Measuring Risk in Epidemiology

Transcript a b
c d

Measuring Risk in Epidemiology

Welcome to “Measuring Association and Risk in
Epidemiology.”
My name is Jim Gale. I am professor emeritus in
the Department of Epidemiology at the University of
Washington School of Public Health and Community
Medicine. I spent 12 years as a local health officer in
Kittitas County, Washington, and have also been a US
Public Health Service Epidemic Intelligence Service officer.

About this Module

This module and others in the epidemiology series from
the Northwest Center for Public Health Practice are
intended for people working in the field of public health
who are not epidemiologists but would like to increase
their familiarity with and understanding of the basic terms
and concepts used in epidemiology. Before you go on with
this module we recommend that you become familiar, if
you haven’t already, with the material presented in the
following modules, which you can find on the Center’s
Web site:
• What Is Epidemiology in Public Health?
• Data Interpretation for Public Health Professionals
• Study Types in Epidemiology
It is particularly important to understand the concepts of rates, ratios, and
measures of disease frequency, such as incidence rates and death rates. These
concepts are covered in the modules “What is Epidemiology in Public Health
Practice?” and “Data Interpretation for Public Health Professionals.”
If you want to review definitions of epidemiological terms, you may access the
glossary at any time from the attachments link at the top of
the screen. You can also find the transcript in the attach-
ments. You may want to print a copy now and make notes
on it as you go through the module.

Learning Objectives
In this module we will explore the meaning of risk and
association and their relationship to incidence rates in
epidemiology. We will also look at measures of risk and
their use and limitations in considering causal relationships.
By the end of this 45-minute module you should be able:
• To define risk as it is used in public health practice

Northwest Center for Public Health Practice 1

Measuring Risk in Epidemiology
Transcript a b
c d

• To identify measures of association and risk as they are used in epidemiology

• To interpret relative risk and odds ratios and be familiar with their calculation
using 2x2 tables, and
• To interpret the following measures of risk differences: attributable risk, popula-
tion attributable risk, and population attributable risk percent.

Defining Risk
Epidemiologists use the term risk to mean the probabil-
ity of an outcome (often a negative outcome) in a speci-
fied period of time. In epidemiology, risk usually implies
a quantifiable concept, such as the risk of dying or the
risk of a heart attack, rather than a more general concept
such as the risk of offending someone by speaking frankly.
In this module, I will use risk, probability, and likelihood
interchangeably, since they’re measured the same way.
Why is it important to know about risk? In the practice
of public health, we are faced with many choices. A quan-
titative estimate of risk is useful in making decisions about
a course of action or intervention as well as how to allo-
cate finite resources of time and money.

Risk vs. Association

In epidemiology, an association means a correlation, often
between an exposure and an outcome.
I’d like to pause for a moment to talk about correla-
tion and causation. Correlation is the situation in which
two or more variables, in this case exposure and outcome,
change at the same time. For example, as exposure to
the sun increases, the incidence of some types of cancer
increases. It’s important to remember that just because
both variables change, it does not mean that one causes
the other to change. In other words, correlation does not
equal causation. Some other unknown factor may be caus-
ing both the exposure and the outcome to change.
Deciding the degree to which an association might be causal, occupies a great
deal of an epidemiologist’s thinking and effort. Where an association is thought to
be causal, as in the association between smoking and lung cancer, we often use the
term risk factor in referring to the exposure we are comparing.
Measures of association refer to specific mathematical expressions that measure
the degree to which an exposure, such as exercise or smoking, is associated with
an outcome of interest, such as health status or disease. Some measures of associa-

Northwest Center for Public Health Practice 2

Measuring Risk in Epidemiology
Transcript a b
c d

tion we’ll explore later in this module are relative risk, odds ratio, attributable risk,
risk difference, population attributable risk, and population attributable risk percent.
Now we’ll pause for the first of several interactive exercises that will allow you to
answer questions about the material we have just covered. Please note that the
exercises can take several seconds to load.

