Risk ratio or odds ratio: which should I use?
Use the risk ratio when your studies are randomised trials or cohorts and the outcome risk is directly estimable — it is what readers think a ratio means. Use the odds ratio when the design forces it, principally case-control studies and analyses adjusted by logistic regression. If the outcome is common, an odds ratio will look like a larger effect than it is.
The difference in one line
- Risk ratio (RR)
- The ratio of two probabilities: the chance of the event with the intervention, divided by the chance without it. "Twice as likely" means what a reader assumes it means.
- Odds ratio (OR)
- The ratio of two odds, where odds are events divided by non-events. It has convenient mathematical properties and is routinely misread as a risk ratio, including in published abstracts.
Why the odds ratio exaggerates when the outcome is common
When an outcome is rare, odds and risks are numerically close and the two ratios nearly coincide. As the outcome becomes common the odds ratio moves further from the risk ratio, always away from 1 — so it looks like a bigger effect.
A worked case: if 60% of the control group and 80% of the treated group have the outcome, the risk ratio is 1.33 — a third more likely. The odds ratio is 2.67, because the odds go from 1.5 to 4. Both are correct. Only one of them will be read correctly by someone skimming your abstract.
A common rule of thumb puts the divergence at outcomes above roughly 10%. Treat that as a prompt to check, not a threshold to trust — the gap grows smoothly, it does not switch on.
When you have no choice
Case-control studies sample on the outcome, so the underlying risk is not estimable from the design and the odds ratio is the only valid ratio measure. Any analysis adjusted with logistic regression reports odds ratios, so a review of adjusted estimates will usually be pooling odds ratios whether or not you would have chosen them.
Do not convert odds ratios to risk ratios after the fact using an assumed baseline risk unless you are prepared to defend that baseline. The conversion is only as good as the number you supply, and it silently imports an assumption into a result that looks like data.
What to report either way
Pool on the log scale, present on the natural scale, and state which measure you used and why in the protocol rather than the discussion. Alongside the ratio, give an absolute measure — a risk difference, or an expected number of events per 1,000 at a stated baseline risk. That is the number a guideline panel needs and the number a ratio cannot supply.
Keep the direction of the comparison consistent across studies before pooling. A study reporting "odds of remaining event-free" is the reciprocal of one reporting "odds of the event", and mixing them silently cancels real effects.
Where this answer stops
Both are relative measures, and a relative effect says nothing about how much the intervention is worth. A doubling of risk means something very different at a baseline of 1% than at 40%. Report an absolute measure alongside whichever ratio you choose.