Odds Ratios

 

 

First of all, what are odds? Odds are simply a ratio of the probability that an event will occur versus the probability that the event will not occur, or probability / (1-probability).

 

For example, if you go fishing and you catch 3 largemouth bass and 1 trout, then the odds of catching a trout = [(1/4)/(3/4)] = 1/3 = 0.33.

Note that this differs from risk (or probability): the risk of catching a trout is equal to

(# of trout caught) / (total # of fish caught) = 1/4 = 0.25.

 

Odds ratios, therefore, are simply a ratio of odds; in general they refer to the ratio of the odds of an event occurring in the exposed group versus the unexposed group.

 

For example, if you compare your luck with fishing with no bait versus fishing with Billy Bob's Bassassinator Bait and cast your line 100 times using each method, a 2x2 table would show the following:

 

 

# of times caught

# of times not caught

Total # of casts

Bassassinator

50

50

100

No bait

2

98

100

 

The odds of catching a fish with the Bassassinator is 50/50 or 1.0.

The odds of catching a fish with no bait is 2/98 or 0.02.

Therefore, the odds ratio for catching a fish with the Bassassinator vs. no bait is

1.0/0.02 = 50.

The probability  of catching a fish with the Bassassinator is 50/100 or 0.50.

The probability  of catching a fish with no bait is 2/100 or 0.02.

Therefore, the relative risk for catching a fish with the Bassassinator vs. no bait is 0.50/0.02 = 25.

 

So what exactly does this odds ratio tell us? Odds ratio can be used to give us an idea of how strongly a given variable may be associated with the outcome of interest compared to other variables. Odds ratios are simply a different way of expressing this association than relative risk since they compare odds rather than risk of an event; however, they are sometimes very close to each other, such as when the outcome of interest is rare.

 

For example, in the example above, we can say that using the Bassassinator Bait is strongly associated with catching fish compared with using no bait. If we also compared another group, such as using worms for bait, and found that 25 of 100 casts using worms yielded fish, the odds for catching a fish using worms would be 0.33, with an OR = 16.5. We could conclude that Bassassinator Bait is better than worms at catching fish, and worms are better than nothing.

 

Assuming that this trial is in fact a RCT, we can conclude that Bassassinator Bait improves the odds of catching a fish by 50 times compared to using no bait; or we could say that the risk or chance of catching a fish is increased 25 times in fishermen using Bassassinator Bait compared to those using no bait.


Why do we use OR instead of RR in case-control studies? To be able to calculate relative risk, we compare the risks of outcome in different groups. In case-control studies, we already know what the outcome is and we separate groups into those with the outcome vs. controls. Our objective in such studies is to try to identify risk factors that are more strongly associated with one group than the other; thus, risk and therefore relative risk cannot be calculated from these studies. We use odds ratios instead, which can give us a measure of how strongly the risk factor is associated with the outcome.

 

For example, if we suspect that Bassassinator Bait is associated with catching more fish, then we could take 100 successful fishermen and compare them with 100 fishermen who were unable to catch any fish and find out how many in each group used Bassassinator Bait. Since we select the outcome in both groups, we cannot calculate the relative chance (risk) of catching fish in the Bassassinator Bait users because we do not know the chance of catching fish in the general population (who are non-Bassassinator users), and therefore we have no comparison group. However, we can compare the odds of the use of Bassassinator Bait in those who caught fish vs. those who were unable to catch fish by calculating the odds ratio. So for example:

 

 

Bassassinator use

No Bassassinator

Caught fish

40

60

Caught nothing

20

80

 

Odds of Bassassinator use in those who caught fish = 40/60 = 0.67.

Odds of Bassassinator use in those who caught nothing = 20/80 = 0.25.

Odds ratio of Bassassinator use in successful vs. unsuccessful fishermen = 0.67/0.25 = 2.7.

 

You can say that the odds of use of Bassassinator were 2.7 times greater in successful fishermen vs. unsuccessful fishermen in this study. This implies an association between use of Bassassinator bait and catching fish. However, remember that many other things could have contributed to this apparent association: chance alone could have accounted for this difference (helpful to know the 95% CI for the OR); the sample selected for both groups could have been skewed to favor Bassassinator users in the group that caught fish; or perhaps the successful fishermen were more likely to recall using Bass. Bait (recall bias).

 

Please note that we cannot conclude that Bassassinator increases the risk of catching fish by 2.7 times; all we can conclude is that successful fishermen were 2.7 times more likely to have used Bassassinator Bait than unsuccessful fisherman in this study only, which would lead us to believe that there could be an association with Bassassinator use and successfully catching fish. We would need to do a RCT or prospective cohort study to be able to estimate the magnitude of the effectiveness of the Bassassinator Bait.

 

QUICK HITTERS:

1.       Odds = Probability / (1-probability).

2.       Odds ratio (OR) = ratio of odds of event occurring in exposed vs. unexposed group.

3.       Odds ratio are used to estimate how strongly a variable is associated with the outcome of interest; in prospective trials, it is simply a different way of expressing this association than relative risk.

4.       In case-control studies, we separate groups by their outcomes and retrospectively try to identify variables that appear to be more associated with one outcome than another. Therefore, we cannot deduce a calculable risk because the outcome has already been predetermined. We therefore use odds ratios instead to estimate the strength of association of the variable with the outcome of interest.