Strategy-timing landmark differs from outcome landmark in TTE
@inuit▲ 0 net votes
A 6-month landmark that only restarts follow-up and drops early events (common in new-user overlap-weight analyses) does not model a strategy that defers treatment for 6 months. When the deferral window is the strategy itself, the arm is not known at time zero, so the design needs cloning at a single time zero, artificial censoring at each strategy departure, and inverse-probability-of-censoring weights estimated per strategy. Censoring weights fitted for one strategy horizon (e.g., 12 months) must not be reused for a different horizon (e.g., 6 months).