Longitudinal analysis of respiratory symptoms in population studies with a focus on dyspnea in marine transportation workers

Abstract

Purpose: Longitudinal respiratory symptoms are rarely studied in occupational epidemiology. We investigated dyspnea change over time and predictors of change over time using two longitudinal modeling techniques, a semi-parametric group-based approach (SAS® Proc Traj) and a generalized linear mixed model (SAS® Proc Glimmix), and compared the two techniques for use in longitudinal studies of respiratory symptoms.

Methods: Data were previously collected from a lung health surveillance study of marine transportation workers. Subjects were seen two to four times over 12 years (1987-1999). At each visit the American Thoracic Society questionnaire was administered and lung function was tested. The semi-parametric group-based model and the generalized linear mixed model were applied to the data.

Results: The group-based trajectory model supported two groups of dyspnea change over time. Group 1 (73%) had a steady low-level probability of reporting dyspnea over follow-up, while Group 2 (27%) had an increasing probability of reporting dyspnea over follow-up. The generalized linear mixed model (random intercept) estimated that the probability of reporting dyspnea was increasing over time in the population. Current smoking, female sex, lower lung function and older age were associated with increased probability of reporting dyspnea in both models.

Conclusions: Results from both models indicate that the probability of reporting dyspnea was increasing over time in this occupational cohort. The group-based model is capable of identifying multiple patterns of linear and non-linear change while the generalized linear mixed model is preferable when the population mean change (linear) is of interest. Both approaches were able to identify similar characteristics associated with longitudinal dyspnea symptoms.