RELATIVE RISK VS ODDS RATIO

[mydamj]

I am doing multinomial logistic regression with 4 levels of dependent variable. I need to get the relative risk,

I used proc logistic with link = glogit. I got the odds ratio estimates but the estimates match with the relative risk estimates when I ran each group separately as a univariate model using proc genmod link=log.

so was wondering if the odds ratio estimate is the relative risk in the multinomial model(proc logistic with link=glogit).

Kindly provide your suggestions.

Thanks.

[StatDave]

They are not the same. See this note which shows how to estimate relative risks in nominal, multinomial models.

[Season]

They are not the same.

By definition, odds ratio is the ratio of odds of those who are exposed and those who are not exposed to a certain exposure. It is most frequently used in case-control studies, where (in most cases) no follow-up occurs.

Relative risk is the ratio of incidence of a certain event (usually called "endpoint") between those who are exposed and those who are not exposed to a certain exposure. Relative risk is used in cohort studies and experimental studies (e.g. randomized clinical trails, RCTs), where follow-up from the "zero time" (the time you start observing if the endpoint happened in the cohort) is a must.

Odds ratio and relative risk are fundamental and basic concepts in epidemiology. You can refer to a book on epidemiology for further explanation on the two terms.