Under logistic regression, the resulting odds ratios can be approximated and interpreted as risk ratios only if the outcome is rare (as a rule of thumb, less than 10% prevalence). This is sometimes referred to as the rare disease assumption.
If I develop a multinomial logistic regression model comparing a reference category to two or more other categories, I am unsure if the rare disease assumption need to be met for me to interpret the transformed coefficients as relative risk ratios. My understanding is that multinomial logistic regression directly estimates RRRs, and so an approximate conversion from ORs to RRs based on the rare disease assumption is not required, and hence the rare disease assumption is not required.
For multinomial logistic regression, do each of the non-reference categories need to have a prevalence of roughly less than 10% for the transformed coefficients to be interpreted as RRRs?
