Depressive symptoms in mothers of children with epilepsy: a comparison of growth curve and latent class modeling

摘要

Two common approaches for studying trajectories of change are standard growth curve modeling (GCM) and latent class growth modeling (LCM) (Singer and Willett, Applied longitudinal data analysis. Modeling change and event occurrence. Oxford University Press, New York, 2003; Nagin, Group-based modeling of development. Harvard University Press, Cambridge, 2005). The objectives were to compare the results obtained using GCM and LCM in modeling trajectories of depressive symptoms in a sample of mothers of children with epilepsy; compare the methods in predicting average trajectory and individual trajectories, and; provide general guidelines for implementing these approaches. Findings from the two modeling strategies were different: GCM suggested a quadratic change in depressive symptoms over time. Addition of the time-varying covariate, family functioning, produced a final model that explained 25, 20, 31, and 18% of the residual intra-individual, as well as inter-individual variation in the intercept and slope (linear, quadratic), respectively. Results from the LCM suggested five distinct trajectories of depressive symptoms: low stable (30%), sub-clinical (39%), moderate decreasing (15%), moderate increasing (9%), and high decreasing (7%). Adding the family functioning variable resulted in a model that replaced the sub-clinical trajectory with borderline and moderate decreasing with high increasing. Both the GCM and LCM adequately described the average trajectory of maternal depressive symptoms with signed differences of 0.61 and 0.75 and −2.39 and −2.54 for the unconditional and conditional models, respectively. There was considerable variation in capturing individual trajectories. For approximately 14 and 9% of individuals, both models under and overestimated depression scores by at least five points. Although GCM and LCM perform equally well in predicting average and individual trajectories of change, they are used most efficiently under different circumstances. Where individuals are expected to share a homogeneous trajectory, GCM should be used; however, where individuals do not follow a common trajectory, LCM is more appropriate.