I have a small question. Usually in survival analysis, the event is something bad, e.g death, relapse of a disease, etc...Now what if I have data in which the event is positive, e.g., time to healing, where of course healing is defined properly. Can I use the same methods of survival analysis (KM, log rank, cox,...), or is there any difference ? What about the censoring mechanism ?


No reason not to do this. You'll often see time to response as an outcome in clinical trials. You can also have outcomes which are somewhat more neutral, e.g., treatment duration.

Censoring is the same, but you should consider whether censoring is actually uninformative (a requirement for KM, Cox) - if not you should use competing risks framework (same goes for negative outcomes).

  • $\begingroup$ so I just use the same methods, only that my survival plot will look like hazard ? (increasing function) $\endgroup$ – user3275222 Nov 8 '15 at 9:16
  • $\begingroup$ Your survival function will still be monotonic decreasing, but you can use the cumulative incidence function (1 - S) if you want something monotonic increasing. $\endgroup$ – tristan Nov 8 '15 at 9:29
  • $\begingroup$ how can it be decreasing ? if the event is positive (e.g. healing), than the probability of healing is getting better with time, or at least should be, if for instance the treatment is constant (not a one-off treatment). $\endgroup$ – user3275222 Nov 8 '15 at 9:34
  • 1
    $\begingroup$ The survival curve represents the proportion of the population which has not experienced the event (positive, negative or neutral). Survival is obviously a slightly loaded term which might be causing the confusion. Just treat it the same as if the event is a failure and if you want a graph that goes up then do 1-S. $\endgroup$ – tristan Nov 8 '15 at 13:22

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.