Non-linear association for proportional hazards models

[mmazariegos]

Dear all,

I am trying to assess the association between breastfeeding and risk of type 2 diabetes, so far the association appears to be nonlinear. Do you know how I can assess the non-linear association when using a Cox proportional hazards regression model?

Thanks,

Monica 

[ballardw]

It might help get a more targeted response by showing the code and the output you used and to identify the non-linear association.

Some description of the input data, such as a few rows of example data, might also help.

[mmazariegos]

Hi Ballardw,

I am using the proc phreg for Cox regression model,  the time variable is "follow_timedx_yrs" (person-years), the case is "definicion11_1" (values 0 and 1) and the exposure are the dummies for lactationcat.

Below the code to run the unadjusted model and the output: 

/*Unadjusted model*/
proc phreg data=diabetes2;
model follow_timedx_yrs*definicion11_1(0) = lactationcat_1 lactationcat_2 lactationcat_3 lactationcat_4/rl;
run;

Attached the output. 

 I see that there is a non-linear association, I mean I would expect to have decreased risk for diabetes for the longer duration of breastfeeding, but this is not the case.  

Some time ago I run regression models with restricted cubic splines using the code of Dr. Rick Wicklin but now I don't know how to do it when using Cox regression models.

Thanks, 

Monica