Statistical challenges in the analysis of dynamic traits: implications for pharmacogenomic clinics

Analysis of dynamic traits is statistically challenging for several reasons. Since most of the dynamic traits result in irregular sparse longitudinal measurements, a unified approach for jointly modeling the mean trajectories and the underlying covariance structure is essential. When the traits are bivariate or multivariate in nature, modeling the covariance structure is really challenging. For the pharmacogenomic clinics, it is extremely important to have a comprehensive study of the whole biological system. In other words, if the traits under consideration result in some events (e.g., death, disease), then a joint analysis is required for the observed dynamic traits and the event-time. In statistics, there is a vast literature on such joint modeling using parametric, nonparametric and semiparametric approaches. In this article, we will discuss different aspects of modeling the longitudinal traits, their limitations and importance to pharmacogenomic clinics.