Ratios, Proportions, and Rates

Before we go any further, let’s review three terms we’ll be
using in this module: ratios, proportions, and rates.
A ratio compares two dissimilar things by dividing one
quantity by another. For example, say a group of people
has 5 women and 7 men. The ratio of women to men is 5
to 7.
We can also express the relationship of women to the
entire group as a proportion, or a percentage, of the total
number of people.
A proportion or a percentage is a special kind of ratio,
because the group affected, or the numerator, must come
out of the population at risk, which is the denominator. For example, our group
contains 12 people. The proportion of women in the group is 5/12. We can also
express this fraction as 42 per 100, or 42%. The value of a proportion never exceeds
1, or 100 percent.
A rate is also written as a numerator and a denominator, but a rate has an element
of time. You will often see person-years used as a denominator in calculating rates.
A multiple of 10 is often used to create a numerator with at least one number to
the left of the decimal point. For example, the risk of dying last year from stomach
cancer may be 0.000023. However, for most people, it’s easier to think about this
rate when it is expressed as 2.3 deaths per 100,000 per year.
Rates can be written in several ways. For example, the rate of 8 deaths per
100,000 from suicide in 20 to 24 year olds in Oregon in 1995 could also be
expressed in these ways.
If you want to review the arithmetic of counts, numerators, denominators, ratios,
proportions, and rates, they are well covered in slides 11 through 14 in the module
“Data Interpretation for Public Health Professionals.”
Before we go on to look at measures of risk, or measures of association as we
usually call them, we should also look at a tool often used in calculating these
measures, the 2 by 2 table.

Northwest Center for Public Health Practice 3

Measuring Risk in Epidemiology
Transcript a b
c d

2 x 2 Tables
Epidemiologists use 2 by 2 tables, which are also some-
times called contingency tables, to record and analyze the
relationship between variables. Since we’re often compar-
ing two exposures and two states of illness, the 2x2 table
has four cells, or options. The usual format of the table
is to list the outcomes or disease categories in the verti-
cal columns, and the exposures, or attributes, in the table
rows.
In this example, we’ll label the exposure rows exposed
and non-exposed, and we’ll label the outcome columns,
sick and well.
Just for this example, we’ll represent the counts in the four cells as a, b, c, and d.
A key point to remember about 2x2 tables is that the data in the cells are actual
counts, not rates.
As you go through the module, you’ll see how to use a 2x2 table to calculate vari-
ous measures of association and risk.

Data Sources for Calculating Risk

As I mentioned before, risk is the same as the probabil-
ity, chance, or likelihood that something will happen.
Risks can be inferred from published death rates, from
incidence rates from an existing disease registry, such as
a cancer registry, or from attack rates from an outbreak
investigation. But more frequently special studies, such
as cohort studies, must be carried out to determine these
rates in populations that are or are not subject to specific
exposures. The way in which we calculate risk depends
on the type of study we conduct. (For more information
about these study types, see the module Study Types in
Epidemiology.)
An incidence rate is the number of new cases of illness occurring in a population
over a specific period of time, usually a year, divided by the total population at risk.
An attack rate is a specific form of an incidence rate, indicating the incidence of
disease in a population during an epidemic or outbreak, usually less than a year.
A death rate is the number of deaths (in general, or due to a specific cause) in a
population during a specific time (usually a year) divided by the total population.

Northwest Center for Public Health Practice 4

Measuring Risk in Epidemiology
Transcript a b
c d

A Problem with Proportions

I’d like to emphasize an important point.
As you interpret reports or news stories, always ask
yourself if statements implying risk are based on rates. In
particular, data from such sources as in-patient records
of hospitals must be interpreted with caution, since you
usually do not know the population from which each
group comes. The report is probably based on only a
proportion of cases within one facility, not on all persons
at risk in a defined population. It would be very difficult
to define the population from which these different age
groups are drawn, except under special circumstances.
To make this point clearer, let’s look at this list of hypothetical cases of stroke. It
might be tempting to say that the risk of dying is highest in the 60 to 69 year age
group. But we don’t know the number of persons at risk in each age category; that
is we can’t define the denominators to calculate risks for any of the age categories.
Although we may know in general, that strokes increase with age, we couldn’t use
the data in this list to reinforce that assumption, because there could be alterna-
tive reasons for the number of strokes. All the younger persons may have left the
community for work elsewhere, for example, which would mean that the denomi-
nators would be smaller, but the rates might be higher.
The point I’m making is always to notice whether a report uses proportions or
rates when talking about risk. Keep in mind that proportions cannot be used to
make statements regarding risk.
Let’s pause now so you can answer some more questions about what you have
just learned.

