Aortic and Cardiac Structure From Routine CT Predict Cardiovascular Risk Beyond PREVENT and Coronary Calcium.
Authors
Affiliations (7)
Affiliations (7)
- Cardiovascular Imaging Research Center (CIRC), Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA.
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, Massachusetts, USA.
- Cardiovascular Imaging Research Center (CIRC), Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA; Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, Massachusetts, USA.
- Hinda and Arthur Marcus Institute on Aging Research, Hebrew SeniorLife, Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, USA.
- Cardiovascular Imaging Research Center (CIRC), Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA; Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, Massachusetts, USA; Radiology and Nuclear Medicine, GROW and CARIM Maastricht University, Maastricht, Netherlands.
- Cardiovascular Research Center and Center for Genomic Medicine, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA; Program in Medical and Population Genetics and Cardiovascular Disease Initiative, Broad Institute of Harvard and MIT, Cambridge, Massachusetts, USA.
- Cardiovascular Imaging Research Center (CIRC), Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA; Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, Massachusetts, USA. Electronic address: [email protected].
Abstract
Cardiovascular disease prevention relies on accurate risk assessment; however, existing scores are imprecise. Routine imaging may be opportunistically used to predict risk. The authors tested whether computed tomography (CT)-derived cardiac and aortic structure predicts major adverse cardiac events (MACE) beyond standard-of-care scores. The authors developed a least absolute shrinkage and selection operator model to predict cardiovascular mortality using "radiomics" features describing cardiac and aortic structure from 13,437 lung cancer screening CTs from the NLST (National Lung Screening Trial). They compared this score to the PREVENT (Predicting Risk of Cardiovascular Disease Events) tool and the coronary artery calcium (CAC) score in patients with routine chest CT and no prior MACE from Mass General Brigham. They calculated discrimination using Harrel's C-index and MACE rates in high-risk groups by the PREVENT score (≥7.5% risk) or the radiomics score (≥3.0% in men, ≥1.5% in women). In external testing, (n = 14,577, mean age 61.1 ± 8.6 years, 47.5% male), 6.2% had incident MACE over a median of 5.7 years of follow-up. The radiomics score had higher discrimination for MACE than PREVENT (C-index 0.66 [95% CI: 0.64-0.68] vs 0.61 [95% CI: 0.59-0.63]) and was complementary to CAC (combined C-index 0.69 [95% CI: 0.67-0.71] vs CAC alone 0.66 [95% CI: 0.65-0.68]). High-risk patients by the radiomics score but not PREVENT had 3.6-fold higher MACE incidence than low-risk patients by both scores (23.1 [95% CI: 16.7-30.2] vs 6.5 [95% CI: 5.5-7.5] MACE per 1,000 person-years). Aortic surface-to-volume ratio, left ventricular volume, and left atrial short-axis length were among the most predictive features of MACE. CT-derived structural cardiac and aortic radiomics identified high-risk patients missed by clinical scores and further stratified risk among CAC risk groups. High-risk patients may benefit from intensified primary prevention.