# Plasma ceramide Cer24:0 and insulin resistance: associations with TyG and TG/HDL-C in a multicenter study of coronary artery disease cohorts

## TL;DR

This study shows that the ceramide Cer24:0 is strongly linked to insulin resistance in patients with coronary artery disease, suggesting it could improve risk assessment.

## Contribution

The study identifies Cer24:0 as a novel biomarker for insulin resistance in coronary artery disease patients using multiple analytical methods.

## Key findings

- Cer24:0 showed a direct conditional association with the TyG index in mixed graphical models.
- Higher Cer24:0 levels were linked to increased TyG scores in double machine learning analyses.
- A ceramide-augmented model achieved 77% accuracy in classifying clinical insulin resistance.

## Abstract

Insulin resistance (IR) is a key metabolic determinant of cardiovascular risk. Ceramides, a class of bioactive lipids, have been linked to IR; however, their clinical associations in patients with established coronary artery disease (CAD) are incompletely characterized.

In a prospective, multicenter observational cohort of adults with established CAD (n=987), we quantified plasma ceramide species (Cer16:0, Cer18:0, Cer24:0, Cer24:1) and surrogate IR indices (triglyceride–glucose index (TyG), metabolic score for insulin resistance (METS-IR), and triglyceride–to–high-density lipoprotein cholesterol ratio (TG/HDL-C)). Mixed graphical models (MGM) were used to estimate conditional associations within a multivariable network. Double machine learning (DML) with causal-forest estimators provided covariate-adjusted association estimates and probed robustness values (RV) to unmeasured confounding. Exposures were standardized per standard deviation. We also developed a ceramide-augmented model to classify clinical IR, prespecified as TyG ≥ 9, and quantified discrimination by the area under the receiver operating characteristic curve (ROC–AUC).

Ceramides correlated with IR indices. In MGM analyses, Cer24:0 showed a direct conditional association with TyG (partial r=0.23; 95% CI, 0.17–0.29), independent of other variables in the network. In DML analyses, per 1-unit increase in ln (Cer24:0) was associated with higher TyG (estimate, 0.459; 95% CI, 0.252–0.665; P = 0.001); These estimates demonstrated moderate RV to unmeasured confounding, with an RV (theta) of 0.223. A ceramide-augmented model classified clinical IR with an ROC–AUC of 0.770 (95% CI, 0.741–0.799).

Across complementary analytic frameworks, Cer24:0 consistently exhibited positive associations with lipid-centric IR metrics among adults with established CAD. Although observational, these findings suggest that circulating ceramide profiling—particularly Cer24:0—may refine metabolic risk stratification beyond conventional indices in cardiology practice.