Controlling for Time Dependent Confounding using Marginal Structural Models to Assess Epoetin Alfa Dose Relationship with Mortality
View Presentation View Presentation
*Ouhong Wang, Amgen, Inc. 

Keywords: Time Dependent Confounding, Marginal Structural Models, Inverse Probability of Treatment Weighting

A number of recent publications have studied the link between mortality, hemoglobin and Epoetin alfa (EPO) dose in the dialysis patient population using observational data. Most of these studies have not attempted to consider EPO measured repeatedly over time whilst simultaneously adjusting for the time dependent confounders, e.g., hemoglobin. Similar problem occurs frequently with observational databases in naturalized settings. Marginal Structural Models (MSM) are designed to use inverse probability of treatment weighting to adjust for this type of confounding. Under the assumptions of no unmeasured confounding and model misspecification, the parameter estimates from the MSM have causal interpretations. Here we apply MSM to a large dialysis provider database to examine the relationship between EPO dose and mortality. Sensitivity of the MSM to model specifications will be presented.