Shared Versus Specific Effects of Treatment on Multiple Outcomes in Clinical Trials Using Latent Variable Modeling

• Activity Number: 449
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Social Statistics Section
• Author(s): Melanie Wall*+

Abstract

It is common in clinical trials to have multiple correlated outcome variables, yet typically a single primary outcome is chosen and traditional statistical analysis of treatment is performed. Further analyses taking account of the multiple variables may include analyzing each outcome separately or performing MANOVA. However, in some types of studies particularly in psychiatry it may be reasonable to consider the multiple outcome variables as indicators of a shared latent variable construct which itself is being affected by the treatment intervention. In such settings it may even be argued that the latent construct itself is the primary outcome of interest, rather than any one of the specific outcome variables measuring it. We describe a method to quantify the treatment effect on the latent variable and additionally identify how much of the treatment effect on the multiple observed variables is attributable to shared effects through the latent variable versus specific effects for each variable. An example is shown of several anxiety scales measured in a clinical trial for alcohol dependence patients with a broad spectrum of internalizing psychopathology.