Extreme value analysis has the potential to improve spirometry interpretation

Spirometry is a measure of lung function used to make clinical decisions regarding the diagnosis and management of respiratory conditions. The goal of spirometry interpretation is to relate impairments in mechanical lung function to pulmonary pathology. In many circumstances, interpretation of measured spirometric values relies on comparisons with a 'healthy' reference population. Such inferences assume that spirometric values in healthy populations follow a Gaussian distribution with impaired values concentrated in the lower tail.

We hypothesised that impairments in lung function mechanics generate spirometric values that follow a non-Gaussian distribution as predicted by extreme value analysis. We used a Gumbel distribution to model lung function impairment. We compared the Gaussian healthy distribution to the Gumbel impaired distribution to calculate the relative probability for impaired versus healthy values. The relative probability provides an objective measure of the likelihood that a particular spirometric value is within the healthy or impaired distribution.

The potential usefulness of the relative probability was demonstrated in simulated cases, providing an objective delineation of the zone of uncertainty in the transition from normal to impaired lung function.

Considering the spirometric measures from people with lung function impairment as a separate distribution from people with healthy lung function provides a more analytic assessment of impairment. Applying extreme value analysis, the relative probability of a spirometric measurement being impaired versus healthy and a more precise definition of the zone of uncertainty potentially provides more objective discrimination of the intersection between the healthy and impaired lung function ranges.