The rare disease assumption states that odds ratio (OR) only approaches relative risk when the disease (=outcome) is rare, i.e., < 10%. There are papers that show that outcomes with higher prevalences may lead to more extreme ORs (biased away from the null value of 1.0).
Is this effect independent of the prevalence of the exposure? Or to put it differently: is the OR affected, if the exposure is very common?
the OR should not be affected by the prevalence of the exposuer
if your outcome is common, i would suggest alternative models such as mod. poisson and log-binomial regression
Agree. The prevalence of the exposure should not affect the OR.
High prevalence doesn't affect the estimate of the OR, but makes the OR a poor approximation of the risk ratio.
Outcome prevalence up to 20% is usually ok. I do not see why exposure prevalence is relevant at all
As you are working in Germany, I'm shure that you are able to read the PPT enclosed.
Slide 9 gives a numerical example for what has been said above. There is also a visualization of the behavior of OR and RR under various scenarios (slides 12pp).
ORs are by nature different from RRs. The challenge is when you are trying to apply the interpretation of an RR to an OR. If your desire is to have an estimate that has the interoperation of a RR then you should use a different analytic approach such as a log-binomial model rather than resting on the rare-disease assumption. ORs will always be more extreme than RRs (i.e. further from 1) regardless of the prevalence of the outcome.
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology