On Modeling Longitudinal Pulmonary Function Data

In this paper we have discussed how random effects can be included in linear models to accommodate the analyses of longitudinal pulmonary function (REM) and categorical (GEE) types of respiratory data. It was suggested that REM analysis be used for continuous observations that are normally distributed or that can be transformed to have near normal distributions and that GEE be used for categorical or non-normally distributed data. Two methods were reviewed, parallel plots and within-subject regression fitting, which can assist in determining the order of random effects to be included. Using a sample data set of longitudinal FEV1 measures, we outlined the steps that should be taken for selecting the within-subject error structure, the order of random effects, elements of the between-subject covariance matrix, and selecting the most important or predictive fixed effects. Lastly, two different types of residual plots were illustrated, conditional and marginal, which can be used to detect outliers and possible trends from underfitting the observed data. All options discussed herein are not available in all the programs currently available for doing the REM and GEE modeling.