PH assumption: categorical variable

[Marko Deinders]

After checking the PH assumption I found one variable (4 categories, 1=ref.) to interact with time. The 3rd and 4th category had p-values <0.05. However, the 2nd category did not show a significant interaction with time (p=0.83).

How can I remove this time varying part (of the 2nd category) from the model?

I read something about making indicator variables for each category. http://people.upei.ca/hstryhn/nova11/srv_exc_3_sol.pdf
Maybe there are other suggestions?

Kind regards,

Marko

[Steve Samuels]

Welcome to Statalist, Marko! I suggest that you read the FAQ, especially section 12. If you don't show us the commands you used and the results (using CODE delimiters), it will be very difficult to advise you.

[Marko Deinders]

My question: How can I remove this time varying effect of _Ifi_2?

It should be: How can I remove this time varying effect from _Ifi_2?

If possible (and allowed), I want to report the HR for _Ifi_2 without the time varying effect
and for _Ifi_3 & 4 with the time varying effect.

[Steve Samuels]

You had the right idea: create indicator variables for each category, then omit the one for category 2 in the tvc() option. You can most easily create the indicators with the generate() option of tabulate:

tab fi, gen(fi)
des fi*
stcox fi2 fi3 fi4 $confounders, robust tvc(fi3 fi4) texp(_t)

Aside: If you have Stata 11 or higher, then you don't need xi; use factor variables instead. See the help for fvvarlist.

See the FAQ about cross-posting: if you post in another forum, link to that post; if you resolve your problem in either place, then you close the question in the other. You've already had a serviceable answer at Cross Validated http://stats.stackexchange.com/quest...ract-with-time.

[Marko Deinders]

Thank you for your answer!

See also: http://stats.stackexchange.com/quest...ract-with-time