Search

Joyce D Schroeder,1 Ricardo Bigolin Lanfredi,2 Tao Li,3 Jessica Chan,1 Clement Vachet,4 Robert Paine III,5 Vivek Srikumar,3 Tolga Tasdizen2 1Department of Radiology and Imaging Sciences, School of Medicine, University of Utah, Salt Lake City, UT, USA; 2Department of Electrical and Computer Engineering, Scientific Computing and Imaging Institute (SCI), University of Utah, Salt Lake City, UT, USA; 3School of Computing, University of Utah, Salt Lake City, UT, USA; 4Biomedical Imaging and Data Analytics Core, SCI, University of Utah, Salt Lake City, UT, USA; 5Division of Pulmonary and Critical Care Medicine, School of Medicine, University of Utah, Salt Lake City, UT, USACorrespondence: Joyce D SchroederDepartment of Radiology and Imaging Sciences, School of Medicine, University of Utah, 30 North 1900 East, Rm #1A71, Salt Lake City, UT 84132, USATel +1 801 581 7553Fax +1 801 581 2414Email joyce.schroeder@hsc.utah.eduBackground: Chronic obstructive pulmonary disease (COPD), the third leading cause of death worldwide, is often underdiagnosed.Purpose: To develop machine learning methods to predict COPD using chest radiographs and a convolutional neural network (CNN) trained with near-concurrent pulmonary function test (PFT) data. Comparison is made to natural language processing (NLP) of the associated radiologist text reports.Materials and Methods: This IRB-approved single-institution retrospective study uses 6749 two-view chest radiograph exams (2012– 2017, 4436 unique subjects, 54% female, 46% male), same-day associated radiologist text reports, and PFT exams acquired within 180 days. The Image Model (Resnet18 pre-trained with ImageNet CNN) is trained using frontal and lateral radiographs and PFTs with 10% of the subjects for validation and 19% for testing. The NLP Model is trained using radiologist text reports and PFTs. The primary metric of model comparison is the area under the receiver operating characteristic curve (AUC).Results: The Image Model achieves an AUC of 0.814 for prediction of obstructive lung disease (FEV1/FVC < 0.7) from chest radiographs and performs better than the NLP Model (AUC 0.704, p< 0.001) from radiologist text reports where FEV1 = forced expiratory volume in 1 second and FVC = forced vital capacity. The Image Model performs better for prediction of severe or very severe COPD (FEV1 < 0.5) with an AUC of 0.837 versus the NLP model AUC of 0.770 (p< 0.001).Conclusion: A CNN Image Model trained on physiologic lung function data (PFTs) can be applied to chest radiographs for quantitative prediction of obstructive lung disease with good accuracy.Keywords: machine learning, chronic obstructive pulmonary disease, quantitative image analysis, natural language processing

Marni Stott-Miller,1 Hana Müllerová,2 Bruce Miller,3 Maggie Tabberer,4 Céline El Baou,5 Tom Keeley,4 Fernando J Martinez,6 Meilan Han,7 Mark Dransfield,8 Nadia N Hansel,9 Christopher B Cooper,10 Prescott Woodruff,11 Victor E Ortega,12 Alejandro P Comellas,13 Robert Paine III,14 Richard E Kanner,14 Wayne Anderson,15 M Bradley Drummond,15 Victor Kim,16 Ruth Tal-Singer,17 Aili L Lazaar3 1GSK R&D, Epidemiology: Value, Evidence and Outcomes, Uxbridge, UK; 2AstraZeneca, Cambridge, UK; 3GSK R&D, Discovery Medicine, Collegeville, PA, USA; 4GSK R&D Patient-Centred Outcomes: Value, Evidence and Outcomes, Uxbridge, UK; 5CEBSTAT Consultancy Ltd, London, UK; 6Cornell Medical College, New York, NY, USA; 7Division of Pulmonary and Critical Care at the University of Michigan, Ann Arbor, MI, USA; 8Children’s of Alabama, Children’s Health Research Unit/University of Alabama, Birmingham, AB, USA; 9Pulmonary and Critical Care Medicine, Johns Hopkins University, Baltimore, MD, USA; 10David Geffen School of Medicine at UCLA, Los Angeles, CA, USA; 11UCSF Department of Medicine, San Francisco, CA, USA; 12Wake Forest School of Medicine, Winston Salem, NC, USA; 13Carver College of Medicine, University of Iowa, Iowa City, IA, USA; 14Division of Pulmonary and Critical Care Medicine, University of Utah, Salt Lake City, UT, USA; 15Division of Pulmonary Diseases and Critical Care Medicine, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA; 16Department of Thoracic Medicine and Surgery, Lewis Katz School of Medicine at Temple University, Philadelphia, PA, USA; 17COPD Foundation, Washington, DC, USACorrespondence: Aili L Lazaar GSK, 1250 S. Collegeville Road, Collegeville, PA 19426-0989, USATel +1 484-923-3730Email Aili.L.Lazaar@gsk.comBackground: Chronic cough and phlegm are frequently reported chronic obstructive pulmonary disease (COPD) symptoms. Prior research classified chronic mucus hypersecretion (CMH) based on the presence of these symptoms for ≥ 3 months, called chronic bronchitis (CB) if respiratory infection symptoms were present for 1– 2 years (Medical Research Council [MRC] definition). We explored whether the COPD Assessment Test (CAT), a simple measure developed for routine clinical use, captures CMH populations and outcomes similarly to MRC and St. George’s Respiratory Questionnaire (SGRQ) definitions.Methods: We identified CMH in the SPIROMICS COPD cohort using (a) MRC definitions, (b) SGRQ questions for cough and phlegm (both as most/several days a week), and (c) CAT cough and phlegm questions. We determined optimal cut-points for CAT items and described exacerbation frequencies for different CMH definitions. Moderate exacerbations required a new prescription for antibiotics/oral corticosteroids or emergency department visit; severe exacerbations required hospitalization. Results were stratified by smoking status.Results: In a population of 1431 participants (57% male; mean FEV1% predicted 61%), 47% and 49% of evaluable participants had SGRQ- or CAT-defined CMH, respectively. A cut-point of ≥ 2 for cough and phlegm items defined CMH in CAT. Among SGRQ-CMH+ participants, 80% were also defined as CMH+ by the CAT. CMH+ participants were more likely to be current smokers. A higher exacerbation frequency was observed for presence of CMH+ versus CMH− in the year prior to baseline for all CMH definitions; this trend continued across 3 years of follow-up, regardless of smoking status.Conclusion: Items from the CAT identified SGRQ-defined CMH, a frequent COPD trait that correlated with exacerbation frequency. The CAT is a short, simple questionnaire and a potentially valuable tool for telemedicine or real-world trials. CAT-based CMH is a novel approach for identifying clinically important characteristics in COPD that can be ascertained in these settings.Keywords: COPD, SGRQ, exacerbation, CAT, cough, phlegm

