Time dependent hazard ratio estimation using instrumental variables without conditioning on an omitted covariate

摘要

Confounding is a threat to all observational studies. The factors that influence the outcome of interest may also influence the selection of treatment. In randomized studies, intention-to-treat estimators are generally consistent estimators of the intention-to-treat effect but are not consistent estimators of the treatment effect. Instrumental variables (IV) influence treatment choice but otherwise have no effect on the outcome (exclusion restriction) and are independent of any other causes of the outcome (randomization assumption).