It is important to note that the odds ratio is not an estimate of the relative risk. It has a different definition, and different properties.

The relative risk is easy to understand, but can lead to meaningless predictions. For example, if your probability of disease is 80% and you develop a risk factor that has a relative risk of 2, then your probability is now 160%!

On the other hand, no matter what your prior probability of disease is, there is no odds ratio that can make that probability greater than 100%.

@Cristian is right that the odds ratio gets close to the relative risk when the probability of disease is small, but that's not the same thing as being an estimate of it.

The OR is an estimate of the RR (as the HR when performing survival analysis)

OR can be performed when the outcome is rare when this is rare the two are equivalent (a statistician would say unbiased estimate...) http://en.wikipedia.org/wiki/Odds_ratio

In case controls (retrospective) studies for cancer, for example, the OR could be used as a RR estimate. When possible epidemiologist always prefer to run expensive prospective cohort study and this to have a better control of covariates and avoid recall bias (also a biger sample size is preferred to reduce estimate confidence interval). Can you imagine a retrospective study about gastric cancer in wich you ask patients how much meat they had in the last 10 years? So, it is more reliable to run a prospective study with food questionnaires performed every year...

best ^_^

It is important to note that the odds ratio is not an estimate of the relative risk. It has a different definition, and different properties.

The relative risk is easy to understand, but can lead to meaningless predictions. For example, if your probability of disease is 80% and you develop a risk factor that has a relative risk of 2, then your probability is now 160%!

On the other hand, no matter what your prior probability of disease is, there is no odds ratio that can make that probability greater than 100%.

@Cristian is right that the odds ratio gets close to the relative risk when the probability of disease is small, but that's not the same thing as being an estimate of it.

Ratio of incidence in exposure group to incidence of non-exposure is relative risk. This is computed in cohort or prospective studies where frequency of new cases of disease (incidence) out of total exposed or unexposed to a risk factor. for example, smoking status is exposure and disease is lung cancer. Smokers and non-smokers are enrolled in cohort to observe disease in long term follow up. A RR=15 interpreted as smokers are 15 fold risk of getting lung cancer as compared to non-smokers.

Odds ratio is computed in case-control studies. These are retrospective studies as against prospective nature. These are short duration and quick studies where for example available lung cancer cases and non-cases are enrolled and look back to their smoking history presence or absence. Then one can compute ratio: number of smokers divided by number of non-smokers among lung cancer case group and similar ratio for non-lung cancer case group. Each ratio is odds of smokers to non-smokers (It is like odds of winning a bet ). The ratio of the two ratios each for case and non-case is odds ratio (OR). This is mathematically equivalent to RR when disease is rare. OR can be interpreted as relative risk if the disease being studied is rare.

Cohort/ prospective study

Disease No disea Smoker n1

Non smoker

? ? n2

Case-control / retrospective study

Disease No disease

Smoker ?

Non smoker ?

N1 N2

Hello,

nothing much to add, I think Ronan and Satyanarayana resumed it very well. One important thing to know is that you cannot use RR in case control studies, but instead OR. A few papers that might help you. If you cannot get them, let me know and I can forward them to you. They are all really good and short.

- Peter Cummings. The Relative Merits of Risk Ratios and Odds Ratios. Arch Pediatr Adolesc Med. 2009;163(5):438-445

- David A. Grimes, MD, and Kenneth F. Schulz. Making Sense of Odds and Odds Ratios. Obstet Gynecol 2008;111:423–6

- Carsten Oliver Schmidt, Thomas Kohlmann. When to use the odds ratio or the relative risk? Int J Public Health 53 (2008) 165–167

- Christopher L. Sistrom. Proportions, Odds, and risks. Radiology 2004; 230:12–19

hope this help.

Thanks Francois, would you please send me these papers.

I personally do not like the Odds Ratio metric, I find it misleading and a poor metric because of issues regarding its sensitivity and for a lack of transparency. The attached paper is a good overview of common problems associated with the OR metric...

Thanks Luisiana.

I think Ronan summed it up correctly. There is nothing wrong with the OR - it has useful properties that make it more suitable for some situations than others. The claim about transparency seems to me subjective (I find OR more transparent and easy to understand than RR and more useful for the work I do).