Abstract 13975: Proteomic Scores for Cardiac Magnetic Resonance Imaging Phenotypes and Heart Failure Risk
Abstract
Background: Early detection of patients at risk of developing heart failure (HF) is critical. While detection of subclinical cardiac remodeling, a precursor of clinical HF, is possible by cardiac magnetic resonance imaging (CMR) derived measures of ventricular and atrial structure and function, it is difficult to implement at the population level. Circulating proteomics may offer a molecular readout of these remodeling processes. We developed proteomic scores as molecular proxies for CMR-derived cardiac phenotypes and examined their associations with incident HF. Methods: Using CMR and plasma proteomics data from 1,101 UK Biobank participants, we examined 17 left ventricular (LV), left atrial (LA), and right-heart structure/function phenotypes. LASSO models were trained in 70% of participants and evaluated in the remaining 30%, with predictive performance assessed using test-set r derived from Pearson correlation. Protein-predicted CMR phenotypes were calculated as weighted sum using the LASSO-derived coefficients and standardized as z-scores using baseline proteomics from 51,441 non-overlapping participants, including 2,394 incident HF cases. Fine-Gray model was used to assess the association between the protein-predicted CMR phenotypes and incident HF, adjusting for risk factors in PREVENT. Results: We created protein-predicted scores for 15 of the 17 CMR phenotypes based on LASSO-selected proteins. Test-set predictive performance was highest for LV myocardial mass ( r = 0.59), RV end-diastolic volume ( r = 0.45), and RV end-systolic volume ( r = 0.40). 11 protein-predicted CMR phenotype scores were strongly associated with incident HF. Higher protein-predicted LA minimum and maximum volume were associated with increased incident HF (HRs of 1.69 and 1.63, respectively), while higher protein-predicted LA ejection fraction was inversely associated with incident HF (HR 0.55). LV remodeling phenotypes were also consistently associated with HF. Protein-predicted LV end-diastolic volume and LV stroke volume were associated with incident HF, with HRs of 1.42 and 1.44, respectively. Three RA CMR phenotypes were associated with incident HF. Adding protein-predicted CMR phenotypes to fully adjusted models improved HF risk discrimination by up to 2.6%. Conclusions: Circulating proteomic signatures of CMR phenotypes were associated with incident HF and improved HF risk prediction, supporting plasma proteomics as a scalable proxy for imaging-defined cardiac remodeling.