I have a dataset of 1500 patients for a time-to-event study. Almost half the population died at the moment of inclusion into the study which means that they have had a follow-up time of zero. Now I want to report the follow-up time which is around 5 days while the surviving patients were followed up for several years.

Is that a problem that needs solving? I have a feeling that this is not representative of my study population. I have looked at the reverse-kaplan-meier method but I dont know if that is applicable to my current situation.

  • $\begingroup$ I'm confused by the issue. A population with half the members dying instantly and the other half surviving for several years should have a mean follow-up time measured in years, not days. Statistics properly computed from a population should of course be representative of that population, but I don't understand what that 5 day number is supposed to represent. $\endgroup$ Jan 14 at 19:52
  • $\begingroup$ Due to the index event being cardiac arrest (which is associated with a high risk of death) the proportion of people who survive the event is very low. The rest may live for a short while and pass away then. This is how the 5 days came to be. The rest of patients have survived years without issue. I hope this clarifies any confusions. $\endgroup$
    – Fabian
    Jan 14 at 20:08
  • $\begingroup$ Why not subdivide your population, much as is done for life expectancy estimates where deaths during or immediately after birth are separated? In other words, report separately on those who died within a defined short period after inclusion. $\endgroup$
    – whuber
    Jan 14 at 20:35

You could perhaps report mean follow-up time among censored patients rather than all patients. One could argue that follow-up time among patients with an event really says more about whatever event you're studying rather than the design of your study. Your study was designed with the capability to track patients for several years, and the fact that many patients don't survive long does not change the nature of your experimental design. It sounds like you want to capture aspects of the study design itself, but the fact that many patients fail to live through the follow-up period doesn't indicate that you couldn't have followed them if they had lived, which the very short mean follow-up period from all patients seems to imply. By using censored patients only, you capture your ability to follow patients over time when not limited by the disease itself.


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.