predict/model.matrix for aftreg objects

[adibender]

So I've read multiple times now, that eha doesn't provide functions for prediction for aftreg objects. Is there a specific reason for that? What would be a straight forward way to obtain the model matrix, variance-covariance matrix, perdictions, etc. in case of left-truncated data?

[adibender]

Ok, just realised,

• model matrix can be obtained by specifying x=TRUE in the call to AFT and
• variance is stored in the var element of the aftreg output.

So, what remains is, why no prediction function? Is it somehow not usefully defined for models with left-truncated data?

[goranbrostrom]

Answers:

First question: No. Second question: I don't know (but see your own answer, or the code). Third question: I haven't found it useful in my own research. Fourth question: It certainly would be useful in some applications.