Abstract Body: Introduction: Current screening criteria for abdominal aortic aneurysm (AAA) are based on clinical factors, such as age and smoking history, but do not include biological factors that may better reflect disease pathogenesis. As the overall prevalence of AAA declines, there is a clinical need for new approaches to AAA screening that increase yield through improved identification of high-risk individuals.

Hypothesis: We hypothesized that a proteomic risk score (ProRS) incorporating plasma protein abundance data could improve prediction of AAA over clinical factors alone.

Methods: We performed a cross-sectional analysis of nearly 37,000 participants in the UK Biobank Pharma Proteomics Project with plasma protein abundance data for 274 cardiometabolic proteins. A sparse ProRS was developed using regularized regression, with the most well-established clinical risk factors (age, sex, and pack-years of smoking) included as unpenalized covariates. Additional clinical factors, including diagnosis of hypertension (HTN) and hyperlipidemia (HLD), as well as the protein abundance data, were penalized. The regularization parameter was selected to balance both performance and sparsity of the final model.

Results: The generated sparse ProRS contained well-established clinical risk factors as well as a single plasma protein – matrix metalloproteinase 12 (MMP12). Overall performance and discriminatory utility of this model was higher than an identical model without MMP12 (difference in Brier score 2.1 × 10^-4, 95% credible interval 2.0-2.3 × 10^-4; difference in AUROC 0.021, 95% credible interval 0.020 - 0.022). Within the cohort, current AAA screening recommendations applied to 4.6% of the population and captured 30% of cases, whereas screening 4.6% of the population at highest risk by ProRS captured 52% of cases. Among individuals with incident AAA, plasma MMP12 abundance was independently associated with time to rupture or repair (HR 1.86, 95% CI 1.39–2.50, p < 0.001). Additionally, MMP12 level improved discrimination of AAA in an external cohort (SIMPLER; difference in AUROC 0.084, 95% CrI 0.081 - 0.087).

Conclusions: A parsimonious ProRS incorporating a single matrix metalloproteinase known to be implicated in AAA pathogenesis substantially improved prediction of AAA compared to clinical factors alone. These results may be used to inform personalized screening strategies for AAA grounded in individual biological risk.