Is Achieving a Fully Race-Neutral Approach to Lung Function Classification Even Possible?

Amin Adibi, Mohsen Sadatsafavi, Emily Brigham, Surya Bhatt

Abstract:

Airflow obstruction is assessed by estimating “normal” lung function using population-derived reference equations and measuring deviations from it using Z-scores (1). In 2023, the American Thoracic Society recommended replacing race-specific Global Lung Function Initiative (GLI)-2012 reference equations with race-weighted GLI-Global equations to address concerns about perpetuating race-based medicine and normalizing lower lung function in underserved populations.

Several studies refer to race-neutral Z-scores or percent predicted values. However, the definition of race neutrality is often unclear. We propose a more structured framework to define race neutrality and evaluate the statistical properties of spirometric values from different equations.

Race neutrality of spirometric values can be defined in at least three ways: 1) eliminating the collection of race information for spirometry (race-agnostic encounters), 2) ensuring that spirometric values have the same prognostic implications regardless of race (race-agnostic interpretability), and 3) ensuring that values do not vary statistically between racial groups and cannot predict race (race-unawareness). Criteria 2 and 3 align with notions of sufficiency and independence in algorithmic fairness literature, respectively. We demonstrate that no commonly used spirometry outputs can simultaneously satisfy all definitions of race neutrality and that the applicable definition depends on the context.