Friday, February 03, 2006
Re: st: Cox PH question
On 3 Feabh 2006, at 20:09, Ricardo Ovaldia wrote:
> A reviewer of a manuscript that we recently submitted > to a top medical journal stated that the "independent > variable, known to be positively skewed , be > log-transformed prior to inclusion in a Cox > proportional hazards model". As far as I know there is > no normality assumption for the Cox model, > additionally transforming this complicates the > interpretation of the reported hazard ratios. Am I > missing something or is the reviewer wrong? Other than > for interpretation, is there another reason to > transform an independent variable in the Cox PH model?
The distribution of the variable will affect the ease of interpretation of the hazard ratios. Given the distribution, a hazard ratio associated with an n-fold increase in the predictor variable might make more sense than a 1-unit increase.
My own hunch is that a variable with a funny distribution might better be re-expressed in, say, deciles or quintiles. This makes the hazard ratio easily explainable. You can also do this for variables on different scales, making your hazard ratios comparable across model terms. And quantiling will deal with zero values, otherwise a tough problem on a log scale.
Ronán Conroy email@example.com
* * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/