st: RE: Checking goodness of fit of stcox when using the tvc() option

["Simmons, Nicole" <nsimmons@jhsph.edu>]

I am using stcox in Stata12.

I am running a cox regression model on a large survival data set with over 30,000 observations and 9,000 failures.  Because the proportional hazards criterion was violated with almost all covariates and the deviance residuals were skewed, I am running a model in which a number of covariates are interacted with ln(_t) using the tvc() option.  

I have obtained a model that appears good and seems to fit the data well using the AIC criterion.  I would like to look at the deviance, martingale and Schoenfeld residuals to see if I have successfully fit the data with this model. However, the Stata 12 documentation says these options are not available using tvc.  I tried to use the other option of splitting the data at failure times but that succeeded only in crashing Stata, I gather because of the large number of failures.

Is there any other way to check the goodness of fit of my survival model with interactions with functions of time, or to visually examine the fit of the predictions?

Thanks
Nicole Simmons
Johns Hopkins University
nsimmons@jhsph.edu

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/