Lipoprotein(a) and atherosclerosis risk across estimated glucose disposal rate strata: evaluating synergistic...

High lipoprotein(a) [Lp(a)] levels and insulin resistance (IR) are both established risk factors for atherosclerosis. Nevertheless, the interrelationship between these two factors and their combined effects on atherosclerosis remain poorly understood. This research aimed to explore the distribution characteristics of Lp(a) across individuals at varying IR degrees and to evaluate the combined ability of Lp(a) levels and the IR surrogate marker, estimated glucose disposal rate (eGDR), in predicting atherosclerosis.

This retrospective study involved data from 10,753 participants who received health examinations at the Third Xiangya Hospital (2017-2025). Carotid ultrasound was employed to diagnose atherosclerotic plaques. Logistic regression and restricted cubic spline (RCS) analysis were used across eGDR quartiles to examine the relationship between Lp(a) and plaque risk. The incremental benefits of incorporating both Lp(a) and eGDR in predicting atherosclerosis were further assessed via machine learning-based analyses.

In this study, the median Lp(a) concentration of the participants was 12.4 mg/dL, with 14.4% exhibiting values greater than 30 mg/dL. RCS analysis identified a significant nonlinear (U-shaped) relationship of Lp(a) with eGDR, whereas peak Lp(a) levels were observed within the uppermost eGDR quartile (Q4). Lp(a) levels demonstrated an independent positive link to atherosclerotic plaque risk; notably, this association remained significant in both the lowest and highest eGDR quartiles. Furthermore, machine learning analyses indicated that the incorporation of Lp(a) and eGDR conferred a modest gain in the model's discriminative capacity when forecasting atherosclerotic plaque, with the AUROC value increasing slightly from 0.7810 to 0.7820.

Lp(a) levels exhibited a positive linear association with atherosclerotic plaques in the overall population, which varied across eGDR strata. Furthermore, the interaction between Lp(a) and eGDR was significantly relevant for the prediction of this condition. The integration of Lp(a) levels and the eGDR into machine learning models modestly improved the identification of high-risk individuals. This strategy supports more precise clinical interventions, ultimately contributing to improved cardiovascular health outcomes.