Integrative Phenogroups Based on Lung Imaging and Spirometry from Young Adulthood to Midlife and Associations with Subclinical Cardiac Structure and Function.
Authors
Affiliations (9)
Affiliations (9)
- Department of Preventive Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA. Electronic address: [email protected].
- Department of Preventive Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA.
- Division of Cardiology, Department of Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA.
- Division of Pulmonary and Critical Care Medicine, Department of Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA.
- Kaiser Permanente Northern California Division of Research, Pleasanton, CA.
- Department of Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, MA USA.
- Department of Medicine, Brigham and Women's Hospital, Boston, Massachusetts.
- Department of Preventive Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA; Division of Cardiology, Department of Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA.
- Department of Preventive Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA; Division of Pulmonary and Critical Care Medicine, Department of Medicine, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA.
Abstract
Chronic lung disease and heart failure (HF) commonly co-occur, share modifiable risk factors, and are preceded by a prolonged, heterogeneous, and subclinical phase that is poorly defined. Does the use of unsupervised machine learning identify distinct lung phenogroups, and are these phenogroups associated with cardiac structure and function? Participants from the Coronary Artery Risk Development in Young Adults study who completed lung imaging with computed tomography (CT), spirometry, and echocardiography were included. Gaussian mixture models were used to cluster ten lung features from CT and spirometry over 30-years into mutually-exclusive phenogroups. Multivariable adjusted linear and logistic regression estimated associations between lung phenogroups and cardiac structure and function parameters from year 30 (Y30) echocardiograms. Among 2,302 participants (mean age 25.1 ± 3.6 years; 58% female; 44% Black race), four lung-heart phenogroups at Y30 were identified: (1) ideal (2) emphysema-predominant with obstructive physiology; (3) mild interstitial/lung injury; and (4) substantial interstitial/lung injury with restrictive physiology. Compared with the ideal phenogroup, the substantial interstitial/lung injury group showed higher CT-measured lung injury (39% vs 1%) and interstitial change (12% vs 0.6%), worse cardiac remodeling, including higher LV mass/height (mean difference 4.4, [2.8, 6.1] and global longitudinal strain (%) (0.7, [0.2, 1.2]), and higher odds of Stage B HF (1.18 [1.08, 1.29]; all p<0.05). These findings were consistent among never-smokers. Machine-learning identified four distinct lung phenogroups in midlife, each defined by diverse subclinical lung and associated with different patterns of cardiac remodeling. Early subclinical lung features are associated with adverse cardiac remodeling and may increase risk of development of cardiopulmonary diseases.