Evaluating Time-Varying Service Implement on Longitudinal Outcome using Joint Analysis Approach
*Haiqun Lin, Yale School of Public Health
When treatment or exposure (such as health service usage) is time-varying longitudinal studies, standard methods for estimating the effect of the treatment may be biased when there exist time-dependent confounding variables. Robins' marginal structural models (MSM) are designed to adjust for observed and measured time-dependent confounding variables under the assumption of no-measured confounding. Our joint modeling approach relaxes such assumption. We first formulate a joint model that can account for unmeasured (as well as measured confounding variables) for the joint sequences of the time-varying treatments and the associated responses of each subject. The marginal estimate of the effect of the time-varying treatment can be obtained by just fitting weighted generalized estimation equation (GEE) model for longitudinal data using weights constructed from the joint models.