I am attempting to use and illness-death multi-state scenario to model a time-dependent covariate in a competing risks analysis following Beyersmann et al. 2012 (Competing Risks and Multi-state Models with R, Springer New York, 2012). In general all subjects would start in the "healty state" = 0 with a possibility to transition into the "sick state" = 1 (the time-dependent variable). Either of these states may then transition into one of two competing absorbing states "discharge" = 2, "death" = 3. My specific question is can a patient start in either state 0 or 1 or do all patients need to start in state 0 for this model to work. My data consists of patients starting in both the healthy and the sick states at the beginning of observation. Thanks.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
If you have left-censoring your approach should correct for that. Kapetanakis et al. (2013) propose an approach how to do that (page 704-706).
See http://onlinelibrary.wiley.com/doi/10.1002/sim.5534/pdf.
