how to calculate eventprob for proc power coxreg ?

[NinaWestern]

Hi everyone,

I am trying to estimate a sample size using proc power; coxreg 

In order to estimate ntotal, I need to specify eventprob=

Where can I get this value from?

Here is the code:

proc power;
coxreg
sides=2
alpha=.05
hazardratio=.564
stddev=0.477
power = 0.8
ntotal= .;
run;

Thank you!

[SteveDenham]

You get eventprob in one of two ways. If you have historical or analytical results available, you can use that value. If you don't, but do have some idea of the range where it might be, you can plug in a "row vector" of possible values, and PROC POWER will calculate ntotal for each of those values, given the other parameters.

SteveDenham

[NinaWestern]

Thank you, @SteveDenham.

I know its simple, but just want to make sure I am using the right values from the historical data:

In my case, we have a control and treatment group where we assess the recidivism rate over time. After 12 months, in the control group, there is a 20% of recidivism (2/10 get arrested again), while in the treatment group it's 15% (1.5/10 get arrested).

At 18 months, these rates are higher (25% and 18%).

how do I calculate the eventprob from those values?

Thank you, I really appreciate your help!

[SteveDenham]

I think you just plug in the value of interest.  Here is code, with the two control rates and the two treated rates you have:

proc power;
coxreg
sides=2
alpha=.05
hazardratio=.564
stddev=0.477
power = 0.8
eventprob = 0.15 0.18 0.2 0.25
ntotal= .;
run;

The only thing that bothers me is the hazardratio estimate, but if you are comfortable with that, this should work for you.

SteveDenham

[NinaWestern]

@SteveDenham

Thank you!

The hazard ratio is taken from the Cox regression using the historical data set. So it is across all time points, B =-.574, HR = .564 (95% CIs: .233, 1.36), p = .202

The data collection started a while ago but is ongoing. Based on that data (current n=188), we are trying to figure out the final sample size to reach 80% power.  

Does this make sense?

Should the HR in the proc power syntax be recalculated at 12 months and 18 months, to correspond with the event rates? 

This is my first time doing this type of calculation.

[SteveDenham · Accepted solution]

To meet your goals, I would suggest updating the HR at fixed points.

SteveDenham

[NinaWestern]

@SteveDenham, thank you!

I ran it as per your suggestions - it does male sense.
Appreciate all your help!