bysort group: lsens

[joao maroco]

Dear all,
I am trying to find out the best cutoff point for a logistic problem.
This is my MWE with standard data from stata:

sysuse auto
gen group = cond(mpg <20, 1, 2)
bysort group: logit foreign c.mpg, nolog
lsens , gensens(sensitivity) genspec(specificity)
bysort group: generate float youden_index = sensitivity + specificity - 1
bysort group: egen float max = max(youden_index)
bysort group: list mpg if youden_index == max

"bysort group" is not allowed to "lsens" function. So I generated spec and sens for the all data set. The problem is that I have a lot of spec and sens missing for group 1 and thus can not get the mpg corresponding to the maximum youden. Is there a way to generate spec and sens for each group independently? Or what am I doing wrong?

Thanks in advance!

[joao maroco]

I think I got it to work, but I don't get the same mpg values if I run the following code multiple times. Can someone please try it? I am on Stata17:

sysuse auto
gen group = cond(mpg <20, 1, 2)
drop specificity sensitivity cutoff youden_index max 
bysort group: logit foreign c.mpg, nolog
lsens, genprob(cutoff) gensens(sensitivity) genspec(specificity) nograph
generate float youden_index = sensitivity + specificity - 1
bysort group: egen float max = max(youden_index)
bysort group: list mpg if youden_index == max

I get 22 for group 2, but for group 1 it oscillates from 17 to 19 to ... :-(