A Predictive Model using Artificial Intelligence on Chest Radiograph in Addition to History and Physical Examination to Diagnose Chronic Obstructive Pulmonary Disease
Jatuporn Wanchaitanawong1, Apichart So-Ngern2, Panaya Tumsatan3, Wipa Reechaipichitkul1, Itthiphat Arunsurat1, Pailin Ratanawatkul1, Worawat Chumpangern1
Affiliation : 1 Division of Pulmonary and Critical Care Medicine, Department of Medicine, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand, 2 Division of Sleep Medicine, Department of Medicine, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand, 3 Department of Radiology, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand
Objective: Spirometry is the gold standard for chronic obstructive pulmonary disease (COPD) diagnosis. Some patients are unable to perform spirometry. The study aimed to evaluate the factors associated with COPD and create the predictive model for COPD diagnosis.
Materials and Methods: A cross-sectional study between January 1, 2020, and December 31, 2020, at Srinagarind Hospital included subjects aged ≥40 years who had productive cough or dyspnea >3 months without lung parenchymal disease. Information from history taking, physical examination, chest x-ray (CXR) and spirometry were collected. The stepwise backward multiple logistic regression was performed to evaluate the factors associated with COPD.
Results: One hundred and eight subjects were enrolled; 46 COPD and 62 non-COPD. The independent factors associated with COPD diagnosis were cigarette smoking ≥30 pack-year, body mass index (BMI) <22 kg/m2, wheezing on forced expiration, modified Medical Research Council Dyspnea Scale (mMRC) ≥2 and emphysema interpreted by AI. The model consisting of these factors showed an area under the receiver operating characteristic curve 0.86 (95% CI, 0.77 to 0.92) for COPD diagnosis. The sensitivity and specificity were 8.7% (95% CI, 2.4 to 20.1%), 100% (95% CI, 94.2 to 100%). The positive predictive value and negative predictive value were 100% (95% CI, 39.8 to 100%), 59.6% (95% CI, 49.5 to 69.1%).
Conclusion: The model consisting of factors including cigarette smoking ≥30 pack-year, BMI <22 kg/m2, wheezing on forced expiration and presence of emphysema on CXR interpreted by AI had high specificity for COPD diagnosis. The model could be used as a diagnostic tool for those who are unable to perform spirometry.
doi.org/10.35755/jmedassocthai.2021.S04.00049
Keywords : Factors; Model; COPD diagnosis; Artificial intelligence
