Growth curve mixture models

doi: 10.3969/j.issn.1002-0829.2012.06.009 Division of Biostatistics, Thomas Jefferson University, Philadelphia, Pennsylvania, USA *Correspondence: [email protected] Psychiatric studies often collect longitudinal data to characterize the natural history of disease in a cohort or to evaluate the effect of behavioral or pharmaceutical interventions. For example, in a recent partially randomized study comparing escitalopram and nortriptyline in the treatment of depression, several depression scales were measured weekly over the 3-month course of treatment.[1] While the primary outcome measure of such studies may be a binary indicator of improvement at the end of treatment, analysis of the full longitudinal profile that makes optimal use of all available data to model rates of change over time may be more informative. For example, in the escitalopram/nortiptyline study, analysis of dichotomous outcomes adjusted for time participating in the study showed no difference between drugs, while analysis of the longitudinal profiles did indicate different patterns of improvement in the two groups over time.[2]