# How to model gender specific values/variables as a predictor variable in the regression model?

My research question is to check whether the Body fat is associated with Hypertension onset. I am using Body fat as a categorical variable (i.e according to the value of body fat, the person will be categorized lean, overweight or obese).

But the classification of body fat for a healthy individual is different for males and females and their age. Is that fine to use the male's body fat classification and adjust the model with sex and age? if not, what is the right way to do this? The dependent variable in the model is 0 or 1 (i.e hypertensive or non-hypertensive) and the predictors are body fat classes and few other continuous values ...

• Why not categorize each person as lean/overweight/obese acording to his/her sex-and-age-specific norms and use this categorization as a predictor? Commented Jul 24, 2019 at 6:19

OK, I'll post my comment as an answer. So:

Why not categorize each person as lean/overweight/obese according to his/her sex-and-age-specific norms and use this categorization as a predictor?

mkt's answer is also fine, but I'd stick to categorization. It'll give you conclusions like "Being obese increases odds of hypertension by 50% compared to being lean".

Including continuous body fat into your model will give you something like "each additional percent point of body fat (I assume body fat in in percents) increases odds of hypertension by 2%". Plus, if you include interactions with sex and age as mkt suggested you'll get different effects of body fat for each sex-age combination like:

• "for 20-40 years old males, each additional percent point of body fat (I assume body fat in in percents) increases odds of hypertension by 4%"
• "for 20-40 years old females, each additional percent point of body fat (I assume body fat in in percents) increases odds of hypertension by 2%"
• "for 41-60 years old males, each additional percent point of body fat (I assume body fat in in percents) increases odds of hypertension by 6%"

and so on ...

It's your choice what suits you better.

I'd recommend just using the continuous body fat values instead of binning them into groups. That allows you to avoid any problems with different classification thresholds, and also uses all the information you have (information is thrown out when you bin continuous data). Then you can do a logistic regression with body fat, sex and age as predictors, with appropriate interaction terms.