topic Cox proportional hazards model with time-dependent covariates in SAS Programming

Cox proportional hazards model with time-dependent covariates

Lars_Andersen — Wed, 04 Nov 2015 13:26:30 GMT

As part of an analysis (I can provide additional details if needed) I need to obtain the estimate of the linear predictor (i.e. the β’x part of S(t) = S0(t) ∙ exp(β’x)) for each time point for each patient from a cox proportional hazards model with time-dependent covariates. However, I cannot figure out how to do that in SAS when the model includes a time-dependent covariate. In a simple cox model, it is relatively easy to get by specifying “baseline” and “xbeta”:

proc phreg data = data;

model time*event(0) = variables;

baseline covariates = data xbeta = score;

run;

However, the “baseline” statement does not work when you introduce a time-dependent covariate. Does anyone know how I can obtain the estimates?

Much appreciated,

Lars

Re: Cox proportional hazards model with time-dependent covariates

BTAinRVA — Wed, 04 Nov 2015 15:46:44 GMT

Lars,

Have you seen this:

http://support.sas.com/resources/papers/proceedings12/168-2012.pdf

Brian

Re: Cox proportional hazards model with time-dependent covariates

Lars_Andersen — Wed, 04 Nov 2015 15:54:40 GMT

Hi Brian,

Yes, I did previously see that but it was not clear to me that it answers my questions. Do you suggest I use the "counting process" method?

Thanks,

Lars

Re: Cox proportional hazards model with time-dependent covariates

BTAinRVA — Wed, 04 Nov 2015 16:04:46 GMT

I would try the programming statement method first as it seems much simpler and gives the same results.

Re: Cox proportional hazards model with time-dependent covariates

Lars_Andersen — Wed, 04 Nov 2015 16:10:01 GMT

As noted in my original post, the programming method and "baseline" statement cannot be used for models with time-dependent covariates.