If I have disease free survival data (defined as whether or not a particular disease has been diagnosed or not along with the time to that event or loss to follow up) and also overall survival data, how do I deal with deaths that occur without the disease event? Are these censored or should I exclude such patients from the disease-free survival (dfs) analysis? I plan to run dfs analyses for several particular types of disease separately.

  • 1
    $\begingroup$ You can do both. I've even seen both types reported in the same paper. "all cause mortality" is one DV and "disease" is another DV. For the latter, deaths from other causes are censored. $\endgroup$ – Peter Flom Sep 30 '12 at 12:11
  • 2
    $\begingroup$ Censoring is perfectly ok, even though they're not "really" censored (you know that they won't develop the disease, as they're dead). Another approach is competing risks analysis, where you consider death as a competing event. If you're interested I can give you some references. $\endgroup$ – boscovich Sep 30 '12 at 13:24
  • $\begingroup$ @PeterFlom do you mean that deaths from the disease in question but undiagnosed in the patient before death should be included, or censored? $\endgroup$ – jetistat001 Sep 30 '12 at 17:18
  • $\begingroup$ @andrea I'd love some references but I'd prefer it even more if you gave a summary of them in an answer :) Perhaps a competiring risks analysis is uitable. $\endgroup$ – jetistat001 Sep 30 '12 at 17:19
  • $\begingroup$ It depends on the goals of the analysis. $\endgroup$ – Peter Flom Sep 30 '12 at 17:33

My interpretation of disease free survival is that the only event is diagnose of return of the disease. Any other event be it patient withdrawal from the study, lost to follow-up for any other reason or death is a censored event because at that time the defined "event" had not occurred and there is no way for it to either occur or for the investigator to ever find out if it occurred.

You should not remove patients that died. That creates potential bias. With survival the whole idea of censoring is to use the incomplete observations and not create bias that could occur if you threw out the incomplete observation.

In comparing treatments I find in agreement with Peter's remarks I have seen it done (and have done myself) analyses of time to recurrence (where death by other causes are censored) and all cause mortality. Death by disease specific cause is another way such data can be analyzed.

  • $\begingroup$ So, if I'm looking at disease A, then I should censor all deaths including deaths from disease A that were undiagnosed but are listed as the cause of death (this shouldn't happen but obviously it can happen if the disease goes undetected)? The only events I should include are diagnosis of the disease. Is that right? $\endgroup$ – jetistat001 Sep 30 '12 at 17:17
  • 1
    $\begingroup$ If death is from a recurrence of the disease that went undetected that seems to me to be a different story. If couldn't happen without the disease reoccurring. Then I think you should count the death as a recurrence of the disease. Of course the subject could not have been disease free until death, The recurrence must have happened earlier. But diagnosis of recurrence happens some unknown time after the actual recurrence anyway. $\endgroup$ – Michael R. Chernick Sep 30 '12 at 17:32

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.