Why remove a constant before calculating a correlation? Correlation is invariant to a constant.
Anyway, the following bash code does what I need in my script, I'll put it here for posterity. Note that I'm doing this on a one-brick file; if you're reading this in the future and trying to apply it to your situation you will also have to pick out the brick you want.
# get the statpar line for the correlation for this file
statpar=$(3dinfo $file | grep 'statcode = fico;')
# get just the numbers (everything after the last equals sign)
statpar=${statpar##*=}
# replace allthe spaces with commas for input to ccalc
statpar=${statpar// /,}
# get the thresholds that correspond to the desired p-values for this one
t001=$(ccalc -expr "fico_p2t(0.001 ${statpar} )")
