B0956
Title: Combining latent class growth models and marginal structural models to estimate the effect of treatment trajectories
Authors: Denis Talbot - Laval University (Canada) [presenting]
Awa Diop - Universite Laval (Canada)
Caroline Sirois - Universite Laval (Canada)
Jason Guertin - Universite Laval (Canada)
Bernard Candas - Universite Laval (Canada)
Abstract: Latent class growth models (LCGMs) are becoming increasingly popular to summarize a time-varying treatment in a few trajectory groups. Standard approaches like a confounder-adjusted regression model or an inverse probability of trajectory groups weighted regression can yield biased estimators of the effect of the trajectory groups because some variables can have a double role of confounders and mediators. We propose to combine LCGMs with marginal structural models (MSMs) to adequately control for time-dependent confounders. The parameter of interest is defined as the projection of the true MSM onto a working model characterized by the LCGM trajectories. Inverse probability of treatment weighting, g-computation and targeted maximum likelihood estimators are proposed. These estimators are evaluated and compared with standard approaches using simulation studies. We also discuss an extension of the LCGM-MSM approach where the trajectory groups can be time-varying using a history-restricted MSM framework. The motivation was a real-world reevaluation of the efficacy and safety of statins for the primary prevention of cardiovascular disease among older adults using population-wide administrative health databases. The application of the LCGM-MSM approach in these data will be illustrated.
