I'm examining associations of various risk behaviors within a clinical cohort on having a suppressed or un-suppressed HIV viral load using logistic regression.

My concern is that I have covariates where the probability of falling into the cell of a contingency table would be zero. I've seen this called "structural" zeros, not sure that's correct.

For example: E(Y)= Bo + B1*gender + B2* msm + B3.....+ e
gender: (0) male (1) female
msm: (0) not msm (1) men-who-have-sex-with-men

By definition there will always be zero where gender=1 and msm=1, thus precluding inclusion of both the gender and msm variables in the regression equation (I think).

Would a solution be to create a factor variable such that: (0) female heterosexual (1) male heterosexual (2) men-who-have-sex-with-men. And would this essentially be an interaction term of gender*msm?

Thanks for the help!