Radiomics-based risk prediction of heart failure using CT calcium score exam

Case Western Reserve University, Cleveland OH

Investigators

Abstract

Radiomics-based risk prediction of heart failure using CT calcium score exam Project Summary Heart failure (HF) is a major cause of morbidity and mortality in the United States. There is a compelling need for personalized pre-emptive HF risk prediction to facilitate precise preventive strategies to alleviate the population burden. However, currently, there are no widely validated models for HF risk prediction, which hinders the timely identification of a pre-emptive initiation of therapies in at-risk patients. Non-contrast low-dose CT scans for coronary artery calcium scoring (CT calcium scoring, CTCS), already widely utilized for risk assessment in atherosclerotic cardiovascular disease, offer an opportunity to identify key pathophysiologic pathways causally or consequentially linked with HF risk. The central premise of this transformative proposal is that a reliable and reproducible CTCS-based HF-specific clinico-radiomic risk model (CTCSHF) will enable the precision identification of patients at risk for HF. We will leverage the University Hospitals CLARIFY program, the world’s largest free CTCS program (>100k unique participants with ~15K increase per year and > 3000 HF events) and our Houston Methodist HeartScan CTCS program (> 50k unique participants with ~10K increase per year and ~2k HF events) in addition to 2 external prospective NIH-funded cohorts (CARDIA and CRIC), that together will provide a robust opportunity for model derivation and validation. Together with institutional expertise in computational image analysis and cardiovascular imaging, we will develop and validate a comprehensive machine learning based analysis of CTCS to identify key image-based biomarkers (radiomics) corresponding to 5 pathophysiologic domains linked with HF risk. They include cardiac remodeling (size/shape), atherosclerosis (coronary/vascular calcification), hemodynamics (aorta/pulmonary artery, size/valvular calcium), visceral adiposity (liver and epicardial adipose tissue), and sarcopenia (skeletal muscle, bone density), combined with clinical factors and demographics, to predict future HF events. In Aim 1, we will develop and validate an automated radiomic extraction tool from CTCS. In Aim 2, we will develop an HF risk prediction model (CTCSHF) incorporating CTCS-derived radiomics and clinical risk factors in >160,000 participants from 4 large well-charactered prospective cohorts with > 5000 incident HF events. In Aim 3, we will explore CTCSHF model fairness across socio-racial groups and investigate the utility of fairness-aware clinico-radiomic HF risk models (i.e., in subgroups of race and socioeconomic status) in improving accuracy. Improved characterization of HF risk will advance knowledge of cardiometabolic disease phenotypes and support clinical therapeutic decision-making and patient counseling for improved adherence.