Identifying Undiagnosed COPD Using Machine Learning-Based Classification.

Area of Science:

• Pulmonary Medicine
• Medical Informatics
• Machine Learning

Background:

• Chronic Obstructive Pulmonary Disease (COPD) is a major global health concern, causing significant morbidity and mortality.
• A high percentage of COPD cases remain undiagnosed, leading to disease progression and increased healthcare burdens.
• Current screening tools are limited, with few leveraging machine learning (ML) for early detection.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) classification model for identifying undiagnosed COPD.
• The model exclusively utilizes data commonly available in primary care settings.
• To assess the model's performance using metrics like sensitivity, precision, and AUROC.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (2007-2012).
• Included participants aged 20+ with complete spirometry and no prior COPD diagnosis.
• Defined undiagnosed COPD as post-bronchodilator FEV1/FVC < 0.7; trained and compared four ML models.

Main Results:

• The logistic regression model achieved the highest Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.84.
• Sensitivity was 0.72, but precision was low at 0.12, indicating many false positives.
• False negatives were younger, had fewer pack-years, and a higher proportion of never-smokers than true positives.

Conclusions:

• An ML model using primary care data can identify individuals with undiagnosed COPD.
• The model requires further refinement to enhance precision and undergo external validation.
• Clinical implementation necessitates ensuring generalizability and improved accuracy.