0

I am investigating covariate X for 8 cancer types which all have different survival rates and will possibly show different impacts of other covariates like age, sex etc. Therefore I will run 8 separate cox-ph regressions.

However, I doubt X will behave completely differently for each cancer type. If it has no significant effect for 6 types, that probably informs us it won't for the remaining two types and vice versa for a significant effect. So there should be a lower sample size than a regular power analysis * 8. But how can I incorporate this correlation to reduce the sample size?

kjetil b halvorsen
  • 77,844
CineyEveryday
  • 53
  • Many relevant Qs here: https://stats.stackexchange.com/questions/373890/separate-models-vs-flags-in-the-same-model, https://stats.stackexchange.com/questions/386877/how-to-interpret-manova-results-after-adding-gender-splitting-the-file/413140#413140, https://stats.stackexchange.com/questions/486373/is-there-a-benefit-to-splitting-the-data-by-gender-or-age-range-when-building-pr/486461#486461 and search this site! – kjetil b halvorsen Aug 21 '23 at 16:36
  • This is not a power problem. It is an estimation/precision problem. See https://onlinelibrary.wiley.com/doi/10.1002/sim.7992 – Frank Harrell Aug 21 '23 at 17:07

0 Answers0