Abstract:
|
Latent class and latent profile analysis are useful multivariate techniques for describing subgroups in study samples based on selected variables. These techniques are probabilistic in nature, where classes are described based on empirical evidence and interpretability. In this talk, we discuss how to apply these techniques using complex clinical research data. We discuss assumptions made in using latent profile analysis (LPA) on 200 clinical trial participants with a serious mental illness and diabetes. LPA in this sample explored differentiation between subgroups that were characterized on the basis of selected dimensions within a biopsychosocial framework. We identified two trivial subgroups (characterized by high and low scores on psychosocial measures) using a standard application of LPA. In making additional assumptions in line with clinical theory, we identified five meaningful subgroups. We also evaluated a secondary auxiliary model to describe relationships between latent classes and other clinical factors.
|