STEMI vs NSTEMI
The ST Elevation (STE) Myocardial Infarction (STEMI) vs. Non-STEMI paradigm has been the universal definition and approach for AMI since the year 2000, when it formally replaced the previous Q-wave vs. non-Q-wave paradigm [3]. According to this paradigm, STE, as defined in the Fourth Universal Definition of MI, is the single marker which determines whether a patient gets immediate or delayed treatment for acute coronary occlusion. STE is inappropriately accepted as a surrogate for this emergency diagnosis, based on conventional wisdom and dogma.
Where does this dogma come from? The Fibrinolytic Trialists Therapy meta-analysis published in the Lancet in 1994, composed of all nine large (at least 1000 patients) randomized, placebo-controlled thrombolytic trials for AMI from the 1980s [4]. Only 4 of the 9 trials required STE (of varying amounts, without specifying the measurement method), and in the remaining 5 trials, no ECG findings were required. Streptokinase was the treatment in 7 trials, and 38% of patients had greater than six hours of chest discomfort despite the fact that the efficacy of thrombolytics drops off sharply after just a few hours [4]. Without any subgroup analysis into ST Elevation or ST depression, thrombolytics reduced mortality at one month, with a number needed to treat of 56. Subgroup analysis found that thrombolysis in unspecified STE and bundle branch block (BBB) had mortality benefit, whereas unspecified ST depression (STD) showed a non-significant mortality harm. It also found no benefit of reperfusion therapy in patients with inferior ST elevation. The universal simplistic and thus erroneous conclusion from this data was that STE is the way to determine which patients need emergent reperfusion.
But this is only true if ECG interpretation is crude, treatment is late, and streptokinase is the treatment. What about acute Q waves, terminal QRS distortion, subtle ST elevation, isolated ST depression, or hyperacute T-waves? What about QRST morphology, or QRS/ST-T proportionality? In LBBB and ventricular paced rhythm, which of the cases have OMI and which do not? How about all the pseudo- STEMI patterns which mimic STEMI? What about mortality using treatment other than streptokinase, such as percutaneous coronary intervention (PCI), or about accounting for the timing of treatment?
None of these were considered, and there have been no further placebo-controlled randomized reperfusion trials of any kind for AMI, much less for other ECG findings of AMI or for PCI vs. placebo or for immediate vs. next day PCI for either OMI or for STEMI. Instead, decades later, we are still basing the reperfusion decision almost entirely on this dichotomous finding of STE or not. The 4th universal definition of MI defines STEMI as STE in 2 contiguous leads with at least 1 mm measured at the J point, except for V2-V3 which vary by age/sex (1.5 in women, 2 in men >40 and 2.5 in men <40) [5]. This is based on 1220 patients, 248 of whom were diagnosed with AMI by elevation of Creatine Kinase-MB fraction, from a 1980's cohort; it is not based on angiographic outcomes or any other proof of OMI [6].
STEMI criteria are promoted as a simple and effective way to teach and identify acute coronary occlusion, but fail in both respects. Interrater reliability and sensitivity of physician identification of STEMI is poor, including among interventional cardiologists [7]. Automated interpretation is no better: a prospective validation of STEMI criteria found that computer interpretation of STEMI criteria was only 21% sensitive for OMI based on the first ECG and 30% based on serial ECGs, while blinded cardiologists were only 49% sensitive. In total, STEMI criteria had 37% false positive and 70% false negative rates [8].
Furthermore, the STEMI paradigm perpetuates these failures by denying the possibility of false negatives. We have called this the “no false negative paradox”. If the ECG meets STEMI criteria and there is OMI, this is a true positive; and if there is no OMI, it is a false positive. If the ECG does not meet STEMI criteria and there is no OMI, then it is a true negative. However, if the ECG does not meet STEMI criteria and there is OMI, i.e. STEMI(−)OMI, it is not considered a false negative because it is by definition a “NonSTEMI.” No matter how deadly the clinical outcome, no matter how much other clinical evidence of OMI were present other than STE, no matter how much benefit the patient may have received by earlier diagnosis and reperfusion, a patient with NSTEMI is treated as though delayed management is almost always justified because the paradigm excludes the possibility of a false negative. While NSTEMI guidelines recommend immediate invasive strategy for patients with hemodynamic or electrical instability or recurrent/refractory chest pain despite medical management [9], a recent study found this was only followed in 6.4% of very high risk NSTEMI patients [10].
This problem has deadly results. A meta-analysis of >40,000 NSTEMI patients with their first AMI found that, on next day angiogram, 25% had a persistent total 100% occlusion, without collateral circulation, and a significantly greater mortality rate compared to those with an open artery, and in spite of the fact that those with occlusion were on average 15 years younger [11].
