency physicians. The standard exercise stress test was still so bad (sensitivity 45% and specify 85%) that it probably shouldn’t be used. Stress echocardiography and nuclear stress testing were better, with sensitivities and specificities approaching 80%, although no modality was accurate enough to either rule in or rule out disease.
Finally, in the few RCTs we have, stress testing provided no benefit to patients, but did expose them to the risks of more invasive procedures.
Overall, the literature points a pretty poor picture when it comes to stress test accuracy. However, even if the test was fantastic, the more important question might be, what should we do with the results? The next post tackles that question by looking at the value of revascularization in patients who aren’t having a STEMI.
Aldous S, Richards AM, Cullen L, Pickering JW, Than M. The incremental value of stress testing in patients with acute chest pain beyond serial cardiac troponin testing. Emergency medicine journal : EMJ. 2015. PMID: 26511125
Cotarlan V, Ho D, Pineda J, Qureshi A, Shirani J. Impact of clinical predictors and routine coronary artery disease testing on outcome of patients admitted to chest pain decision unit. Clinical cardiology. 2014; 37(3):146-51. PMID: 24255007
Fanning JP, Nyong J, Scott IA, Aroney CN, Walters DL. Routine invasive strategies versus selective invasive strategies for unstable angina and non-ST elevation myocardial infarction in the stent era. The Cochrane database of systematic reviews. 2016.
Foy AJ, Liu G, Davidson WR, Sciamanna C, Leslie DL. Comparative effectiveness of diagnostic testing strategies in emergency department patients with chest pain: an analysis of downstream testing, interventions, and outcomes. JAMA internal medicine. 2015; 175(3):428-36. PMID: 25622287
Hermann LK, Newman DH, Pleasant WA, et al. Yield of routine provocative cardiac testing among patients in an emergency department-based chest pain unit. JAMA internal medicine. 2013; 173(12):1128-33. PMID: 23689690
Khare RK, Powell ES, Venkatesh AK, Courtney DM. Diagnostic uncertainty and costs associated with current emergency department evaluation of low risk chest pain. Critical pathways in cardiology. 2008; 7(3):191-6. PMID: 18791408
Mahler SA, Riley RF, Hiestand BC, et al. The HEART Pathway randomized trial: identifying emergency department patients with acute chest pain for early discharge. Circulation. Cardiovascular quality and outcomes. 2015; 8(2):195-203. [pubmed]
Napoli AM. The association between pretest probability of coronary artery disease and stress test utilization and outcomes in a chest pain observation unit. Academic emergency medicine : official journal of the Society for Academic Emergency Medicine. 2014; 21(4):401-7. PMID: 24730402
Poldermans D, Bax JJ, Schouten O, et al. Should major vascular surgery be delayed because of preoperative cardiac testing in intermediate-risk patients receiving beta-blocker therapy with tight heart rate control? Journal of the American College of Cardiology. 2006; 48(5):964-9. PMID: 16949487
Reinhardt SW, Lin CJ, Novak E, Brown DL. Noninvasive Cardiac Testing vs Clinical Evaluation Alone in Acute Chest Pain: A Secondary Analysis of the ROMICAT-II Randomized Clinical Trial. JAMA internal medicine. 2018; 178(2):212-219. PMID: 29138794
Stergiopoulos K, Brown DL. Initial coronary stent implantation with medical therapy vs medical therapy alone for stable coronary artery disease: meta-analysis of randomized controlled trials. Arch Intern Med. 2012;172:(4)312-9. PMID: 22371919
Morgenstern, J. Stress Tests Part 3: Stress test accuracy, First10EM, March 13, 2019. Available at:
6 thoughts on “Stress Tests Part 3: Stress test accuracy”
Great article – very thought provoking. I appreciate the details you have gone into with each study.
I am reviewing some articles on how effective stress testing is at risk stratification (ie actually predicting MI/death). Appreciate all of the articles you’ve mentioned here, but how do you address some of the other literature ( for example, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3366298/#R16; https://www.ncbi.nlm.nih.gov/pubmed/15472640) where the risk of MI/death is much higher in those patients w/ positive studies. The article does mention that a negative test doesn’t mean zero risk- still means we miss a significant portion of patients – BUT the patients with positive tests do seem to have much higher risks.
There are a couple problems with the article you link. A big concern is that it only looks at “cardiac death” which is a meaningless concept, and never reports all cause mortality – so I am not sure there is even a difference. (Considering the inaccuracy of the test for coronary artery disease, it would be surprising if it predicted hard outcomes. As the literature above notes, it definitely doesn’t predict hard outcomes in ED patients.)
The article doesn’t apply to the above discussion. I am looking at the use of stress test after a negative emergency department workup. From what I can tell, these are not ED chest pain patients. They are high risk patients, many with known coronary disease (table 1 – seems like 30-50% had prior MIs or revascularization procedures) in an outpatient setting – although they are a little unclear on that. In that population, this study does indicate that stress tests have a minimal predictive value: A positive test was associated with a 10% rate of the outcome and a negative test is associated with a 2% outcome. Because of the sketchy definition of the outcome, I am not sure that difference is real, but even if it is, it seems pretty clear the information is not usable clinically. From a patient standpoint – those risks are pretty similar. More importantly, what are you going to do with those results? All the patients in this trial need medical management. And seeing as there is pretty definitive evidence that invasive management doesn’t help these patients (who are not having an MI), there is nothing else to add. And I think we can see that in the RCT by Young above (2009) – they randomized a similar population to either stress test or not, and there was no change in patient outcomes.
We are short on RCTs of stress test, but we have many of coronary CT angiogram. The CCTA is clearly a more accurate test – but across many RCTs, it clearly doesn’t help patients. If a better test can’t help patients, how can the inferior stress test possibly help? (For a sommary of the CCTA literature, see this post:
So stress test is inaccurate. That doesn’t mean it doesn’t have some minimal predictive value, but the discrimination between the two groups in the trial you cite isn’t big enough to help clinically, and ultimately there is no change in management that results from the test, so there is still no role for stress testing.
thanks for the reply! not debating any of your other points (agree that we do not have interventions that work). just trying to understand the literature out there from cards and ours and why there is a difference between the findings.
i have another question to pick your brain about. one of the main reasons why stress testing is so low yield seems to be because we miss so few MIs/cardiac deaths. However, this is presumably in the low risk group. What about the intermediate risk group? At my institution, the low risk (by HEART score) are being discharged home. It’s the intermediates who are staying for stress testing. Presumably, the risk of MI/deaths are a little higher than 0.2% in this group and the prognostic utility of stress testing is a little bit higher as well? Have you see any literature for this group of folks?