We are already on the third post of this series, and we still aren’t going to look at the data on stress testing. For a series about stress testing, that might sound ridiculous, but in order to understand the stress testing studies, I think it’s important to understand the outcomes used in these studies. Talking about cardiovascular outcomes might seem rather simplistic, and it is true that many readers might want to skip straight to the next post, but there is actually a lot of complexity hidden in some rather simple sounding terms.
Death is an important, patient oriented, objective outcome. It is about as good as we get. Of course, it is not the only important patient oriented outcome, but when we are talking about MI, changes in mortality will always be important. The only thing to say about mortality, when analyzing these studies, is that some studies have very long (multi-year) follow-up periods, in which case death will frequently be unrelated to the chest pain that prompted enrollment in the study. These deaths can be difficult to assess in isolation, which is why control groups are so important.
Cardiac specific mortality
I don’t think disease specific mortality is a useful outcome, but it will be used in many of these studies. Over long follow-up periods, many patients will die of causes unrelated to their chest pain, so researchers attempt to count only those deaths than might be cardiac related. This fails for 2 reasons. First, aside perhaps from autopsy (which is essentially never done), there is no valid mechanism for determining cause of death. The cause of death is generally determined either by consensus of the researchers or by whatever happens to be written on the death certificate. Neither of these techniques is accurate and both introduce significant bias into the research. The second reason that I dislike disease specific mortality is that it is frequently used in misleading ways. People were very excited about the decrease in death due to bleeding in the WOMAN trial. However, there was no change in mortality overall. That means that either TXA actually decreases death due to bleeding, but kills an equal number of patients through another mechanism, or we are not very accurate at recording the cause of death and the reported differences were fictitious. Either way, the patient still dies. It is a statistical game, not an important outcome.
Non-fatal myocardial infarction
Non-fatal MI is a somewhat nebulous concept. It certainly sounds like an important outcome, but it may, in fact, be the most misleading of the cardiac outcomes used in these studies. Why are people worried about having heart attacks? Mostly, it is because they are scared they will die. However, non-fatal MIs are by definition non-fatal. The thing to remember is that non fatal MI is actually a surrogate outcome. However, it is not a surrogate for death, which is the most feared outcome of myocardial ischemia. So what exactly do nonfatal MIs mean for our patients? It isn’t clear. Most of the events are simply brief rises in troponin that are probably meaningless. Some non-fatal MIs probably result in worse patient outcomes in terms of long term CHF or exercise tolerance, although that link is unclear, and rarely studied. The important thing to keep in mind is that, despite sounding very important and objective, “non-fatal MI” is a laboratory diagnosis without clear immediate implications for our patients. (That being said, in my analysis I still count this is an important outcome. I just try to keep in mind that it is a surrogate for other less well defined patient oriented outcomes).
Revascularization is another outcome that sounds incredibly important and objective until you start exploring the data. Revascularization refers either to percutaneous interventions like stents or to coronary bypass grafting. In the emergency department, when we rush people off for a stent because of an acute occlusional MI, it is an important life saving procedure. As a result, we sometimes fall into the trap of thinking of all stents as beneficial. As is explored in depth in part 4, outside of acute MI, stents don’t help. In these trials, revascularization and MI are separated, meaning that the patients who fall into the revascularization category don’t have an MI (otherwise they would have been recorded in the MI category). That means that the revascularization procedures were almost certainly unnecessary (and therefore maybe even harmful). Even more problematic, from a research standpoint, is that revascularization performed on people with normal troponins and normal ECGs (the non MI group) is a completely subjective decision. Patients with positive stress tests will often get an angiogram, whereas those with negative stress tests will not. However, we know that none of these patients needed a stent. Therefore, the stress test is not predicting the need for a stent (there was no need), but is actually directly causing the stent to be placed. (A combination of partial verification bias and incorporation bias in these studies.) This is the reason that my discussion of stress testing focuses on MI and death as outcomes, but ignores revascularization in those studies.
The problem with unstable angina is that it is a subjective diagnosis. There are debates about whether it even exists in the era of high sensitivity troponins. (Braunwald 2013) I think it probably still exists, as is evidenced by numerous case reports on Dr. Smith’s incredible ECG blog, but it is rare and definitely over-diagnosed. The problem with unstable angina is that the diagnosis is entirely subjective. By definition, there has to be a non-ischemic ECG and negative biomarkers. So how do we distinguish unstable angina from the many non-cardiac causes of chest pain? We guess. It is an educated guess, but I don’t think we are very good at it. This becomes a huge problem in the stress testing literature. In unblinded studies (all of them), a patient with a positive stress test is called unstable angina, but the same patient with a negative stress test is called non-cardiac chest pain. This is called incorporation bias. In fact, some studies go as far as to use the results of the stress test in the definition of unstable angina. (Scheuermeyer 2012) When the results of the stress test determine your diagnosis, that diagnosis clearly cannot be used to determine the accuracy of the stress test. In a properly blinded study, unstable angina might be an appropriate outcome to consider (although still nebulous). However, none of these studies were blinded, which I why I think you have to focus on the objective outcomes of death and MI in assessing the accuracy of stress testing.
Major adverse cardiac events (MACE)
MACE is a very common outcome used in these trials. It is usually a combination of death, MI, and revascularization (including PCI and CABG), although sometimes the definition also includes things like unstable angina, hospitalization, and repeat emergency department visits. It is a fundamentally flawed outcome, because it combines things that are really important (death, and to a lesser extent MI) with meaningless outcomes like revascularization. This becomes a big problem when interpreting a number of these trials, as frequently only the composite MACE is reported, without any information about its individual components.
- This post is part of a series exploring the practice of stress testing.
Braunwald E, Morrow DA. Unstable angina: is it time for a requiem? Circulation. 2013; 127(24):2452-7. [pubmed]
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
Justin Morgenstern. Stress Tests Part 2: Defining Cardiovascular Outcomes, First10EM, 2019. Available at: