Introduction We outline a novel approach for longitudinal modelling of lung function with long-term follow-up in which within-patient variation over time is described by a stationary (mean-reverting) stochastic process, and apply these techniques to a unique dataset of cystic fibrosis patients in Denmark. The aim is to quantify how lung function changes in chronic lung diseases.

Methods The Danish CF register contains data collected on a monthly basis with up to 30 years of follow-up. Our statistical analysis framework is that of a linear mixed effects model with longitudinally structured correlation. Using open-source software we describe how to partition the variability in the data into three components (between and within patient, and measurement error) using the empirical variogram. A parametric model for lung function decline can then be developed. We apply this approach to explore the effect of age, birth cohort and infection status on lung function decline.

Results The dataset contains 70 448 measures on 479 patients seen between 1960 and 2009. The empirical variogram shows slowly decaying long-term correlation (>15 years) in FEV1, with half of the variability in lung function explained by within person variation. The mean rate of lung function decline is 0.96% per year (95% CI 0.86 to 1.07). There is a significant cohort effect, and chronic infection significantly increases the rate of lung function decline.

Conclusions We apply a novel modelling approach to demonstrate that lung function in early life in the Danish cystic fibrosis population is correlated with lung function over 15 years later.