0

This question is based in this other question that I asked Cox Model and proportional hazards , but are different things.

What I did first is check for which variables the proportional hazards assumption is met, then I saw that all the variables violated the proportional hazards assumption. In this case the alternative that I have is use the stratified cox model.

I have the following variables

  • Group: 1, 2, ..., 9

  • Sex: 1 female and 0 male

  • Weight

  • Age

Before make the stratified cox model, I run the Cox model, and the only significant variable was Group. I did it manually using Collet's model selection and after that I used the step() function in R, and the results are the same.

In this case I have just one significant explanatory variable, and this variable don't met the proportional hazard assumption. What I should do?

Community
  • 1

1 Answers1

1

If the proportional hazards assumption is violated, then the associated concept of the hazard ratio, typically assessed in Cox proportional hazards analysis, is squirrelly at best. To demonstrate that there are differences among multiple survival curves, like those for your various Groups, you could do a non-parametric extension of the log-rank test as provided by the survdiff() function in the R survival package. More comprehensive analysis might be possible with fully parametric survival models. As I mentioned in my answer to your related question, it might be wise to examine why certain Groups have hazards non-proportional to those of other Groups, a process that might be simplified by choosing another reference Group.

EdM
  • 57,766
  • 7
  • 66
  • 187