In this article, we address statistical techniques appropriate for examining longitudinal changes in cancer symptom clusters. When the cluster structure is not pre-determined, researchers may examine symptom clusters either at each time point or use composite scores to examine the symptom clusters across time points. When the cluster structures are pre-determined, the statistical techniques depend on the research assumptions or purposes. Multilevel modeling, generalized estimating equations, latent growth curve modeling, and multivariate repeated-measure analysis of variance are good choices for exploring whole cluster changes over time. Alternately, confirmatory factor analysis and path analysis are appropriate techniques for examining changes in symptom relationships within clusters over time. Each technique is described, with examples and strengths and weaknesses.
Keywords: cancer; generalized estimating equations; latent growth curve modeling; longitudinal designs; multilevel modeling; symptom clusters.
© 2014 Wiley Periodicals, Inc.