Systemic inflammatory and metabolic indexes for aortic dissection: a machine learning analysis

Inflammation plays a crucial role in the development of aortic dissection. The systemic immune-inflammation index (SII) and systemic inflammation response index (SIRI) are widely used surrogate markers for systemic inflammation, while the neutrophil to high-density lipoprotein cholesterol ratio (NHR), platelet to high-density lipoprotein cholesterol ratio (PHR), lymphocyte to high-density lipoprotein cholesterol ratio (LHR), and monocyte to high-density lipoprotein cholesterol ratio (MHR) incorporate high-density lipoprotein, reflecting both inflammatory and metabolic status. However, there is currently a lack of studies systematically comparing the associations between different inflammatory and metabolic surrogate markers and the risk of AD. Therefore, this study aims to investigate the relationship between these six inflammatory-metabolic indexes and the risk of aortic dissection, as well as evaluate their discriminative value for this condition. This study aims to clarify the relationship between the inflammatory metabolic markers and aortic dissection. This study enrolled a total of 1,959 patients. First, Propensity score matching (PSM) was applied to minimize confounding bias. Subsequently, multivariable logistic regression models were used to evaluate the associations between inflammatory indexes and AD risk. Restricted cubic spline (RCS) analysis was performed to explore potential nonlinear dose–response relationships. The discriminative performance of inflammatory indexes was assessed using receiver operating characteristic (ROC) curve. Furthermore, six machine learning algorithms were implemented for feature importance analysis. After adjusting for potential confounding factors, we identified SII 1.002 (95% CI: 1.001–1.002), SIRI 1.984 (95% CI: 1.702–2.352), and NHR 1.692 (95% CI: 1.499–1.928) as independent risk factors for AD development, while LHR 0.211 (95% CI: 0.143–0.304) served as an independent protective factor. Dose–response analysis demonstrated a linear association between NHR and AD risk (P for Nonlinear = 0.02), whereas SII, SIRI and LHR showed nonlinear relationships (P for Nonlinear < 0.001). ROC curve analysis revealed that the four inflammatory indexes had AUC values ranging from 0.740 to 0.835 for distinguishing AD and its clinical subtypes. Among the six machine learning algorithms evaluated, the XGBoost model demonstrated the best performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.883. Among these six systemic Inflammatory and Metabolic indexes, the SII, SIRI, NHR, and LHR indexes demonstrate significant associations with the presence of aortic dissection. Further prospective validation in larger, multicenter cohorts is warranted to confirm their clinical utility before integration into diagnostic workflows.