Cox Regression with Age as Timescale and Age varying variables

[Dana Cullen]

Hi,

I'm trying to fit a model to understand the effect of a drug types for a condition on survival. I'm using stcox and my analysis is for older ages (60-85 at baseline) with at most 6 years of follow up. Because in old age, the effect of a drug starts to diminish as age becomes the predominant force on mortality, the hazards are not proportional. To allow for this I have thought that I can run separate cox models for the current age bands for example:

stsplit current_age, at (60,10,100)
stcox i.drug if current_age<70
stcox i.drug if 69<current_age<80
stcox i.drug if 79<current_age<90

I thought this was sensible because I have 7 other predictors (bmi, sex, comorbidity status ect.) that also vary by age band. However, could I not also include a time varying interaction?

Would the model then be specified as:

stcox i.drug#i.current_age

or does it need to be:

stcox i.drug##i.current_age

or:

stcox i.drug i.drug#i.current_age

As age is my timescale, I don't think the second option makes sense, because you can't estimate the hazard for your timescale can you? So it makes sense to use the third option? If the later term (i.drug#i.current_age) is significant, then I keep the term in my model, yes? Would the HR would then be HR from i. drug multiplied by the HR for i.drug#i.current_age?

I can't seem to find a lot of guidance online using stcox with age as the timescale, so any help would be much appreciated!

[Phil Bromiley]

You didn't get a quick answer - you'll increase your chances of a useful answer if you follow the FAQ on asking questions - provide Stata code using code delimiters, Stata output, and sample data using dataex. Also simplify your question as much as you can. You ask a lot of questions here which also lowers your chances of an answer.

I'm sorry but I don't use this technique and can't provide you with a direct answer about stcox.