A Reclassification of Thoracic Aortic Aneurysm Using Machine Learning

Objective Traditional guidelines for aortic aneurysm replacement use a diameter threshold varying based on presumed etiology of the aneurysm: 1) Bicuspid aortic valve, 2) Sporadic (heritable and non-heritable), and 3) Syndromic aortic disease1. In this study, machine learning cluster analysis was compared to the AHA guidelines for patients with thoracic aortic aneurysms. Methods A machine learning clustering approach categorized patients based on a series of clinical variables for a 190-patient cohort. The clustering method was compared to a systematic grouping algorithm based on AHA guidelines. Clusters and group were compared using geometry, histopathology and biomechanical profile. Results Four relevant clusters were observed using this approach. Of these, three closely follow the AHA guidelines. The first cluster overrepresented patients with bicuspid aortic valves (P<0.01), matching the bicuspid group. The second cluster features older patients with more comorbidities and family history, corresponding to the sporadic group. The third cluster was younger with syndromic features (P<0.01) along with higher incidence of positive genetic testing (P<0.05). This is associated with the syndromic group. A fourth cluster, not well represented in the established guidelines featured patients with higher incidence of severe aortic root dilation, aortic insufficiency and negative genetics (P<0.01) with histological and biomechanical findings similar to the BAV-like and syndromic-like clusters. Conclusions Machine learning cluster analysis has identified a fourth aortopathy clusters distinct from the AHA group classifications. This fourth cluster has similar histologic and biomechanical findings as the BAV and syndromic group, suggesting an underlying genetic cause and prompting re-evaluation of our traditional classification.