On Jun 15, 2016, at 9:12 AM, takiy berrandou wrote:

Hello,

thank you for the answer. it's true i was'nt clear enough.

the problem with glm()/gam or coxph() when the spline terms are used (specialy spline of 3rd degree) the calculated coefficients for the spline terms are difficult to interpret.

Not just difficult, but also generally unwise to attempt interpretation of individual spline coefficients.

for example for spline of degree=3 and 3 knots, there are 7 beta's calculated, and to represent the result clearly we need to calculate OR/HR against a reference.

Natural splines are third degree, although generally the number of knots is 1 less than the degree of the polynomial.

just making exp(beta) does'nt have a propre meaning.

On SAS there are some MACRO's like %regspline which help to make some calcul to figure out the OR's.

The predict.* functions will generally handle that difficulty naturally. Each of hte regression functions will have a specific `predict` method for the class of model object. You simply provide the values for the covariates in a dataframe and a prediction for for total of associated terms will be returned. See:

?predict.glm

Std errors are generally available if the proper parameters are offered to the function. This procedure should be exemplified in the help documents and vignettes of the packages with whatever regression function you anticipate using. On the Rhelp mailing list you are requested to present questions with sufficient code to support a coding response.

--

David

Thank's again for the answers

2016-06-15 17:56 GMT+02:00 David Winsemius <

dwinsemius@comcast.net>:

On Jun 14, 2016, at 7:43 AM, takiy berrandou wrote:

Hello,

I'm looking for a package or function, which calculate OR/HR from spline

model regression. for example in SAS it exist some MACRO helping to do that

easiely.

i had make some research on the forum here and on the web but without any

succes.

It's not yet clear what you want to do. Odds ratios are easily calculated from logistic regression models (typically constructed with the glm function the stats package) and hazard ratios are easily calculated from survival models (typically constructed with survreg or coxph in the survival package). All those functions accept spline terms in their formula versions. There are many worked examples that could be found with google searching since google now properly interprets the letter "r" as referring to the computer language. You can get a more focussed search using rseek.com.

David Winsemius

Alameda, CA, USA

--

Takiy BERRANDOU

0618916037

takiyeddine at gmail.com

David Winsemius

Alameda, CA, USA