The failure of STEMI criteria has led to attempts to broaden the paradigm to include “STEMI equivalents”, “semi-STEMIs” or “subtle STEMIs.” Yet these terms have no universal meaning, are never mentioned in the ACC/AHA guidelines, and in actual practice are ignored by cardiologists. These terms also perpetuate the cognitive bias inherent in the term “STEMI,” with its focus on ST Elevation. The STEMI paradigm leads providers to believe that STE is a valid surrogate marker of OMI, to believe that isolated ST segment amplitudes out of QRS context are all that matter, and that other ECG findings of OMI are irrelevant, when in fact the entire QRST, including all of the relative amplitudes and durations between the QRS and the ST-T are critical. The STEMI/NSTEMI false dichotomy prevents us from learning that all ECG findings are inherently proportional, which explains why subtle STE not meeting STEMI criteria can be diagnostic, while STE of greater amplitude than STEMI criteria can be present without ischemia. Just as disturbing, “STEMI” makes providers believe that OMI, as a disease process, is almost entirely defined and managed by the electrocardiogram; in fact, the ECG may be entirely nondiagnostic in the presence of OMI, and thus other modalities may be necessary for the diagnosis.
The STEMI paradigm prevents improvement in understanding of OMI, limits research on the ECG in AMI and research into ECG algorithms, and stifles evolution of the AMI paradigm. To our knowledge, no study of automated ECG algorithms has used an OMI study group; rather, studies of the accuracy of automated ECG interpretation tools invariably start by looking at a cohort of STEMI patients as the study group; thus, they begin with a group of true positives without any false negatives.
For all these reasons, the term “STEMI” is an important cause of the limitations of our current paradigm, and we must reframe the paradigm if we are to understand OMI and maximize the benefit of reperfusion for our patients. In the STEMI paradigm, the test (STE) is the definition of the pathology (OMI), so reperfusion therapy is based on presence or absence of STE on the ECG rather than presence or absence of underlying OMI. Instead we propose a new paradigm based on that actual underlying pathology: Occlusion MI (OMI) [12].
How can we identify OMI? Contrary to the STEMI paradigm, OMI is not defined by the ECG: refractory ischemia is an indication for urgent angiography even with a normal ECG, a recommendation made by current NSTEMI guidelines [9] but rarely followed [10]. ECGs are very insensitive for NOMI, but they don't need to be sensitive because urgent reperfusion is not necessary. We can therefore wait for a delayed diagnosis by use of troponin and delayed angiography/PCI for NOMI, because there is no ongoing ischemia. The exception is an occluded artery which has reperfused and has very unstable thrombus putting the artery at high risk of re-occlusion. Fortunately, ECGs, if read by true experts, are very sensitive for OMI and for reperfused occlusions, both of which identify patients requiring urgent angiography regardless of the initial troponin level [16,17]. Similarly, while most patients with unstable angina can wait for delayed angiography, ECG signs of reperfusion can identify those with an occlusive thrombus that rapidly spontaneously reperfused before infarction but is at risk for reocclusion.
Because the ECG is an instantaneous indicator of acute coronary occlusion, it is the most rapid means of identifying OMI. Other modalities such as point of care ultrasound can complement subtle ECG findings [18]. However, we cannot rely on advances in high sensitivity troponin assays to rapidly identify STEMI(−)OMI. No matter how sensitive or rapid, troponin is a delayed marker of AMI that reflects damage from the precious previous hours. An initial negative troponin on a patient with acute chest pain cannot rule out OMI, and an elevated troponin cannot determine if the injury is due to occlusion [19].
How do we know that STEMI(−)OMI can be identified by expert interpretation? In the DIFOCCULT study, cardiologists trained in the OMI paradigm and blinded to the outcome identified 28% of “NSTEMI” as OMI, and those patients had mortality similar to STEMI patients and much higher than NSTEMI whose ECGs did not show occlusion [16]. In our study, emergency physicians trained in the OMI paradigm (one senior, and one junior, faculty) could identify OMI with twice the sensitivity as STEMI criteria without a loss of specificity (see Table 1) [17]. Even among ECGs identified by STEMI criteria, the experts identified OMI on ECGs recorded a mean of 3.0 h and median of 1.5 h earlier. If implemented in the real world, this would result in an additional salvage of myocardium, with attendant reduction in mortality and morbidity, but the current paradigm constrains such implementation. Seven critical findings were identified which helped to identify OMI: hyperacute T waves, pathological Q waves along with subtle STE, terminal QRS distortion, reciprocal STD and/or reciprocal T wave inversion, subtle STE not meeting criteria but with other features, any amount of primary STD maximal in V1–4, and any amount of STE in inferior leads with any STD/T wave inversion in aVL. See Table 2.
