|
Abstract:
|
In observational studies frequently confounding-by-indication complicates analysis and interpretation of relationship between treatments and outcomes. For example, patients with a more severe form of a disease are at a higher risk of mortality, and they are often exposed to higher doses of certain treatment as well. In order to study the independent relationship between dose and mortality, the researcher must consider the possibility of time dependent confounding: Physicians use previous markers for severity to decide on subsequent drug dose, and drug dose last administered affects the next markers assessment. Here we consider different methods to approach the time dependent confounding problem, including propensity scores, time dependent Cox regression, restriction analysis, marginal structural models, structural nested models, Instrumental Variables, etc.
|
