Hi all,
I'm working on a dataset that I want to use to study racial disparities in access to a surgical procedure before and after a new surgical device was introduced onto the market. I want to use a competing risk analysis (subhazard Fine and Gray model), where patients can either get the surgery (failure=surgery) or die while waiting for the procedure (competing risk). I want to compare the subhazard ratio between race categories within each period (before and after device inntroduced), and then determine racial differences between pre- vs. post-implementation eras. Finally, I want to censor pre-implementation patients at the date of device implementation so that they do not contribute time to the post implementation era. Here is how I coded it:
1) Generated survival times for each period for censoring purposes/avoiding survivorship bias:
*Generate survival time pre implementation*
gen survival_time_pre = (removal_date - original_eligibility_date) if implementation_period==0
removal_date is when they were either operated on or when they died.
*Generate survival time post implementation*
gen survival_time_post = (removal date - original_eligibility_date) if implementation_period==1
2) *Binary event variables for survival analysis.
gen surgery=cond(removal_category==1, 1,0)
label variable surgery "Surgery"
gen died=cond( removal_category==2, 1,0)
label variable died "Died"
3) *Create binary variable to set up for competing risk*
gen surgery_died=surgery
replace surgery=2 if died==1
tab surgery_died
label define surgery_died 0 "Other" 1 "Surgery" 2 "Died"
label values surgery_died surgery_died
4) Competing risk analysis in pre-implementation era where failure==surgery
stset survival_time_pre, failure(surgery_died==1)
stcrreg i.race, compete(surgery_died==2)
stcrreg i.implementation_period##i.race, compete(surgery_died==2)
margins i.implementation_period#i.race
The issue with this is that I get a lot of colinearity when I introduce the interaction term. The biggest challenge is that I do not know if my approach to generating survival times for pre and post implementation eras separately is correct.
I really hope this makes sense. Please let me know if it does not and I will attempt to clarify. Thank you!!
Emile
I'm working on a dataset that I want to use to study racial disparities in access to a surgical procedure before and after a new surgical device was introduced onto the market. I want to use a competing risk analysis (subhazard Fine and Gray model), where patients can either get the surgery (failure=surgery) or die while waiting for the procedure (competing risk). I want to compare the subhazard ratio between race categories within each period (before and after device inntroduced), and then determine racial differences between pre- vs. post-implementation eras. Finally, I want to censor pre-implementation patients at the date of device implementation so that they do not contribute time to the post implementation era. Here is how I coded it:
1) Generated survival times for each period for censoring purposes/avoiding survivorship bias:
*Generate survival time pre implementation*
gen survival_time_pre = (removal_date - original_eligibility_date) if implementation_period==0
removal_date is when they were either operated on or when they died.
*Generate survival time post implementation*
gen survival_time_post = (removal date - original_eligibility_date) if implementation_period==1
2) *Binary event variables for survival analysis.
gen surgery=cond(removal_category==1, 1,0)
label variable surgery "Surgery"
gen died=cond( removal_category==2, 1,0)
label variable died "Died"
3) *Create binary variable to set up for competing risk*
gen surgery_died=surgery
replace surgery=2 if died==1
tab surgery_died
label define surgery_died 0 "Other" 1 "Surgery" 2 "Died"
label values surgery_died surgery_died
4) Competing risk analysis in pre-implementation era where failure==surgery
stset survival_time_pre, failure(surgery_died==1)
stcrreg i.race, compete(surgery_died==2)
stcrreg i.implementation_period##i.race, compete(surgery_died==2)
margins i.implementation_period#i.race
The issue with this is that I get a lot of colinearity when I introduce the interaction term. The biggest challenge is that I do not know if my approach to generating survival times for pre and post implementation eras separately is correct.
I really hope this makes sense. Please let me know if it does not and I will attempt to clarify. Thank you!!
Emile
