Because multiple colinear variables like that would make the confidence intervals blow up because not every person in the study is obese much less the same degr

Because multiple colinear variables like that would make the confidence intervals blow up because not every person in the study is obese much less the same degree of obese, and adjusting for colinearity is a basic part of data modeling. Just like how smokers may partake in worse health practices in general, but not every smoker does, and not every smoker is unhealthy to the same degree, thus you can find an effect across all smokers, like O2 deficiency, even a health nut smoker.