This whole category of decision making basically consists of taking observable things and then using them to infer other things despite the correlations often only being barely better than a coin toss. It's the same logic by which the police harass you more if you check more bad demographic checkboxes.

You can make an argument that it's different because the stuff being measured is at the other end of the "how easily can they change it" spectrum but that doesn't change the fundamental accuracy of the correlation. Something like this shouldn't be used for anything serious.