UBC Theses and Dissertations
An evaluation of two existing methods for analyzing longitudinal respiratory symptom data Arrandale, Victoria Helen
Due to the complexities of analyzing repeated binary outcomes, changes in respiratory symptoms over time are rarely studied. In fact, most respiratory epidemiology studies to date have not taken full advantage of longitudinal symptom data. This thesis evaluated discrete mixture models (SAS® Proc Traj) and generalized linear mixed models (SAS® Proc Glimmix) with respect for their applicability to six basic respiratory symptom research questions. These methods are both capable of handling repeated binary outcome data and permit inclusion of time varying covariates. These two techniques were then applied in a case study. Results from the evaluation of the methods indicated that Proc Glimmix can model the predictors of respiratory symptoms as well as population trends in symptom reporting over time. But Proc Glimmix is not suitable for modeling pattern or shape of change over time. In contrast, Proc Traj models patterns of change over time, and identifies multiple subgroups within the population. Proc Traj is not capable of modeling overall population trends. Both methods have statistical limitations that researchers need to understand; to help with this a simple guide describing both techniques was compiled. The case study utilized longitudinal data from a population of marine workers and focused on the outcome breathlessness, or dyspnea. Results from both Proc Traj and Proc Glimmix models indicated that the probability of reporting dyspnea changed over time in this population. Proc Traj models identified two distinct patterns of change in the population (one increasing over time, one steady over time). Proc Glimmix models identified several factors that were associated with dyspnea reporting; older age, childhood asthma, smoking and being female were associated with more dyspnea, whereas better lung function and current exposure to respiratory irritants were associated with less dyspnea. The overall conclusion was that both Proc Traj and Proc Glimmix models are suitable for analyzing repeated binary respiratory symptom data and researchers are encouraged to consider their use. Proc Glimmix is best for modeling the predictors of reporting a symptom at the population level, while Proc Traj is suited for modeling multiple subgroups in the population and their patterns of change over time.
Item Citations and Data