Study Control
Disease a b
No disease c d
A total row can be added, as below, to simplify calculations
Study Control
Disease a b
No disease c d
Total a+c b+d
1. What is risk?
In everyday language, the word risk has many closely related meanings, which are often
associated with bad outcomes. In statistical and epidemiologic terms, risk is simply an expression
of probability. It is usually, but not invariably, the probability of an adverse event such as disease,
injury, or death. Sometimes the nature of the risk depends on the perspective of the observer. For
instance, there is a risk that you will lose money (an adverse event) if you buy a lottery ticket;
but, if you do not win, the lottery organizers will profit (a beneficial event).
8. Give examples of how incidence risk and incidence rate are determined.
Let's take an extreme example of a population of 1000 with a serious disease that causes 720 of
them to die in 1 year. The incidence risk of mortality in that population will be 720 / 1000 = 0.72
/ year.
The simplest way to work out the average number of the population who are disease-free over the
year is to calculate the mean of those disease-free at the beginning (1000) and end of the year (1000
- 720 = 280). In this case, the average is (1000 + 280) / 2 = 1280 / 2 = 640. The incidence rate of
mortality in that population is 720 / 640 = 1.125. Unlike a risk, incidence rate can, therefore, exceed
1 in some circumstances. An alternative way of describing the denominator for incidence rate is the
mean time that one person will need to wait before contracting the disease.
9. Do these differences have significant effect on study results?
The difference between calculated values for incidence risk and rate is minimal in the majority of
cases. Two factors are responsible for this:
Most diseases affect quite small proportions of a population, and few will leave the group at risk.
The rate denominator often remains stable as new persons at risk replace those who leave the
group. For instance, a population remains moderately stable if births roughly balance out deaths.
10. What are high and low values for absolute risk?
There is no standard. A single value of absolute risk in isolation does not tell you whether it is
high or low. The only way that this can be done using absolute risk is by comparing findings in
two or more sets of observations.
11. Give an example of how high and low values for absolute risk are determined. How did the
risk of divorce in the United Kingdom compare with that in Greece in 2002?
Government statistics show that the risk of divorce in Greece in 2002 was 1.1 per 1000
population, while the risk of divorce in United Kingdom in 2002 was 2.7 per 1000 population.
The risk of divorce was clearly greater in the United Kingdom than it was in Greece. The
conclusion is straightforward in this case because such figures are usually derived from
centrally kept national registers covering the whole population. If only a sample of each
national population had been taken, statistical tests would have been required to establish that
the differences in the risks were significant.
18. The above example does not seem misleading. Why is it?
First, let us examine the benefit of removing the food additive since it showed a higher relative
risk than the medication. This removal would lead to 4 fewer cases of disease in a population of
1 million. Despite its lower relative risk, however, avoiding the medication would lead to 20
fewer cases in a population of 1 million (1.2 cases in 10,000 - 1 case in 10,000 = 0.2 cases in
10,000 = 20 cases in 1 million). The difference in background risk between the two diseases
can have a profound effect, and it is often best to calculate the actual anticipated change in
terms of raw numbers of cases.
22. This seems confusing. Is there an easier way to calculate the odds ratio?
Yes. Fortunately, the raw figures can also be used. Referring to the generic 2 2 table, the odds
of a positive outcome in the study population would be a/c. The odds of a positive outcome in
the control population would be b/d. Therefore, the odds ratio would be
23. Why is it typically impossible to calculate relative risk from a case-control study?
A case-control study involves selection of one group with disease (or other outcome of interest)
and one without. The number of individuals with the risk factor under investigation in each
group is then determined. Remember, risk can be calculated from a sample only if it reflects the
proportions in the whole population.
25. Why is the probability of having a risk factor in the presence or absence of disease seldom
calculated from the results of a case-control study?
There is no more than a very loose relationship between the risk of disease in the presence or
absence of a risk factor and the probability of having that risk factor in the presence or absence
of the same disease. This information would be of academic interest only and of no predictive
use at all.
Because the whole population is recorded, risks can be calculated. The risk of pulmonary
barotrauma with a low FVC is 39/7586 = 5.1 in 1000 ascents. The probability of a low FVC in
the presence of pulmonary barotrauma is much greater at 39/122 = 0.47 and is clearly not a good
measure of the risk of each ascent undertaken.
