Leveraging deep learning applied to chest radiograph images to identify individuals at high risk of chronic obstructive pulmonary disease: a retrospective model validation study

Background Chronic obstructive pulmonary disease (COPD) is a leading cause of mortality, yet early detection remains challenging. This study assessed whether deep learning applied to routine outpatient chest radiographs (CXRs) can identify individuals at high risk of incident COPD. Methods Using cancer screening trial data, we previously developed a convolutional neural network (CXR-Lung-Risk) to predict lung-related mortality from a CXR image. In this retrospective model validation study, we externally validated whether CXR-Lung-Risk was associated with incident COPD from routine CXRs. We identified outpatients without lung cancer, COPD, or emphysema who had a CXR taken from Jan 1, 2013, to Dec 31, 2014, at Massachusetts General or Brigham and Women's Hospitals in Boston, MA, USA. The primary outcome was 6-year incident COPD. Discrimination was assessed using area under the receiver operating characteristic curve (AUC) compared with the TargetCOPD clinical risk score. All analyses were stratified by smoking status. A secondary analysis was conducted in the Project Baseline Health Study (PBHS) to test associations between CXR-Lung-Risk with pulmonary function and plasma protein abundance. The PBHS study is registered with ClinicalTrials.gov, NCT03154346. Findings The primary analysis consisted of data from 12 550 ever-smokers (mean age 62.4 years [SD 6.8], 6135 [48.9%] male, 6415 [51.1%] female) and 15 298 never-smokers (mean age 63.0 years [8.1], 6550 [42.8%] male, 8748 [57.2%] female). 1562 (12.4%) of 12 550 ever-smokers and 580 (3.8%) of 15 298 never-smokers developed COPD within 6 years. CXR-Lung-Risk had additive predictive value beyond the TargetCOPD score for 6-year incident COPD in both ever-smokers (CXR-Lung-Risk + TargetCOPD AUC 0.73 [95% CI 0.72-0.74] vs TargetCOPD alone AUC 0.66 [0.65-0.68], p<0.0001) and never-smokers (CXR-Lung-Risk + TargetCOPD AUC 0.70 [0.67-0.72] vs TargetCOPD alone AUC 0.60 [0.57-0.62], p<0.0001). In secondary analyses of 2097 individuals in the PBHS, CXR-Lung-Risk was associated with worse pulmonary function and with abundance of SCGB3A2 (secretoglobin family 3A member 2) and LYZ (lysozyme), proteins involved in pulmonary physiology. Interpretation In this external validation, a deep-learning model applied to routine CXR images identified individuals at high risk of incident COPD, beyond known risk factors. Patients at high risk might benefit from diagnostic spirometry and subsequent preventive care. Copyright (c) 2025 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).