The odds ratio is a technical term often used in statistics, and especially in medical statistics and epidemiology. The odds ratio is the ratio of the relative incidence (odds) of a target disorder in the experimental group relative to the relative incidence in a control group. It reflects how the risk of having a particular disorder is influenced by the treatment. An odds ratio of 1 means that there is no benefit of treatment compared to the control group.
The odds ratio is a difficult concept for non mathematicians, and recommendations for how to teach its use are available.
For a mathematician, the relative incidence of a disorder would be called the 'odds on having the disorder'. An odds ratio is the ratio between the odds on some event under two circumstances.
If we have A individuals with the disorder and B without in the experimental group, and C individuals with the disorder and D without in the control group, the empirical odds ratio is
This example is from the Titanic (example from Power):
142 survived, 709 died
- Odds of survival = 142/709 = 0.20
- Probability (risk or chance) of survival = 142/(142+709) = 17%
308 survived, 154 died
- Odds of survival = 308/154 = 2.00
- Probability (risk or chance) of survival = 308/(308+154) = 67%
- Odds ratio (OR) for survival = 0.20/2.00 = 0.10
- Relative risk (RR) for survival = 17%/67% = 0.25
The odds on a male passenger surviving the Titanic disaster was 1 to 5 against surviving. The odds on a female passenger surviving the Titanic disaster was 2 to 1 in favour of surviving. The odds ratio for survival between men and women is (1/5) / (2/1) = 1/10
Rather than classifying the passengers on the Titanic as male or female, and comparing their odds on survival, one could classify passengers as survivors or not survivors, and compare the odds on being male or female. The odds ratio for sex (male versus female) between survivors and non-survivors is also 0.10. This follows from the algebraic identity .
The identity between the odds ratios is much used in epidemiology, especially in retrospective case-control studies. Consider for example the relation between lung cancer and exposure to asbestos. Instead of sampling from the two populations of people who are exposed to asbestos and people who have not been exposed to asbestos, and studying the odds ratio for lung cancer between the two populations defined by exposure status, one could just as well sample from the two populations of people who got lung cancer and people who did not get lung cancer, and study the odds ratio for asbestos-exposure between the two populations defined by health status. This is extremely valuable when studying illnesses which are rather rare, even in the population exposed to the risk factor of interest.
The odds ratio is generally used to measure the association between a risk factor and disease. However, using the odds ratio to measure the ability of a risk factor to diagnose disease is problematic. The odds ratio should be at least 16 to have reasonable diagnostic ability.
The odds ratio may be the most stable ratio across different prevalences.
For odds ratios less than 1:
For odds ratios greater than 1:
- Anonymous. Odds and odds ratio. Bandolier.
- Prasad K et al. (2008). "Tips for teachers of evidence-based medicine: understanding odds ratios and their relationship to risk ratios". J Gen Intern Med 23: 635–40. DOI:10.1007/s11606-007-0453-4. PMID 18181004. Research Blogging.
- Power M (2008). "Resource reviews". Evidence-based Medicine 13: 92. PMID 18515638.
- Boyko EJ, Alderman BW (1990). "The use of risk factors in medical diagnosis: opportunities and cautions". J Clin Epidemiol 43: 851–8. PMID 2213074.
- Pepe MS et al. (2004). "Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker". Am J Epidemiol 159: 882–90. PMID 15105181.
- Sinclair JC, Bracken MB (1994). "Clinically useful measures of effect in binary analyses of randomized trials". J Clin Epidemiol 47: 881–9. PMID 7730891.
- Altman DG, Deeks JJ, Sackett DL (November 1998). "Odds ratios should be avoided when events are common". BMJ 317 (7168): 1318. PMID 9804732. PMC 1114216.
- Page J, Attia J (2003). "Using Bayes' nomogram to help interpret odds ratios". ACP J Club 139: A11–2. PMID 12954046. Helpful chart
- Furukawa TA et al. (2002). "Can we individualize the 'number needed to treat'? An empirical study of summary effect measures in meta-analyses". Int J Epidemiol 31: 72–6. PMID 11914297.
- McQuay HJ, Moore RA (1997). "Using numerical results from systematic reviews in clinical practice". Ann Intern Med 126: 712–20. PMID 9139558.