Length of stay over multiple hospitalisations with competing risk of death

I’ve a large data set with many variables of older peoples, many of whom have multiple hospital admissions over a year. My aims are with a multivariable model to:
(i) understand which variables are associated with the cumulative length of stay of all admissions within a year &
(ii) predict the cumulative length of stay.

Unfortunately ~25% of subjects die within one-year.

I’ve tried going down the “days alive out of hospital” path - the QQ plots are horrible and it is not giving me quite what I want.

I can recode by (eg) trichotomizing the cumulative LOS outcome and construct six outcomes (three each for those who survived a year and those who didn’t) and then construct a multinomial model. However, I am not satisfied with the loss of information this way.

Any ideas?

One approach could be a multistate model with in hospital and out of hospital as states with exits to each other and death, and death as an absorbing state. Hein Putter’s work in mstate https://cran.r-project.org/web/packages/mstate/vignettes/Tutorial.pdf


Check that this topic was not already started elsewhere on the site. Then please move your topic to a reply to that one.

Thanks Martin. I’ll work on this. I don’t have new data at each potential transition, but may model as transitions each entry and each exit from hospital as well as mortality.

1 Like

Thanks Frank, I did look before I posted, but may not have used all the right terms! There are some discussions which nearly, but don’t quite fit.

1 Like