bauerb wrote: ↑Mon Apr 03, 2023 4:36 pm
I can understand the importance of BMI in the beef cattle business, but I don't see it being important to skiers.
Yeah, it's kind of like judging intelligence based off one math test. Sometimes it correlates, possibly more often than not.
Type I and II Errors:
Yet as it could mislabel a high BMI (lo-fat hi-muscle) person as 'overweight-obese' (Type II Error, false negative, false "unhealthy"), I wonder what a Type II for "underweight" BMI, and Type I (false positive, false "healthy") for "normal" BMI, just to round out the possible error categories.
Perhaps a "normal BMI" error would be for "skinny-fat"?
Methodology Mental Exercise:
Think of this not only for BMI screening, but also for cancer screening, or anything else "unhealthy" a doctor would want to bring up with their patient.
A system that
only errors by "flagging" healthy people is
fine as a quick screening tool
if it is followed up with a more detailed analysis that would catch the error and correct an initial "unhealthy" classification as "healthy", such as a body fat % test (electrical pads, skin pinch, water tank, etc.). I.e. it catches all the "unhealthy" and some "healthy." Yeah, annoying for "healthy" to have to jump a second 'hoop,'
"we think it's a tumor, so are going to have a second look."
By contrast, if a system
only errors by "failing to flag" those it should, it is
worse because while it never puts a "healthy" in the "unhealthy" group, it would let "unhealthy" pass a screening as "healthy." The person would continue along thinking they were good when in fact they are not and their health would continue to degrade,
"test show that you have no tumors" when the patient actually does.
Back to BMI:
It clearly can make the former error. If it can also make the latter error (classify an "unhealthy" as 'ok') with no steps in place to catch
that error, then it has serious issues as a screening tool. The best it could possibly be at that point is merely one of many in a battery of required tests where somehow it could catch an error in another test just as other tests could catch an error with it in a cross-checking fashion.