• Unsupervised Machine Learning Study Identifies 4 Distinct Aortic Stenosis Phenotypes

    Unsupervised machine learning could help in capturing complex clinical presentations in severe aortic stenosis (AS) patients, researchers concluded, after using such a computer program to distinguish four distinct phenotypes of aortic stenosis that differ in mortality.

    The findings were published online Monday, ahead of the Oct. 11 issue of JACC: Cardiovascular Interventions, with authors led by Mark Lachmann, MD, and Elena Rippen, from the Technical University of Munich.

    Patients with severe AS are traditionally categorized into subgroups according to the presence of symptoms and aortic valve gradient (AVG), flow, and left ventricular ejection fraction (LVEF), the researchers noted.

    However, this “hypothesis-driven and divisive clustering” incorporates only a “limited set of characteristics,” they lamented, adding that physicians in real-world settings are commonly faced with disparities between AS-induced hemodynamic burden and extravalvular damage to the heart and pulmonary circulation, which could indicate multiple morbidities besides AS, including atrial fibrillation, coronary artery disease and chronic obstructive pulmonary disease.

    They hypothesized that, because an unsupervised machine learning technology based on agglomerative clustering works on a “theoretically endless set of variables and without the constraint of any a priori assumption,” it could improve identification of patients and aid categorization.

    “Unsupervised clustering approaches have proved extremely successful in analyzing high throughput experimental data, for example, in the field of genomics or cancer biology,” they said, adding that: “Adopting machine learning to cardiovascular medicine could aid in advancing the field of personalized medicine, especially when it concerns sub phenotyping of heterogeneous diseases.”

    The current study’s aim was, therefore, to use machine learning in the form of unsupervised agglomerative clustering of echocardiographic and hemodynamic parameters from right heart catheterization (RHC) from 366 consecutively enrolled patients undergoing transcatheter aortic valve replacement (TAVR) for severe AS at two centers in Munich between January 2014 and December 2020.

    The researchers selected 12 variables from echocardiography and RHC for clustering, covering all stages of potential disease progression from AVG over left heart function and pulmonary circulation to right heart structural and functional parameters.

    Mean age of the study population was 79.8 ± 6.8 years, the mean aortic valve area was 0.78 ± 0.21 cm2, and median survival was 6.3 years, with 50% of deaths occurring within 1.97 years after TAVR. All echocardiographic studies were performed by experienced institutional cardiologists during clinical routine using a commercially available echocardiographic system, the researchers noted.

    The cluster analyses revealed four distinct patient phenotypes, the researchers noted;

    • Cluster 1: 164 patients (44.8%) serving as a reference, presented with regular cardiac function and without pulmonary hypertension (PH).
    • Cluster 2: 66 patients (18.0%) patients with postcapillary PH and preserved left and right ventricular structure and function.
    • Cluster 3: 45 patients (12.3%) with left and right heart dysfunction together with combined pre- and postcapillary PH.
    • Cluster 4: 91 patients (24.9%) with postcapillary PH with dilatation of all heart chambers and a high prevalence of mitral and tricuspid regurgitation.

    For patients in cluster 1, estimated 2-year survival was 90.6% (95% confidence interval [CI]: 85.8% to 95.6%) – similar levels to the 85.8% 2-year survival rate in cluster 2 (95% CI: 76.9% to 95.6%).

    Patients in cluster 4 with dilatation of all heart chambers and a high prevalence of mitral and tricuspid regurgitation (12.5% and 14.8%, respectively) died more often (2-year survival 74.9% [95% CI: 65.9% to 85.2%]; hazard ratio [HR] for 2-year mortality: 2.8 [95% CI: 1.4-5.5]).

    Those in cluster 3 displayed the most extensive disease characteristics, and 2-year survival was accordingly reduced (77.3% [95% CI: 65.2% to 91.6%]; HR for 2-year mortality: 2.6 [95% CI: 1.1-6.2]), the researchers added.

    “By distinguishing phenotypes without the constraint of any a priori assumption (ie, ignoring a hypothesized sequential order of isolated, AS-induced impairments of the heart and pulmonary circulation), this study sheds light from a new perspective on structural and hemodynamic impairments in patients with severe AS and finally demonstrates that structural alterations in left and right heart morphology constitute an equally sensitive indicator of poor prognosis compared with high-grade PH,” they concluded.

    “Unsupervised agglomerative clustering deserves particular attention, as the extent of cardiac and pulmonary circulatory impairments does not necessarily follow a sequential order from AS-induced left heart dysfunction, through PH, to ultimately right heart failure but is also influenced by comorbidities and ageing in general.

    “As predefined cluster assignments could additionally be recapitulated by an [artificial neural network], this study advocates for the use of machine learning technology for individual cluster assignment and hence refined risk stratification prior to TAVR for patients with severe AS in future clinical practice.”

    In an accompanying editorial, Nico Bruining, PhD, and Peter P.T. de Jaegere, MD, PhD, from Erasmus MC, Rotterdam, Netherlands, said the study “clarifies that disease phenotyping by AI reveals different pathophysiological disease states regardless of disease severity defined by the current guideline criteria for stenosis severity.

    “However, it may be clear that AI is an advanced clinical decision support tool that requires human intelligence and supervision for the ultimate interpretation of the findings. Keeping this premise in mind, the future of AI looks bright.”

    Sources

    Lachmann M, Rippen E, Schuster T, et al. Subphenotyping of Patients With Aortic Stenosis by Unsupervised Agglomerative Clustering of Echocardiographic and Hemodynamic Data. JACC Cardiovasc Interv 2021;14:2127-2140.

    Bruining N, de Jaegere PPT. Will Artificial Intelligence Deliver Precision Medicine for Patients With Aortic Stenosis? JACC Cardiovasc Interv 2021;14:2141-2143.

    Image Credit: ibreakstock – stock.adobe.com

This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Review our Privacy Policy for more details