A refined spirometry dataset for comparing segmented (piecewise) linear models to that of GAMLSS

Generalized Additive Models for Location, Scale, and Shape (GAMLSS) are widely used for developing spirometric reference equations but are often complex, requiring additional spline tables. This study explores the potential of Segmented (piecewise) Linear Regression as an alternative, comparing its predictive accuracy to GAMLSS and examining the agreement between the two methods. Spirometry data from nearly 16,600 patients, deemed Grade “A” and “B” acceptable from the NHANES 2007-2012 dataset, was analyzed. The dataset includes both nominal and scalar variables. Reference equations for forced expiratory volume in 1 s (FEV₁), forced vital capacity (FVC), and the ratio (FEV₁/FVC) were generated using GAMLSS (FEV₁, FVC, FEV₁/FVC), Segmented Linear Regression (FEV₁, FVC) and multiple linear regression (FEV₁/FVC). K-fold cross-validation was employed to compare prediction accuracy, using root-mean-square error (RMSE) and correlation coefficients. Agreement in classifying spirometric patterns (i.e. airway obstruction, restrictive spirometry pattern, mixed obstructive and restrictive disorder) was evaluated with the kappa statistic. This study uniquely compares the models by incorporating the lower limit of normal (LLN) using fitted z-scores of –1.645 or –1.96. The dataset is publicly available in SPSS (.sav) and .csv formats through the Mendeley Data repository.