I am investigating the effect of genomic dysmethylation on cancer survival time, with data from multiple different cancers with very different survival curves.

Normally, I would split the cases into high and low dysmethylation, and do a log-rank test on the Kaplan-Meier curves. This isn't possible because for each cancer I have only 5 or so cases where methylation was measured.

However, for every cancer, I do have 100+ cases where survival time was measured, even if methylation wasn't.

Can I do a statistical test where I measure the survival times of the cases where methylation was measured against this "background" survival curve?

Is it possible for me to combine the different groups, so I get a single answer across all the different cancers?




1 Answer 1


You could use relative survival methods to address your question. The observations without methylation measured can be used as the population for the expected survival, and then the survival of subjects with methylation data can be modeled after adjusting for expected survival. The actual implementation depends on your software.

  • $\begingroup$ Thanks, I'll try that. The relative survival methods I've looked at so far require that the background distribution must be almost free of the cause of death in the sample, but I'll keep looking. $\endgroup$
    – Steph
    Sep 3, 2014 at 17:45
  • $\begingroup$ The cause of death only changes the interpretation. You would not be using the expected survival as a comparison to determine whether the rate of events in the data is higher than in the reference population, but only as an appropriate baseline for each cancer type. $\endgroup$
    – Aniko
    Sep 3, 2014 at 18:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.