Debunking the GAMLSS Myth: Simplicity Reigns in Pulmonary Function Diagnostics.

Rationale: Current technical standards advocate using Generalized, Additive Models of Location, Scale, and Shape (GAMLSS) for lung function reference equations. These equations are complicated and require supplementary spline tables.

Objective: (1) To demonstrate that segmented (piecewise) linear regression (SLR) yields similar prediction accuracies as GAMLSS in pulmonary function diagnostics. (2) To determine the agreement between both SLR and GAMLSS.

Methods: The NHANES 2007-2012 database was utilized to construct spirometric reference equations for FEV₁, FVC, and FEV₁/FVC using both SLR and GAMLSS modeling techniques. K-fold cross-validation was used to provide the 95% confidence interval (CI) of the root-mean-square error (RMSE) as indicators of prediction accuracy. Additionally, agreement was assessed between the two modeling techniques in classifying spirometric patterns (standard, airflow obstruction, restrictive, or mixed disorder) using an unweighted kappa statistic.

Results: The RMSE values for FEV₁, FVC, and FEV₁/FVC and correlation coefficients between predicted values and test data were similar between the two techniques. Agreement in classifying spirometric patterns between the two techniques ranged from 0.78 to 0.80 (95% CI).

Conclusions: The findings suggest that simple linear regression for FEV₁/FVC and SLR for FEV₁ and FVC offer prediction accuracies on par with GAMLSS while being more straightforward, parsimonious, and accessible to a broader audience in the field of pulmonary function diagnostics.