Mapping the Zone of Uncertainty in Pulmonary Function Test Interpretation

European Respiratory Society and American Thoracic Society (ERS/ATS) guidelines for pulmonary function test (PFT) interpretation posit the existence of a "zone of uncertainty." However, as reference equations have been defined only for healthy lung function, the precise borders of the zone of uncertainty remain unspecified. To address this limitation, we sought to define distributions of both healthy and diseased lung function and to use these distributions to map the zone of uncertainty.

We used a latent class model to define distributions of healthy and diseased FEV₁/FVC values. To represent the distribution of healthy FEV₁/FVC values, we fit a Box-Cox Cole Green distribution to spirometry from healthy adult participants in the continuous National Health and Nutrition Examination Survey (NHANES). Using the distribution of healthy FEV₁/FVC values, we then fit a latent class model to FEV₁/FVC data from adult patients who underwent pulmonary function testing at the University of Pennsylvania Health System (UPHS) between 2000 and 2023. The distribution of diseased FEV₁/FVC values and the prior probabilities that FEV₁/FVC values had been sampled from healthy and diseased populations were selected to maximize the likelihood of the observed UPHS data. We considered the normal, Box-Cox Cole Green, and Box-Cox power exponential distributions for the distribution of diseased FEV₁/FVC values and selected the distribution that minimized the Bayesian information criterion. We then mapped the zone of uncertainty by identifying the range of FEV₁/FVC values in which the distributions of healthy and diseased lung function overlapped.

Pre-bronchodilator spirometry data were collected from 14,075 NHANES participants-of whom 6,063 were without respiratory symptoms or a history of tobacco use-and 41,437 UPHS patients. Healthy lung function was represented by a Box-Cox Cole Green distribution with a median of 0.81, a coefficient of variation of 0.08, and skewness of 1.70. Diseased lung function was best represented by a Box-Cox power exponential distribution with a median of 0.56, a coefficient of variation of 0.30, skewness of 0.91, and kurtosis of 2.36. In the latent class model, the prior probability that an FEV₁/FVC value was healthy was 76.2%, while the prior probability that it was diseased was 23.8%. The overlap of the distributions of healthy and diseased FEV₁/FVC values defined a zone of uncertainty in the interval [0.64, 0.95]. The FEV₁/FVC values of 33,747 (81.4%) UPHS patients fell within the zone of uncertainty, including 30,288 (99.2%) patients with a normal FEV₁/FVC and 3,459 (31.7%) patients with an abnormal FEV₁/FVC.

This exploratory study presents evidence that in a clinical cohort, most FEV₁/FVC values fall within the zone of uncertainty. Further research is needed to develop optimal ways to represent and incorporate this uncertainty into PFT interpretation.