4
$\begingroup$

I have some time to event data, but the population is only those who had the event (specifically, my cohort is all kidney tx recipients who were readmitted within one year of discharge for a specific event).

Since there is no censoring, what would be the pros and cons of using survival methods (both KM and Cox) vs. median regression (the time to event is highly skewed)? Also for the survival model - I know that this is the complement of the empirical distribution function, but does my interpretation of results change at all in regards to specific language when there is no censoring?

$\endgroup$
7
$\begingroup$

Survival methods are about modeling some time to event data. There is no need for there to be censoring! the methods will work and be more effective without censoring. Time to event data will probably not be well fitted by normal distribution models, so usual linear regression is not indicated. I say you should go with survival methods.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ I guess I'm just used to there always being censoring. I did go ahead with the survival methods, but my alternative I was considering was not mean regression but median (i.e. quantile at tau=0.5) regression which would take into account the non-normality of the time distribution. $\endgroup$ – Scott Jackson Apr 21 '17 at 16:50
  • $\begingroup$ I still think survival methods should be better here than quantile regression. $\endgroup$ – kjetil b halvorsen Apr 21 '17 at 16:55
  • 1
    $\begingroup$ Thanks for this. I guess one advantage of survival analysis is also that it can cope with time dependent features such as age? This is something (quantile) regression cannot cope with to my knowledge. $\endgroup$ – cs0815 Oct 30 '18 at 8:12
  • $\begingroup$ @kjetilbhalvorsen Thank you for clarifying that censored data is not necessary for survival models. However I wonder, what "time" should be entered for the cases, where the event did not occur and no censorship was present? (I use R, so Inf seems to indicate censoring, NA seems to indicate a missing value (which is not correct) and 0 seems to indicate the occurance of an event (which also doesn't apply)) Can you clarify? $\endgroup$ – stats-hb Jun 8 at 13:54
  • 1
    $\begingroup$ I just learned, that in these cases you give these cases a value for the number of times, you have observed the cases: socialsciences.mcmaster.ca/jfox/Books/Companion/appendices/… (page 5) $\endgroup$ – stats-hb Jun 8 at 15:25

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.