Obtaining the Most Accurate, Explainable Model for Predicting Chronic Obstructive Pulmonary Disease: Triangulation of...

Area of Science:

• Public Health
• Environmental Health
• Data Science

Background:

• Chronic obstructive pulmonary disease (COPD) remains a significant health burden in the United States.
• Despite declining smoking rates, understanding and predicting COPD prevalence is crucial.
• This study analyzes COPD in the U.S. from 2016 to 2019.

Purpose of the Study:

• To compare linear and machine learning models for predicting COPD rates.
• To identify key predictors of COPD at the Core-Based Statistical Area (CBSA) level.
• To develop accurate and interpretable models for COPD prediction.

Main Methods:

• Integrated non-personally identifiable data from CDC sources.
• Included variables such as cigarette smoking, race/ethnicity, air quality index, education, employment, and economic factors.
• Applied multiple linear regression and machine learning techniques.

Main Results:

• Machine learning models achieved higher accuracy (85.7% variance explained) than linear models (81.1%).
• Cigarette smoking and household income were the strongest predictors.
• Education, unemployment, and percentages of American Indian/Alaska Native, Black, and Hispanic populations were also significant predictors.

Conclusions:

• Diverse data sources and multiple modeling methods are essential for understanding and predicting COPD.
• Gradient boosted tree models demonstrated superior accuracy by capturing nonlinearities.
• Findings support tailored interventions for specific communities and highlight the need for further research on air quality impacts.