survival analysis: non-proportional hazards model with time varying covariates and multiple outcomes

[Lok Man Tong]

Hi all

I need to conduct survival time analysis on the dataset which contains 15 variables, of which 10 variables are time dependent and there are 5 multiple outcomes. I cannot employ the Cox proportional hazards model as I cannot pass the proportional hazards assumption tests.

May I ask for advices about which models are suitable? Thank you.

[Carlo Lazzaro]

Lok:
do things improve if you apply the -strata- option with -stcox-?

[Lok Man Tong]

Hi Carlo,

Thanks for your suggestion. Strata option limits 5 variables, but I have around 10 time varying variables. So, it cannot help. Does "tvc" option help?

[Carlo Lazzaro]

Lok:
you may want to take a look at stcox PHassumption tests entry in Stata .pdf manual (Examples 1 & 2).

[Steve Samuels]

Remarks and Examples in the manual entry for stcox has sections about g discrete and continuous time-varying covariates. The tvc option is designed for the latter.

Analysis of multiple failure types is described in Mario Cleve's Stata FAQ. If the failure types are "competing", then you will need a competing risks analysis. In Stata, this is done with stcox and stcrreg. For single outcome flexible parametric models, see stpm2. For competing risks, see stpm2cif and stcrprep. These last three commands can be found at SSC.

Your description of your dataset appears be incomplete, as you do not include a variable to indicate whether an event is censored or not. If you haven't already done so, be sure to read the Statalist FAQ, particularly FAQ 12. See also William Lisowski's excellent presentation about the FAQ 12 here.