Automated CT biomarkers for opportunistic prediction of future cardiovascular events and mortality in an asymptomatic screening population: a retrospective cohort study

Abstract

Summary: Background: Body CT scans are frequently done for a wide range of clinical indications, but potentially valuable biometric information typically goes unused. We aimed to compare the prognostic ability of automated CT-based body composition biomarkers derived from previously developed deep-learning and feature-based algorithms with that of clinical parameters (Framingham risk score [FRS] and body-mass index [BMI]) for predicting major cardiovascular events and overall survival in an adult screening cohort. Methods: In this retrospective cohort study, mature and fully automated CT-based algorithms with predefined metrics for quantifying aortic calcification, muscle density, ratio of visceral to subcutaneous fat, liver fat, and bone mineral density were applied to a generally healthy asymptomatic outpatient cohort of adults aged 18 years or older undergoing abdominal CT for routine colorectal cancer screening. To assess the association between the predictive measures (CT-based vs FRS and BMI) and downstream adverse events (death or myocardial infarction, cerebrovascular accident, or congestive heart failure subsequent to CT scanning), we used both an event-free survival analysis and logistic regression to compute receiver operating characteristic curves (ROCs) . Findings: 9223 people (mean age 57·1 years [SD 7·8]; 5152 [56%] women and 4071 [44%] men) who underwent CT scans between April, 2004, and December, 2016, were included in this analysis. In the longitudinal clinical follow-up (median 8·8 years [IQR 5·1–11·6]), subsequent major cardiovascular events or death occurred in 1831 (20%) patients. Significant differences were observed for all five automated CT-based body composition measures according to adverse events (p<0·001). Univariate 5-year area under the ROC (AUROC) values for predicting death were 0·743 (95% CI 0·705–0·780) for aortic calcification, 0·721 (0·683–0·759) for muscle density, 0·661 (0·625–0·697) for ratio of visceral to subcutaneous fat, 0·619 (0·582–0·656) for liver density, and 0·646 (0·603–0·688) for vertebral density, compared with 0·499 (0·454–0·544) for BMI and 0·688 (0·650–0·727) for FRS. Univariate hazard ratios for highest-risk quartile versus others for these same CT measures were 4·53 (95% CI 3·82–5·37) for aortic calcification, 3·58 (3·02–4·23) for muscle density, 2·28 (1·92–2·71) for the ratio of visceral to subcutaneous fat, 1·82 (1·52–2·17) for liver density, and 2·73 (2·31–3·23) for vertebral density, compared with 1·36 (1·13–1·64) for BMI and 2·82 (2·36–3·37) for FRS. Multivariate combinations of CT biomarkers further improved prediction over clinical parameters (p<0·05 for AUROCs). For example, the 2-year AUROC from combining aortic calcification, muscle density, and liver density for predicting death was 0·811 (95% CI 0·761–0·860). Interpretation: Fully automated quantitative tissue biomarkers derived from CT scans can outperform established clinical parameters for presymptomatic risk stratification for future serious adverse events and add opportunistic value to CT scans performed for other indications. Funding: Intramural Research Program of the National Institutes of Health Clinical Center.