Christopher B Cooper,1 Robert Paine,2 Jeffrey L Curtis,3,4 Richard E Kanner,2 Carlos H Martinez,3 Catherine A Meldrum,3 Russell Bowler,5 Wanda O’Neal,6 Eric A Hoffman,7 David Couper,6 Miguel Quibrera,6 Gerald Criner,8 Mark T Dransfield,9 MeiLan K Han,3 Nadia N Hansel,10 Jerry A Krishnan,11 Stephen C Lazarus,12 Stephen P Peters,13 R Graham Barr,14 Fernando J Martinez,15 Prescott G Woodruff12 On behalf of the SPIROMICS investigators1Departments of Medicine and Physiology, David Geffen School of Medicine, University of California Los Angeles, Los Angeles, CA, USA; 2Section of Pulmonary and Critical Care Medicine, Department of Veterans Affairs Medical Center, University of Utah, Salt Lake City, UT, USA; 3Pulmonary and Critical Care Medicine Division, Department of Internal Medicine, University of Michigan Health System, Ann Arbor, MI, USA; 4Veterans Affairs Ann Arbor Healthcare System, Ann Arbor, MI, USA; 5National Jewish Health, University of Colorado School of Medicine, Denver, CO, USA; 6University of North Carolina Marisco Lung Institute, Chapel Hill, NC, USA; 7Department of Radiology, University of Iowa, Iowa City, IA, USA; 8Department of Pulmonary and Critical Care Medicine, Temple University, Philadelphia, PA, USA; 9Department of Medicine, University of Alabama at Birmingham, Birmingham, AL, USA; 10Johns Hopkins University School of Medicine, Baltimore, MD, USA; 11Department of Medicine, University of Illinois at Chicago, Chicago, IL, USA; 12Division of Pulmonary and Critical Care Medicine, Department of Medicine, University of California San Francisco, San Francisco, CA, USA; 13Wake Forest School of Medicine, Winston-Salem, NC, USA; 14Department of Medicine, Columbia University Medical Center, New York, NY, USA; 15Joan and Sanford I Weill Department of Medicine, Division of Pulmonary and Critical Care Medicine, Weill Cornell Medicine, New York, NY, USACorrespondence: Christopher B CooperDepartments of Medicine and Physiology,David Geffen School of Medicine, University of California Los Angeles, 10833 Le Conte Avenue, 37-131 CHS, Los Angeles, CA 90095, USAEmail ccooper@mednet.ucla.eduRationale: Some COPD patients develop extreme breathlessness, decreased exercise capacity and poor health status yet respiratory disability is poorly characterized as a distinct phenotype.Objective: To define respiratory disability in COPD based on available functional measures and to determine associations with risk for exacerbations and death.Methods: We analyzed baseline data from a multi-center observational study (SPIROMICS). This analysis includes 2332 participants (472 with severe COPD, 991 with mild/moderate COPD, 726 smokers without airflow obstruction and 143 non-smoking controls).Measurements: We defined respiratory disability by ≥ 4 of 7 criteria: mMRC dyspnea scale ≥ 3; Veterans Specific Activity Questionnaire < 5; 6-minute walking distance < 250 m; St George’s Respiratory Questionnaire activity domain > 60; COPD Assessment Test > 20; fatigue (FACIT-F Trial Outcome Index) < 50; SF-12 < 20.Results: Using these criteria, respiratory disability was identified in 315 (13.5%) participants (52.1% female). Frequencies were severe COPD 34.5%; mild-moderate COPD 11.2%; smokers without obstruction 5.2% and never-smokers 2.1%. Compared with others, participants with disability had more emphysema (13.2 vs. 6.6%) and air-trapping (37.0 vs. 21.6%) on HRCT (P< 0.0001). Using principal components analysis to derive a disability score, two factors explained 71% of variance, and a cut point − 1.0 reliably identified disability. This disability score independently predicted future exacerbations (ß=0.34; CI 0.12, 0.64; P=0.003) and death (HR 2.97; CI 1.54, 5.75; P=0.001). Thus, participants with disability by this criterion had almost three times greater mortality compared to those without disability.Conclusion: Our novel SPIROMICS respiratory disability score in COPD was associated with worse airflow obstruction as well as airway wall thickening, lung parenchymal destruction and certain inflammatory biomarkers. The disability score also proved to be an independent predictor of future exacerbations and death. These findings validate disability as an important phenotype in the spectrum of COPD.Keywords: disability, frailty, exacerbation rate, mortality, SPIROMICS