0
$\begingroup$

In my research I want to analyze survival of dengue patients. I have data from 2007_2010. I'm interesting analyzing time to death. As survival time, I took the difference in between admission and discharge time of patients. But some of them died, some of survived up to that period and some of them are not discharged in that period. Since we want to use a censoring indicator to fit survival model, how can I handle the censoring indicator with the patients who have survived from that period?

$\endgroup$
1
$\begingroup$

My 2 cents,

  1. You could use the discharge date as the censoring date. For those who are still in the hospital, use the last day as the censoring date. The important part in defining censoring is non-informative. So if a patient who stays at hospital for longer is prone to die, the above definition is no longer non-informative. You may need one extra covariate in your survival model in such case.
  2. competing risk is not applicable here. No events, like discharge, will stop us observing death.
$\endgroup$
  • $\begingroup$ I think I misunderstood the question. I see how it is not a CR issue. I'm going to cross that answer out. Thanks for pointing it out! $\endgroup$ – RayVelcoro Sep 18 '15 at 20:49
1
$\begingroup$

If I understand the question correctly, you have two options:

  1. Rephrase your question to ask about death or survival. In this way, you can define two groups: Those who died and those who are either still in the hospital or have been discharged

  2. Set it up in a competing risks setting. Your competing risks would be death, discharge and survive, discharge and died, no discharge. As purewater pointed out, this is not applicable in this situation

$\endgroup$
0
$\begingroup$

In my view, you have two options.

First, you can censor the discharged patients at the discharging times (make the same status indicator as for the censored patients). However, this leads to dependent censoring. Then you should use something as IPCW (inverse probability censoring weights). This is not necessarily easy to do. There are a few R packages which might be able to do this, but I do not have any experience with them.

The second choice is to rephrase your question. As purewater said, you can not use a competing risks approach here... unless you are willing to change the research question. Instead of time to death, you can ask the question of time to death in hospital. Then I think you can use a competing risk approach, but it remains to you to see if this is a relevant question or not.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.