[R] Plot of coxph tt effects

[David Winsemius]

On Nov 1, 2013, at 11:16 AM, cesar garcia perez de leon wrote:

> Dear all, 
> 
> We are conducting a study in with a set of covariates and a time to event outcome. 
> Covariates b1 and b3 violate proportionality.

Can you describe the basis for that statement?

> We applied a coxph with a „tt‰ term to evaluate the nature of time dependence. 
> 
> Call: 
> coxph(formula = Surv(start - 1, stop, outcome) ~ tt(b1) + 
>    b5 + tt(b3) + b4, data = data, tt = function(x,  t, ...) pspline(x + t/90)) 
> 
> We would like to plot the predicted effects. Following the logic of the tt function, does it make sense to plot the residuals against the results from the model with tt for the upsetting covariates? 
> 
> As far as we know the there has not been updates of survfit for the tt  function. So, we are unsure as to how to plot our curves. 
> 
> I would appreciate if someone can give me an advise or a publication reference that can help us in this matter. 

Is there a reason not to use `cox.zph` and it's `print` and `plot` methods for this purpose? I think it is already offering what you are trying to re0invent.

-- 

David Winsemius
Alameda, CA, USA