The role of biomarkers both as diagnostic and prognostic tools for risk assessment of a disease is emerging. Most experience has focused on relating biomarkers measured at one time point with subsequent outcomes. However, there is an evolving interest in the effect of temporal change in biomarkers over time. In such cases, the statistical challenge is to model the longitudinally measured predictor variable against an outcome variable. The most common practice in the literature can be generalized as a two-stage model in which a measure of the change is first estimated and then used as a predictor variable to be modeled in the second step. Here we propose an alternative modeling approach that integrates the two stages and primarily estimates the association between underlying temporal change and the outcome. Using simulations and application to real data, we demonstrate that our approach results in more consistent and efficient estimates of the measure of variable associations.