[R] cph/nomogram Design/RMS package hazard ratio: interquartile vs per unit

[David Winsemius]

On Oct 20, 2011, at 8:22 PM, renee wrote:

> Hello,
>
> I am constructing a nomogram using cph and nomogram commands in Dr.
> Harrell's Design/RMS package. The HR that I obtain for dichotomous and
> categorical variables are identical to those that I obtain using STATA
> stcox.

When posting to r-help it is advised to produce the code used. One  
might guess that you were talking about output from summary(fit) but  
that would be a guess.

> However, the inter-quartile HR I obtain for continuous variables is
> obviously different, since STATA gives me HR for each unit (year,
> centimeter, etc) like coxph would give.

If you want HR's for single unit difference on the scale of the  
measured units then this should produce those:

exp(coef(fit))

> My question is if this will effect
> the output of the nomogram. I'm assuming that nomogram is  
> constructed using
> hazard between each unit rather than quartiles - is this true?

Yes.
>
> Also, I've found that I do not need to create indicator variables  
> for my
> categorical variables when I use cph. Is this also correct?

If they are factor classed variables, then that is correct.
>
> I appreciate your feedback. Thank you.
>
> ~Renee
>
> --
> View this message in context: http://r.789695.n4.nabble.com/cph-nomogram-Design-RMS-package-hazard-ratio-interquartile-vs-per-unit-tp3923896p3923896.html
> (At least until Nabble deletes it in a year or two.)

-- 

David Winsemius, MD
West Hartford, CT