Machine Learning-Assisted Feature Selection Identifies the Joint Association of Body Mass Index and Periaortic Adipose Tissue as a Risk Factor for Aortic Dissection: A Multicenter Retrospective Study
Journal:
medRxiv
Published Date:
May 1, 2026
Abstract
BACKGROUND: Aortic dissection (AD) is a life-threatening emergency with high mortality. Although elevated body mass index (BMI) is associated with both AD incidence and mortality, the underlying mechanisms remain unclear. Periaortic adipose tissue (PAAT) increases with BMI, and the PAAT of AD shows marked inflammatory infiltration, suggesting PAAT-driven inflammation may contribute to the development of AD. However, no direct evidence links BMI and PAAT to AD. To further elucidate the obesity-inflammation-AD relationship, we aim to quantify the contributions of BMI, PAAT, and their derived indices to the risk of AD. METHODS: This retrospective multicenter study (June-November 2025) quantified PAAT around the descending thoracic aorta with CT angiography (CTA). Logistic regression analyses were performed to identify AD risk factors. Based on the Boruta algorithm (a machine learning feature selection method) and ROC curve analysis, the variable importance for AD risk was assessed. The dose?response relationship between BMI?Volume-derived metric (BMV) and AD risk was further characterized by quartile stratification and restricted cubic spline (RCS). RESULTS: This study enrolled 376 consecutive participants. After adjusting for potential confounders, BMI, smoking, systolic blood pressure (SBP), diabetes mellitus (DM), TC/HDLC, ApoE, PAAT volume (Volume), PAAT fat attenuation index (FAI), and BMV were identified as independent predictors of AD. Volume was the strongest AD predictor with the highest Z-score. Compared with BMI [AUC 0.627, 95% confidence interval (CI): 0.569?0.687] and Volume (AUC 0.716, 95% CI: 0.662?0.772), BMV showed better discriminatory performance (AUC 0.726, 95% CI: 0.673?0.778). RCS showed an approximately linear positive association between BMV and AD risk (P-overall < 0.001, P-non-linear = 0.09). CONCLUSIONS: In this retrospective multicenter study, BMV, a composite measure integrating systemic and periaortic adipose tissue factor, showed a positive association with AD risk, and improved predictive performance beyond BMI, indicating incremental predictive value, pending external validation.
