Purpose: The aim of this study was to utilize bootstrapping to investigate the robustness of latent class trajectories and risk factors of depressive symptoms among mothers of children with epilepsy.
Methods: Data were obtained from a national prospective cohort study (2004-09) of children newly diagnosed with epilepsy and their families in Canada (n = 339). Latent classes of depressive symptom trajectories were modeled using a semi-parametric group-based trajectory modeling approach. Multinomial logistic regression identified risk factors predicting trajectory group membership.
Results: Four trajectories were identified: low stable, borderline, moderate increasing, and high decreasing. Goodness of fit, posterior probabilities, and parameter estimates obtained with bootstrapping were not significantly different from the original sample. Calculation of the root mean square error demonstrated minimal non-ignorable bias for three parameter estimates, which was subsequently removed with additional sampling. Risk factors identified were identical for the original sample and the bootstrap, and differences in odds ratios, as calculated with the method of variance estimation recovery, were not significant.
Conclusions: As examined using a bootstrapping procedure, group-based trajectory modeling offers a robust methodology to uncover potential heterogeneity in populations and identify high-risk individuals.