Global effects estimation for multidimensional outcomes.

Many health studies focus on multifaceted outcomes that are not easily measured with a single variable; examples include studies on quality of life (QOL) and general health. To fully explore such an outcome, researchers typically collect information on multiple endpoints. The resulting measurements constitute multidimensional outcome data. An object of great interest is the overall-or global-effect of a covariate, such as a treatment intervention, on the multidimensional outcome. Quantifying such an effect can be difficult because multiple clinical outcomes are usually measured on different scales; the problem is enhanced by the fact that multiple measurements on a given subject are typically correlated. We present a regression modeling scheme permitting estimation of global treatment effects when multiple continuous endpoints are examined in concert either once or for several times. The global effect is conceptualized as a change in the distribution functions of the outcome variables. It may thus be interpreted as a connection between outcome distribution quantiles for the treatment and control groups. This concept allows the presentation of a global effect as a scalar quantity applicable to all outcomes simultaneously, easing interpretation of results. Model estimation proceeds directly from existing methods for multivariate survival analysis. The assumption that the treatment effect is homogenous across different outcomes is testable. To illustrate the application, we present data analytic results from a motivating example, an analysis of patients' QOL during recovery from lower limb trauma. We also explore the performance properties of global effects estimation through simulation.