I am doing survival analysis on a dataset where the number of right-censored individuals is much greater than the number of observed individuals. There are about 50,000 individuals and only about 5,000 of them are observed. So as you might imagine, the survival curves are not that informative because they never go below about 80%.

So I was wondering what the implications would be by looking at only the observed data (i.e. the 5,000). I would imagine that this would artificially lower survival probabilities on the curve. However, if you are only interested in the relative difference in survival curves for different groups (I am using a regression-based SA model, where the different groups are defined by different values of the covariates) would it matter?

  • 3
    $\begingroup$ Why does the fact that the survival curves never go below 80% mean they aren't informative? If there is a big gap between them, that would inform. $\endgroup$
    – Peter Flom
    Dec 12, 2013 at 16:29
  • $\begingroup$ right but you cannot answer questions like when do you expect 50% of patients of type x to die (with much certainty) $\endgroup$ Dec 12, 2013 at 20:45

1 Answer 1


Only using the observations where subjects failed will bias the data and artificially lower survival probabilities.

Now, you're asking if it will bias estimated group differences if you only look at subjects who failed. It won't bias group differences if failure is independent of group. However, I'm assuming you think failure is dependent on group (and that's why you're studying it).

I would look into the Cox Proportional Hazards model and see if you can use that; it is meant for investigating differences between groups in survival data.

  • $\begingroup$ I agree that you absolutely must not ignore the individuals without observed events. With 5000 observed events you should have a good chance of detecting any important differences among your survival curves even if none go below 80% or so. This is similar to the situation in analyzing survival in better-outcome types of cancer. Patients do care about the difference between, say, 95% and 75% chance of survival. $\endgroup$
    – EdM
    Dec 12, 2013 at 18:45

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.