Association between aortic valve sclerosis and re-infarction after first myocardial infarction points towards a common pathway: results of an observational study using topological data analysis

This editorial refers to ‘Non-stenotic fibro-calcific aortic valve as a predictor of myocardial infarction recurrence’, by V.A. Myasoedova et al. https://doi.org/10.1093/eurjpc/zwae062. Patients after their first myocardial infarction (MI) are more likely to suffer further cardiac events including death despite contemporary management including high-intensity statin/PCSK9 inhibitors, dual antiplatelet, and percutaneous coronary intervention.1 Risk stratification means were developed to predict in-hospital and 6–12-month mortality and subsequent cardiovascular events, such as MI, after discharge.2,3 The hallmarks of those risk scores are the common cardiovascular risk factors, raised cardiac serum and electrical biomarkers, kidney injuries, etc. Aortic valve stenosis of any grade has been shown to be a predictor of poor prognosis in large echocardiography registries.4 Previous studies suggested that non-stenotic aortic valve fibrocalcific remodelling [aortic valve sclerosis (AVSc)] was associated with worse long-term outcomes independent of the traditional cardiovascular risk factors.5–7 Myasoedova et al.8 in their current publication assessed the importance of AVSc in predicting re-infarction in patients after type 1 acute coronary syndrome. They have analysed data of 2036 prospective patients recruited over a 10-year period (between 2021 and 2019), mean age 66 ± 12 years with ST-elevation myocardial infarction (STEMI) (∼52%) and Non-ST-elevation myocardial infarction (NSTEMI) who received guideline-directed medical management including coronary revascularization (100%), DAPT (∼97%), high-intensity statin treatment (∼93%), angiotensine convertase enzyme inhibitor (ACE-I) or angiotensin receptor blocker (ARB) treatment (∼67%), and beta-blocker treatment (∼79%). All the patients had transthoracic echocardiography for non-stenotic aortic valve sclerosis following Gharacholou et al.’s9 criteria such as the following: irregular, non-uniform thickening of portions of the aortic valve leaflets or commissures, or both; thickened portions of the aortic valve with an appearance suggesting calcification (i.e. bright echoes); non-restricted or minimally restricted opening of the aortic cusps; and peak continuous-wave Doppler velocity across the valve < 2 m/s. The incidence of AVSc at the time of admission was 46%, and re-infarction occurred in 14%. Researchers have developed a model using ‘topological data analysis (TDA)’ to assist in predicting the recurrence of MI over a 5-year follow-up period in a subset of patients (n = 1070) termed the study cohort and tested its performance on 966 MI patients from the same institution (test cohort). (Details of TDA are provided in the supplement of the main paper8 and incorporated in this editorial below.) There was no difference between the study and test cohorts in terms of anthropometric variables, cardiovascular risk factors, prior cardiac history, lipid profile, kidney function, in-hospital complication, and treatment on discharge. The TDA has identified 11 subgroups (clusters): 2 clusters with the highest incident of re-infarction over the 5-year follow-up were grouped and termed as a high-risk group, and the 9 other clusters were grouped as a moderate-risk group, with the hazard ratio (HR) 3.8 (2.7–5.4) for the high-risk group compared with the moderate-risk group. The features of the high-risk group were male predominance, hypertension, dyslipidaemia, with a higher prevalence of AVSc, and higher levels of high-sensitive C-reactive protein and creatinine. The 5-year average risk of event rate in the study cohort was 18% for the moderate-risk group and 42% for the high-risk group. The risk of events was similar in the test cohort: 14 and 40% for the moderate- and high-risk groups, respectively. The variables that consistently predicted the year-by-year risk of re-infarction included the following: AVSc, male sex, age, estimated glomerular filtration rate (eGFR), haemoglobin, and dyslipidaemia for the study cohort and AVSc, male sex, STEMI, diabetes, haemoglobin, dyslipidaemia, and smoking for the test cohort. The subtle differences are difficult to elucidate between the cohorts. Nevertheless, after adjusting for all significant cofounders on a year-by-year basis, AVSc was independently associated with an increased risk of recurrent MI both in the study and test cohorts. Interestingly, the AVSc was an independent predictor from year 2 in the study and from year 1 in the test cohort. The researchers must be congratulated for this analysis, especially for using the contemporary modelling approach TDA (Figure 1 with permission). Justification of using TDA had been provided as a supplementary method to the main article.8 Since its outstanding educational value, it is incorporated in this editorial. Representative images of the 11 clusters (A) (the variables as nodes) (B) derived by topological data analysis (with permission by OUP8). ‘Traditional risk prediction models are usually derived by the implementation of supervised approaches (often additive logistic models), on a small number of features, generally pre-selected on the base of clinical experience. In addition, a supervised model trained on an imbalanced data set, in which a group (e.g. patients without events) is over-represented compared with the other one (e.g. patients with events), will suffer from overfitting, leading to low performances. Thus, we used an unsupervised methodology, namely TDA, to give a snapshot of the inferred clinical phenotypes, highlighting analogies and dissimilarities between clusters. Unlike supervised methods, unsupervised approaches excel in discovering intricate patterns within the data that might elude traditional supervised techniques. In a clinical context, where the relationships between variables can be complex and multidimensional, the flexibility of unsupervised models in capturing subtle patterns could be essential for accurate risk stratification and identification of sub-phenotypes (‘precision medicine’); indeed, through unsupervised techniques like clustering and dimensionality reduction, it is possible to reveal latent structures in clinical data, contributing to a more nuanced understanding of risk factors and more generally of disease pathogenesis. In addition, the reduced reliance on labelled data and the shortage of a priori assumptions, also, mitigate the potential biases present in data sets, ensuring that the predictive models are more objective and generalizable across diverse patient populations. However, it is worth remembering that the primary aim of an unsupervised approach is not to generate a prediction model but rather to identify groups of samples with homogeneous characteristics in sizable data sets without any a priori information about the outcome (recurrent MI in this case). Just once the groups (‘clusters’) are identified, it is possible to use this information to assess and stratify the risk, adding the outcome labels for each patient to the analysis. Indeed, in this study, all the TDA findings were obtained without any information about cardiovascular events. The results of the study are intriguing and undoubtedly hypothesis generating that require further external validation before AVSc can be recommended in the clinical routine to predict the recurrence of MI (Figure 2). Further assessment of such population to investigate the progression of AVSc over the follow-up period and to correlate with the outcome can be an interesting proposition. In addition, exploring possible interventions that will have the potential to improve outcomes through modifying the progression or regression of AVSc would be of great relevance. Aortic valve sclerosis as an independent predictor of recurrence of myocardial infarction—summary figure.8 Inflammation and infection have been associated with the development and progression of atherosclerotic disease.10 Cardiac computed tomography (CT) angiography-based assessment of perivascular adipose tissue (PVAT) has been validated as a source of imaging biomarker of inflammation.11 In recent years, substantial evidence emerged recognizing radiomic features of the PVAT (fat attenuation index) as a sensor of vascular inflammation and a marker of cardiovascular events12,13 (Figure 3). Aortic valve sclerosis bares the similar pathomechanism as atherosclerosis via endothelial damage and inflammatory processes involving macrophage mast cells and T lymphocytes.14 Coronary CT angiography-based evaluation of perivascular and peri-aortic inflammation may be a helpful research tool that could facilitate risk stratification and the development of appropriate treatment to minimize future cardiovascular events such as re-infarction. Fat attenuation index mapping derived from the perivascular adipose tissue by computed tomography imaging.