(Survival Analysis) How to perform a logrank test on two survival functions adjusted for some values?

[Dan Smith]

Hello guys. I have a table similar to this:

id	power	category	time
1	35	A	8
2	50	A	34
3	75	B	17
4	50	A	11
5	60	B	50

After setting my survival data with:

sts time

I want to compare the 2 survival functions of the items of category A and B, adjusted for Power = 45

I can do it grafically, typing:

generate power45 = power–45

sts graph, strata(category) adjustfor(power45)

Or, as an alternative:

sts graph, by(category) adjustfor(power45)

My question now is: how can I perform a test on the equality of these 2 adjusted survival functions for know if they are statistically equal or not?
I can't figure it out, I'm pretty sure sts test can't help...

Thank you

[Steve Samuels]

See sts test, which has both logrank and strata() options.

[Dan Smith]

I tried:

sts test category, strata(power45) logrank

but it provides exactly the same output of

sts test category, strata(power) logrank

Is that what you mean Steve?

[Steve Samuels]

Sorry- I overlooked your desire for an adjusted analysis. To do that, use stcox. The score test for a single indicator (two-group) variable is equivalent to the log rank test, but the Wald test for the indicator (Z statistic) tests the same hypothesis and is close. To adjust for power45, add it as a covariate. I recommend the option for robust standard errors. The sts graph commands should have alerted you to possibly non-proportional hazards in the two categories, and you can model this in stcox with the tvc and texp() options--see page 129 of the Stata 13. Manual. To really see what's going on, don't settle for a linear term in power45; use fp to detect a non-linear effect on the log hazard. Also check for interactions interactions of category and power45.

Code snippets

 stcox  category power45, robust

or

stcox  category power45, robust tvc(category) texp(log(_t))
test [main]category [tvc]category texp(log(_t))

[Steve Samuels]

Correction to the test statement for assessing simultaneously the main and tvc interaction affects of category.

stcox  category power45, robust tvc(category) texp(log(_t))
test [main]category [tvc]category

I apologize for the error.