2
$\begingroup$

II've got a bunch of data on hospital admissions and I'm trying to establish if the number of admissions somebody has had is associated with survival time. I also wish to control for a bunch of other covariates (age, sex, ethnicity...). The cohort consists of everybody who was admitted to a particular hospital for the condition of interest over the course of a 5 year period, and I have since obtained their full history of admissions for the condition. Now the question of interest immediately suggests that a Cox PH model is what I should be fitting. The main thing that I need to account for is that those with a single admission versus those with say, more than 5, are likely to have a shorter follow-up period if their first admission just occurred much later. I've been racking my brain to establish whether a) this is violating any assumption of the Cox model, and b) whether I will even have sufficient data/follow-up for the "single admission" group to be able to compare them to the multiple attenders. There is a lot of right censoring in the data however there is a much larger portion of those with more than one admission who have since died compared to the others.

Modelling time from last admission to death does not seem appropriate since the ones with a single attendance are in the study for a shorter period, hence a lot of censoring in this group, hence overestimation of survival (is this a correct interpretation?). For this reason I'm thinking of trying to fit the number of attendances as a time-dependent covariate and using the index admission as the baseline. I may also categorise the number of admissions into 3 groups.

Does this seem reasonable?

$\endgroup$
  • $\begingroup$ Few questions come to mind: (1) is everybody censored at the end of the 5 years period? (2) are the subjects selected only if they have at least one admission during the 5 year period? (3) if a subject is selected, then do you also have the information on admissions before the start of the 5 year period? (4) what would be the "time 0" of this analysis - birth, first admission, some diagnosis time? $\endgroup$ – Theodor Nov 10 '15 at 9:26
  • $\begingroup$ 1) There are 44 events in total, but everybody else is censored at the end of this 5 year period. 2) Yes, everybody has at least one admission as this is how the initially entered the cohort. 3) I have details of all admissions since diagnosis, regardless of whether these occurred during the 5 year "recruitment" period or not. 4) I was intending to use date of discharge of the baseline admission, since everybody has at least one admission. We are trying to get hold of full date of diagnosis which may be a more suitable choice, currently on year of diagnosis is available. $\endgroup$ – Ray Sconosciuto Nov 11 '15 at 10:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.