Comparison of regression models for developing spirometric reference values

Abstract

Background: Modeling all age spirometric reference value was complicated for data have skewed distributions. However conventional statistical approaches for developing reference value such as multiple linear regression or quantile regression have limitations. Objective: To compare three regression approaches and identify the best approach for developing all age spirometric reference value. Methods: Different regression approaches to develop all age spirometric reference value by goodness-of-fit measures were compared including multiple linear regression, quantile regression and Generalized Additive Models for Location, Scale and Shape (GAMLSS). We analyzed spirometric data from healthy, never-smoking Chinese children and adults aged 5 to 80 years in the Second National Lung Function Survey (male, n=3279). Results: Spirometry were strongly associated with age, height and weight, with a large and nonlinear age effect across the age range. Variability depends nonlinearly on the age/height range. GAMLSS showed a much better fit regarding the estimation of spirometry data than multiple linear regression and quantile regression with respect to the generalized Akaike information criterion. Conclusions: GAMLSS seem to be more appropriate than multiple linear regression and quantile regression for modeling spirometric reference values, therefore it is a powerful approach for the construction of such references.