Validation of Questionnaire-based Case Definitions for Chronic Obstructive Pulmonary Disease

Area of Science:

• Pulmonary Medicine
• Epidemiology
• Data Science

Background:

• Questionnaire-based definitions for chronic obstructive pulmonary disease (COPD) lack validation against objective lung function data.
• Previous COPD definitions relied on physician diagnosis, symptoms, and smoking history.
• A novel definition was developed using machine learning for improved accuracy.

Purpose of the Study:

• To validate existing and develop a new COPD definition using National Health and Nutrition Examination Survey (NHANES) data.
• To compare the performance of different COPD definitions against spirometry.
• To identify key predictors for COPD in smokers.

Main Methods:

• Utilized NHANES 2007-2012 data from 7,996 adults aged 40-79 who underwent spirometry.
• Defined 'true' COPD cases based on postbronchodilator spirometry (FEV1/FVC ratio < 0.7 or LLN).
• Developed a gradient boosting model for smokers, incorporating sociodemographics, exposures, clinical variables, and symptoms.

Main Results:

• Spirometry revealed COPD prevalence of 26% in smokers and 8% in never-smokers.
• Existing definitions showed limited sensitivity (18-35%) but good specificity (88-92%) in smokers.
• The new definition, using age, bronchodilator use, BMI, smoking pack-years, and dust exposure, achieved the highest sensitivity (35%) and specificity (92%) in smokers.

Conclusions:

• The developed COPD definition demonstrates improved accuracy for identifying cases in smokers.
• Findings aid in quantitative bias analysis when objective pulmonary function data are missing.
• The study highlights the utility of machine learning in refining epidemiological definitions.