27. What measure can be used to obtain clinically useful predictive information from a case-
control study?
The odds ratio can be used to illustrate the strength of association between risk factor and
outcome in both cohort and case-control studies. A mathematic property inherent within the
odds ratio that makes it particularly useful for case-control studies is that the odds ratio does not
require risk factors to be present in the same proportions as the population from which the
samples were drawn. In addition, the odds ratio possesses symmetry such that
As a result, the odds ratio gives a consistent and meaningful measure of the relationship
between risk factor and disease, whichever way it is calculated. Also, if the outcome of interest
occurs only rarely, the odds ratio is very close to the relative risk.
28. Give a worked illustration of the symmetry of odds ratios.
For this example, we will use the generic 2 2 table at the beginning of the chapter. We saw
earlier that the odds ratio of disease in the presence of a risk factor is given by (a/c) / (b/d) =
ad/bc. Re-titling the generic 2 2 table to represent the findings of a case-control study, in
which the numbers were divided between outcome and risk factor in the same proportions as
before, gives us the following:
The odds of a risk factor in the presence of disease equals a/b. The odds of a risk factor in the
absence of disease equals c/d. The odds ratio is a/b / c/d= ad/bc, the same as the odds ratio
calculated using odds of disease.
29. Give an illustration of the circumstances required for odds ratio to closely approximate to
relative risk.
The relative risk in a cohort study is given by the expression Rc/Rs. In terms of raw numbers
from the generic 2 2 table, relative risk is equal to [a/(a + c)]/[b/(b + d)]. As the outcome of
interest becomes rarer, both a and b tend to zero and the relative risk tends to which is the same
as the odds ratio. So if a disease is rare, the odds ratio is a close estimate of the relative risk.
30. Apply the odds ratio to the example of the submarine escape training tank.
FVC normal
Low FVC Total
or larger
PBT 39 83 122
No PBT 7547 240,168 247,715
Total 7586 240,251 247,837
The results from the submarine escape training tank are repeated below:
Note the odds ratio is identical calculated either way. Also note that the odds ratio and relative risk
are very similar because clinically apparent PBT is rare.
40. Give an example of how relative and attributable risks differ in their sensitivity to
absolute magnitude of risk.
If Rs is 0.075 and Rc is 0.05, then attributable risk is 0.075 - 0.05 = 0.025 and relative risk
is 0.075 / 0.05 = 1.5.
If Rs is 0.75 and Rc is 0.5, then attributable risk is 0.75 - 0.5 = 0.25 but, despite the tenfold
increase in overall risk, relative risk is unchanged at 0.75 / 0.5 = 1.5.
DCI
No DCI
Total
It is widely believed that a PFO, or any other right-to-left circulatory shunt, increases the risk of
DCI. Some 30% of individuals within the general population have a PFO, and some surveys
have shown that a similar proportion of divers have a PFO.
In this study, the absolute risk of DCI with a PFO is 2/500 = 0.004 and the absolute risk of DCI
in an individual with no PFO is 1/500 = 0.002. The relative risk is 0.004/0.002 = 2 and the odds
ratio is 2/0.998 = 2.
43. What if the control group differs from the study group in ways (other than the risk factor
under investigation) that might influence the outcome of interest?
It is not unusual for characteristics, such as age and sex, to influence outcome, independent of
the risk factor being studied. In studies of cardiovascular mortality, for instance, fewer would
be expected to die in a younger population with a greater proportion of females. In such cases,
the odds ratios might need to be modified to account for differences in samples.
44. What techniques can be used to make the odds ratio more representative?
One technique is stratum matching, in which the same number of cases and controls are drawn
from each age/sex group. Another technique is to calculate the odds ratio for each group and
then summate them.
45. How do you calculate the odds ratio in a case-control study with matched controls?
In case-control studies, age/sex matched controls are often selected for each case. Sometimes
more than one matched control is selected for each case to enhance statistical significance. The
resulting information can be tabulated in the following format:
Controls
Risk
factor Risk factor
present absent
Cases Risk factor present a b
Risk factor absent c d
In this case, the odds ratio is calculated from b/c. If you want to
understand why this should be so, read a large reference book
and then let me know, too
References