While for individual cases clinicians shouldn't rely on BMI much, and only use it for screening, in studies it makes sense to use it because it's a useful tool. A person with a BMI of 30 is almost always fatter than a person with a BMI of 20. The outliers (like athletes with big bones) wouldn't change the statistical analysis.

It's in the same boat with IQ, it's a useful tool if you accept it has limited uses.