I am using the episensi and episens commands for external adjustment of selection biases. When I attempt external adjustment for selection bias for a cohort study using the option study(cs), I get the error message "selection bias correction works for case-control data". This limitation is true for the general episens and immediate episensi commands. I am able to obtain a selection bias-adjusted OR using the option study(cs); in fact, the ORs appear valid for cohort and cross-sectional data as well, even with the study(cc) option.
In any event, I would prefer selection bias-adjusted RRs from cohort studies, similar to what is available for the misclassification capabilities of the command.
To demonstrate the problem, here is an example in which I have set the selection probabilities at 100% for all four combinations of exposure and outcome, which should recreate the uncorrected measure of association. I compare this result to that from the cc and cs commands.
cci 300 200 99700 99800 // OR = 1.50 [1.26, 1.80]
episensi 300 200 99700 99800, study(cc) dpscex(c(1)) dpscun(c(1)) dpsnex(c(1)) dpsnun(c(1)) // OR = 1.50 [1.26, 1.80]
csi 300 200 99700 99800 // RR = 1.5 [1.25, 1.79]
episensi 300 200 99700 99800, study(cs) dpscex(c(1)) dpscun(c(1)) dpsnex(c(1)) dpsnun(c(1))
// Error: "selection bias correction works for case-control data"
My question is two-fold: Is there a way to obtain selection-bias adjusted RRs using this command, and (2) are there other packages out there to perform probabilistic sensitivity analyses, especially ones that can also incorporate the estimate's sampling variability?
Thanks, Garth
Garth Rauscher
University of Illinois Chicago School of Public Health
In any event, I would prefer selection bias-adjusted RRs from cohort studies, similar to what is available for the misclassification capabilities of the command.
To demonstrate the problem, here is an example in which I have set the selection probabilities at 100% for all four combinations of exposure and outcome, which should recreate the uncorrected measure of association. I compare this result to that from the cc and cs commands.
cci 300 200 99700 99800 // OR = 1.50 [1.26, 1.80]
episensi 300 200 99700 99800, study(cc) dpscex(c(1)) dpscun(c(1)) dpsnex(c(1)) dpsnun(c(1)) // OR = 1.50 [1.26, 1.80]
csi 300 200 99700 99800 // RR = 1.5 [1.25, 1.79]
episensi 300 200 99700 99800, study(cs) dpscex(c(1)) dpscun(c(1)) dpsnex(c(1)) dpsnun(c(1))
// Error: "selection bias correction works for case-control data"
My question is two-fold: Is there a way to obtain selection-bias adjusted RRs using this command, and (2) are there other packages out there to perform probabilistic sensitivity analyses, especially ones that can also incorporate the estimate's sampling variability?
Thanks, Garth
Garth Rauscher
University of Illinois Chicago School of Public Health
