In a randomized clinical trial, measured and unmeasured confounders are balanced across treatment groups making it possible to draw causal inferences on a clinical outcome when all study participants comply with their assigned treatments. However, standard methods for estimating causal effect of a time-varying treatments can be biased when a time-dependent risk factor on an outcome also predicts subsequent treatment; the past treatment history also predicts subsequent risk factor level. There will be discussions on these issues in addition to various causal inference methods and analysis populations.