Has the Fourth Universal Definition of Myocardial Infarction led to Better Diagnosis and Risk Stratification?

Héctor Bueno; Xavier Rossello; Alfredo Bardají


Eur Heart J. 2021;42(26):2562-2564. 

The 4th Universal Definition of Myocardial Infarction (4UDMI) was developed to improve the diagnosis of myocardial infarction (MI) with its different types, and to better differentiate these from myocardial injury.[1] Ideally, the 4UDMI should be helpful to guide the acute management of patients with troponin elevation, and to improve the prognosis and the long-term preventative strategies for the different types of MI. Although it has been shown that the 4UDMI improves the identification of patients at higher risk for cardiovascular events,[2] several challenges still remain. The 4UDMI requires evidence of myocardial ischaemia to differentiate MI from myocardial injury, but this may actually be a clinical judgement rather than true evidence. For instance, could chest pain at rest in patients with severe heart failure decompensation, acute anaemia, pulmonary embolism, or obstructive cardiomyopathy definitley not be a symptom of myocardial ischaemia? Could the presence of transient ST depression in patients with rapid atrial fibrillation, post-operative anaemia, sepsis, or electrolyte disturbances be ruled out with certainty as a sign of myocardial ischaemia? Furthermore, the 4UDMI uses a pathophysiological approach for classifying MI types, but there may be overlapping pathophysiologies in the same patient. To rule out acute coronary thrombosis (T1MI) from acute myocardial oxygen supply/demand imbalance in patients with chronic coronary artery disease (T2MI) may be difficult to do with clinical and non-invasive tools only. Would it be possible that in some of the aforementioned examples acute coronary thrombosis might be occurring simultaneously (or worse, mainly)? Is it possible to rule this out always without coronary angiography? If the answer is yes, would it be feasible, or wise, or even safe, to perform coronary angiography in all patients with a reasonable doubt for the presence of some degree of coronary thrombosis in the presence of other pathophysiological conditions causing myocardial ischaemia or myocardial stress? This is currently unknown but—most probably—the answer is no. Thus, there will always be a need for clinical judgement without definite evidence in patients with troponin elevation or suspicion of acute coronary syndrome (ACS) and, therefore, some level or uncertainty (read it sensitivity and specificity) to be measured. Finally, does the 4UDMI help in improving the short- and long-term management of patients according to its classification and related risk stratification (Figure 1)? In this issue of the European Heart Journal, a collaborative study by Hung et al.[3] evaluated the accuracy of the GRACE 2.0 score to predict 1-year mortality in patients with a diagnosis of T1MI and T2MI in Scotland and Sweden. This study is an opportunity to ascertain how well our current tools may be overcoming these challenges.

Figure 1.

Steps for the use of the Fourth Universal Definition of Myocardial Infarction and its classification in clinical practice and research. Figures are schematic and not intended to be proportional to fractions or event rates.

The diagnosis of T1MI may be considered as relatively straightforward, but the proportion of T1MI was quite different among the patients with suspected ACS seen in emergency departments from Scotland (55%) and from Sweden (28%). It is true that these differences may be explained by potential differences in the role of emergency departments within each healthcare system, differences in the patient clinical profiles, and in the pre-test probability of ACS between the studied populations,[4] or differences in the level of clinical suspicion in the professionals, but it is also true that other challenges have been described for the diagnosis of T1MI, such as the lack of evidence of a rise and fall of troponin levels in a number of T1MIs.[5] As mentioned before, the diagnosis of T2MI is actually more difficult. The proportion of T2MIs reported in the study also differed between countries: 12% in Scotland and 6% in Sweden. This could be explained by differences in patient profiles as well, but it is more likely that different clinical interpretations of the available data may have played a role. The presence of concomitant conditions, such as infection, heart failure, tachyarrhythmias, or hypertension may be attributed as a direct cause of myocardial injury or as contributors of myocardial ischaemia and lead to a T2MI diagnosis. In a previous report from the SWEDEHEART registry, infections were one of the most common diagnoses at discharge, and were considered the most frequent trigger for T2MI.[6] In contrast, other investigators may interpret troponin elevations in this situations as due to myocardial injury even in the presence of concomitant myocardial ischaemia.[7] There is controversy as to whether criteria for defining T2MI should be specific or wider. Strict criteria should help in reducing ambiguity and subjectivity in the diagnosis of T2MI, and in improving the reproducibility of the findings;[7] however, even using the same stringent criteria, the rate of T2MI in different populations may range between 26% and 34%.[8] Therefore, better guidance on how to interpret these interactions clinically may be needed in future versions of the definition of MI. The current study was done using administrative databases. However, while the concordance between clinical information and data obtained from administrative databases seems to be acceptable for T1MIs,[9] when International Classification of Diseases codes are used to classify all clinical conditions covered by the 4UDMI, the limitations are very important.[10]

