R help - Jul 2010 - Using coxph with Gompertz-distributed survival data.

Try using psurvreg(x, mean=fit2$coef, scale=fit2$scale, dist='extreme')

The psurvreg and qsurvreg functions inherit the same parameterizations
as survreg.  I have trouble myself matching the location/scale form of
survreg to default parameterizations used in the weibull, extreme, etc
distributions.

For a multivariate model the call above would use mean = 
sum(x* fit$coef), where "x" was the vector of values for a
hypothetical
subject.

An alternative is to use predict(fit2, type='quantile'), there is an
example in help(predict.survreg).

Terry Therneau

PS Your title was wrong, the quesion is about survreg not coxph. Also
note that the Gomperz and extreme value distributions are not exactly
the same: the Gompertz adds the restriction that x>0.  This changes the
normalization constant, but if most of the data is far from zero the
difference is very small.