I am doing Cox regression models and KM plots for a data set where the end point is death. In addition there is information about whether the death was cancer-specific or not. So I have three categories:

  1. No death, last seen is the date of censor
  2. Death - cancer-specific
  3. Death - other cause

What I would like to know is what to do with category 3 data when I am looking at cancer-specific survival as my endpoint. Do I censor that data at the date of death OR do I just remove that data from the dataset?

  • $\begingroup$ Is your main research Q more about length of time until cancer or about the causes of the three different outcomes? If the latter, you might try a multinomial logistic regression. $\endgroup$ – rolando2 Jun 3 '11 at 14:49
  • $\begingroup$ Length of time before death after treatment is the main research question $\endgroup$ – danielsbrewer Jun 6 '11 at 10:53

Category 3 should be censored, not removed. Removing them would be similar to removing those in category 1 instead of censoring. The fact that those people were alive and did not die from cancer is useful information.

You should also look at all cause mortality in addition to the cancer specific. There is a chance that some of the non-cancer deaths were indirectly influenced by the cancer (increased stress leads to heart problems, believing they are going to die anyways leads to risky behavior, etc.)

Competing risks models could also be informative (just make sure that you understand what assumptions you are making).


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.