In other words, identifying OMI on ECG requires going far beyond the exclusive focus on ST segment millimeter criteria. First of all, the ECG needs to be interpreted in totality—including heart rate, the entire QRS complex (including acute Q wave, loss of R wave, loss of S wave), the ST segment (including subtle STE, or STD reciprocal to STE including posterior or subtle inferior or lateral), the T wave (including hyperacute T-waves or reperfusion T-wave inversion), and QT prolongation (a subtle but early sign of occlusion), and dynamic change. Hyperacute T-waves are identified by their “bulk” (total area under the curve), not their amplitude, and that bulk is in proportion to the QRS. Many OMI have multiple subtle abnormalities (note the percentages in Table 2 add up to far >100%) that in their totality add up to a diagnostic ECG in the eyes of an experienced interpreter, but are dismissed as “non-specific” by the current paradigm and those who follow it.
Secondly, ST/T changes need to be assessed in proportion to the QRS complex. This is crucial in differentiating OMI from other causes of STE, among them LAD OMI vs normal STE in leads V2-V4, anterior left ventricular aneurysmmorphology vs. LAD OMI, and LBBB/paced rhythm with or without OMI using the ST/S ratio in the Modified Sgarbossa criteria.
Neural network potential and pitfalls
ECGs offer a major opportunity for neural networks [23]. The CDC estimates there are 40 million ECGs recorded in the US per year, which is likely an underestimate. This includes at least 6.5 million US ED visits for chest pain alone, plus another 4 million clinic visits for chest pain, with many more additional ED visits for anginal equivalents such as dyspnea. If only 10% of patients with chest pain have acute AMI, half of whom have OMI, and half of these have STEMI(−)OMI, then 2.5% of patients with chest pain in the ED have STEMI(−)OMI – a “needle in a haystack.” Those with STEMI(−)OMI require immediate recognition and reperfusion, yet nearly all are missed by current automated interpretation and by the STEMI paradigm on which they are based. OMI is the deadly needle in the haystack, and neural networks could play an important role in finding them.
Most research on neural networks has been on rhythms, especially atrial fibrillation detection. This work is so critical because so much of atrial fibrillation is occult and ripe for screening with monitors that are intelligent. OMI is also “occult,” since only a tiny fraction of patients with chest discomfort or dyspnea have OMI, and thus unless its subtle features are instantly recognized by providers, it will be missed until it is too late. In the Emergency Department, rhythm problems on the ECG are relatively easy because the presence of a problem is obvious: the rhythm is fast, slow, or irregular. But when there is chest discomfort or dyspnea, most etiologies are either benign or can wait hours for a diagnosis. The subtle ECG findings of OMI are too difficult for widespread training of providers; only neural networks will be able to fill this void.
By examining ECGs in their totality and applying principles of proportionality, ECG experts can identify OMI with the same familiarity as individual faces. When we look at faces, we recognize them immediately, without measuring the eyes or nose. As with faces, strict measurements of ECGs (ST segments) are useless. Like facial recognition technology, deep neural networks offer the potential of making immediate recognition of OMI widely available and accelerating the paradigm shift.
There are a number of challenges in training and testing neural networks for OMI. We need to start with the right outcome measure for OMI. If neural networks are only designed based on the STEMI criteria they will continue to reinforce its failings. Studies using conventional algorithms [24] that are only based on STEMI databases and only designed to identify STEMI(+)OMI continue to ignore STEMI(−)OMI which is the subgroup with the greatest reperfusion delay.
As a corollary, the control groups must be appropriately chosen. Due to the vast number of Non-OMI ECGs, it is nearly impossible to study a consecutive population of chest pain. As discussed, only a tiny percentage of chest pain patients have STEMI(−) OMI, and thus there would need to be accurate coding of 50 cases without OMI for every case of OMI. Therefore case control studies are much more practical. Many studies of STEMI vs Not STEMI compare STEMI (+) OMI to all other classifications (see Fig. 1, Fig. 2), and thus the control group is contaminated by STEMI (−) OMI [25]. Fig. 2shows the appropriate inclusion of control groups.
Secondly, we need to use serial and post-reperfusion ECGs to identify the underlying pathophysiology. OMI is a dynamic process, including reperfusion and re-occlusion, and ECGs on ED arrival do not represent the state of the artery by the time of the angiogram. Ideal research methods would go beyond brief coronary balloon occlusion studies by recording 12‑lead ECGs at the moment of natural thrombotic occlusion, with continuous monitoring during the course of occlusion, with or without alternating spontaneous thrombolysis and thrombus propagation, but ideal is not always possible. Many studies of algorithm diagnosis of STEMI use only one ECG from each patient. These methodologies ignore the dynamism of the ECG (and the practice of obtaining serial ECGs), with morphologies in constant flux due to the thrombus propagating and lysing, with various degrees of reperfusion and reocclusion. See Fig. 3, Fig. 4.
-USE.png)
-USE.png)
-USE.png)
-USE.png)
-USE.png)
-USE.png)
-USE.png)
-USE.png)
