Statistical modelling of longitudinal lung function data

Abstract

Statistical models were developed for the analysis of longitudinal data obtained from two different studies. For the first longitudinal study, our objective was to evaluate different longitudinal models for predicting the longitudinal decline in lung function measures of grain elevator workers and to assess the goodness of fit of these models. Generalized estimating equations and maximum likelihood methods were used to fit different models. Concordance coefficients 'rc' and 'r'(ώ) were used to assess the adequacy of the model and variance-covariance structure respectively. Pseudo-likelihood ratio test, ƛ was used to test the null hypothesis that the assumed covariance structure is equal to the true covariance structure. An important finding from the random effects models was that there might be more observational error in measuring FVC than FEV1. Another longitudinal analysis was conducted to study the respiratory health effects of initial exposure to grain dust among workers commencing employment in the grain industry in the Province of Saskatchewan. Correlated survival data analysis techniques ere used to determine predictors of bronchial hyperresponsiveness. Consistent estimates of standard errors were obtained by using jackknife, bootstrap and the method proposed by Wei, Lin and Weissfeld (WLW). We conclude that survival analysis is a useful technique to analyze the bronchial hyperresponsiveness data. The estimates of standard errors were very similar for jackknife, bootstrap, and WLW, but different from those obtained using standard likelihood maximum methods. Cox's proportional hazard model based on the data from first longitudinal study, proved to be a useful technique in investigating the relationship between survival time (time to first episode of wheezing) and possible prognostic variables.