The low in-hospital fatality rate observed in both countries (1.5% in Scotland and 2.2% in Sweden) is not in line with others reported in contemporary observational studies,[11,12] and may be explained in part by the use of network organization for patients presenting with ST-segment elevation MI, with higher short-term mortality, who bypass emergency departments (0% in Sweden and 18% in Scotland). The 1-year mortality rate was 50% higher in Scotland compared with Sweden (15% and 10%, respectively), although with an inverse direction compared with the in-hospital mortality, which was higher in Sweden. The study focused on the performance of the GRACE 2.0 score as a predictive tool for patients with T1MI and T2MI. The authors claim that they have demonstrated that the GRACE 2.0 score performs well in the prediction of all-cause mortality and future cardiovascular events in patients with T1MI, although with lower discrimination in patients with T2MI. However, the most striking finding of this study is the mismatch between observed and predicted 1-year event rates, two- to three-fold higher for the former. While the GRACE 2.0 score predicted 1-year mortality rates of 4.9% and 3.8% for T1MIs in Scotland and Sweden, these rates were 15% and 10%, respectively. For T2MIs, predicted death rates were 11.2% and 7.7% and actual rates were 23% in both countries.

This study overlooks one of the relevant dimensions needed to adequately assess the model performance of a risk score (calibration) by focusing too much on the c-statistic (discrimination). Discrimination, measured as the area under the receiver-operating curve—which, for models with binary endpoints such as mortality, is equivalent to the c-statistic— is not the probability that subjects are classified correctly but the probability that for any randomly selected pair of individuals, one with and one without the outcome, the model assigns a higher probability to the individual with the outcome, i.e. how well the predictive model can rank those who will and will not present the outcome of interest.[13] In contrast, calibration measures how well predicted probabilities agree with actual observed outcomes. Both discrimination and calibration are key aspects of predictive performance of prediction models, though they differ in the way in which they are formally assessed. Whereas discrimination can be summarized in a single number, easy to interpret for clinicians with little statistical background, calibration should ideally be assessed graphically by plotting the observed outcome frequencies against the mean predicted outcome probabilities or risks, within subgroups of participants that are ranked by increasing estimated probability. The assessment of this plot might require higher methodological skills and it is therefore often supplemented with formal statistical testing for goodness of fit (Hosmer and Lemeshow test) despite their limitations for sample sizes whuch are too small or too large.[14] Unfortunately, there is a trade-off between discrimination and calibration, and models cannot be perfect in both.[13] In this study, the GRACE 2.0 performs well in terms of discrimination for the risk of 1-year mortality in patients with T1MIs but it seems poorly calibrated, with the GRACE score systematically underestimating the risk beyond death rates over ~5%, particularly in the Swedish cohort. This is relevant because, from a clinical perspective, accurately calculating the absolute risk is important to set up long-term strategies for secondary prevention and follow-up. In a way, this risk score usually provides higher probabilities for the outcome in those who are going to experience it, but this probability does not necessarily match the expected outcome rate.

In conclusion, with the 4DMI we now have better tools for the diagnosis, prognostic assessment, and risk stratification of patients with troponin elevation, suspected ACS, and myocardial infarction. However, more clinical research is warranted to validate the use of the classification provided by the 4DMI in unselected patients, to assess its accuracy and reliability, and to evaluate its clinical implications. Given the discordance between the predicted and observed mortality rates at 1 year found in the study by Hung et al.,[3] we need to wonder whether the GRACE 2.0 may need re-calibration.