st: Fwd: Proportional hazards violation in univariable v. multivariable analysis

[Shervin Shirvani <shervin.shirvani@gmail.com>]

Hi Statalisters,

This is a statistical query but I hope you may be able to provide some guidance.

I have a cancer data-set involving approximately 15 covariables and I
am interested in evaluating the effect of treatment assignment on
overall survival. The treatment assignment covariable is categorical
with 3 possible values.

As it happens, in unadjusted kaplan meier plots, the treated and
untreated curves cross, resulting in a violation of the proportional
hazards assumption. This is confirmed by a significant p-value using
-estat phtest- following -xi: stcox i.treatment-.

However, using -estat phtest, detail- on the final multivariable Cox
regression model, I note that the p-value for the treatment covariable
is no longer significant.

My question is: Is there a need to address proportional hazards in the
multivariable analysis (ie with a time-varying covariable) based on
the above findings?

Thank you very much in advance.
*
*   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/