st: LR test after multiple-record per patient data stcox

[Amit Achhra <amitstata@gmail.com>]

Dear All

The P-values from the likelihood ratio test generated using the
post-estimation user-written command 'reformat, lr' seems to differ
from ones I get performing LR test in a more usual way. I'm performing
survival analysis on multiple-record per patient cohort data after
issuing stset (with id option) and then stcox.

For example, my variable of interest is say X1 which has been
categorised at quintiles (i.e. 5 categories with 3rd as reference
category) for which I want an overall LR test P-value. There are three
other variables included in the model, say X2-X4. I'm stratifying the
model on another categorical variable X5. I'm using Cox regression.

(i) LR test by reformat:
qui xi: stcox i.X1 X2-X4, strata(X5)
reformat, lr

(ii) LR test in a usual way
qui xi: stcox i.X1 X2-X4, strata(X5)
estimates store A
qui xi: stcox X2-X4, strata(X5)
lrtest A

The P-value using (i) seems to be strongly significant as compared to
the one I get using (ii). I usually prefer reformat for convenience.

Also, both the methods display "n= total number of observations" which
is not the same as total number of patients.Please suggest if I'm
missing something? Is there any neater way to get the LR P-value after
such an analysis.

Many Thanks
Amit Achhra, Australia
*
*   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/