If tstop in a multistate model for an event of interest is day zero and tstart must be less than tstop (day 0). Where should I move cases that experienced an event at day 0? I am using mstate package
1 Answer
Section 3.3 of the R time-dependent survival vignette suggests the following for coxph
; mstate
extends the coxph
framework:
One subject died on the day of entry. However (0,0) is an illegal time interval for the
coxph
routine. It suffices to have them die on day 0.5. An alternative is to add 1 day to everyone’s follow-up, e.g., subject 2 who enrolled on Jan 2 1968 and died on Jan 7 would be credited with 6 days. (This is what Kalbfleisch and Prentice do in their textbook.) The result of the finalcoxph
call is the same from either strategy.
The question you have to ask/answer in the multi-state scenario is which state was in place before the event on Day 0. Alternatively, you could just define the starting state for such individuals as the state reached on Day 0. Which of these is best for your application depends on your understanding of the underlying subject matter.