Practice: Ratios, Proportions, and Rates

Exercise 1

Relative Risk (RR)

Relative risk is the first measure of association we will
consider. It is a ratio of the risk of an event (or of develop-
ing a disease) in two groups: persons who are and who
are not exposed to some factor. It can be expressed, or
written, as the incidence in the exposed divided by the
incidence in the non-exposed.
The size of the relative risk is one of the criteria we use
in estimating the strength of a causal relationship between
an exposure and an outcome. The larger the relative risk,

Northwest Center for Public Health Practice 5

Measuring Risk in Epidemiology
Transcript a b
c d

the stronger the evidence is for a causal relationship, all other things being equal. For
example, a relative risk of 10 provides better evidence for a causal relationship than
would a relative risk of 3. But remember, a large relative risk alone does not estab-
lish a causal relationship.
If the relative risk is equal to 1, we say there is no evidence of an association. If
the relative risk is greater than 1, than the exposure is harmful, and if the relative risk
is significantly less that one, we can say there is evidence that the exposure may be
protective.
Now let’s look in detail at how you calculate relative risk.

Calculating Relative Risk

As I said, we calculate relative risk by dividing the inci-
dence in the exposed group by the incidence in the non-
exposed groups. In this fictitious example of a cohort study,
we found that in a factory of 3000 workers, 1000 workers
were exposed to toxic substances, and 2000 were not.
We’ll use this 2x2 table to calculate the relative risk
of developing lung disease among exposed workers
compared to unexposed workers.
During the study period, 800 of the exposed workers
developed some type of lung disease, and only 40 of the
non-exposed workers developed a lung disease .
The incidence of lung disease during the study period among exposed work-
ers was 800 per 1000 person per year. The incidence of lung disease among non-
exposed workers during this time was 40 per 2000. So that we have comparable
rates, we’ll convert this to 20 per 1000. By subtracting both 800 and 40 from their
respective row totals, we can get the number of persons with no lung disease in
each exposure row.
Note that the incidence in exposed workers was 800/1000, or 80%. The inci-
dence in those not exposed was 40/2000, or 2%. Comparing the incidence in the
exposed to the incidence in the non-exposed, we get 80/2, for a relative risk of 40.
We could also say that workers exposed to toxic substances would be forty times
more likely than non-exposed workers to develop lung disease.
This elevated relative risk strongly suggests, but does not prove, a causal relation-
ship between exposure to this toxic substance and developing lung disease.
Notice that this is an example of a cohort study. In case-control studies, relative
risk cannot be calculated directly. It must be estimated using an odds ratio. We’ll
discuss odds ratios later on in this module.

Northwest Center for Public Health Practice 6

Measuring Risk in Epidemiology
Transcript a b
c d

Relative Risk Review

Let’s review a few key points about relative risk. Rel-
ative risk is the likelihood of an event in people who
are exposed to some factor compared to the likelihood
in another group that is not exposed. Relative risk is a
ratio of two incidence rates. To calculate it, you divide
the incidence in the exposed by the incidence in the
non-exposed. 
Early in my work as an epidemiologist, I was part of a
team investigating a large outbreak of St. Louis encephali-
tis, known as SLE, in Houston, Texas. This serious illness is
transmitted through the bite of a mosquito.
We found that the incidence of SLE in people living in central Houston who were
more likely to be exposed to standing water, where mosquitoes could breed, was
13 times the incidence in people living in neighborhoods on the outskirts of the city
who had little or no exposure to standing water. After calculating the relative risk
of SLE relative to exposure to standing water, we concluded that because the rela-
tive risk was higher in the central part of the city, those neighborhoods might benefit
from spraying with chemicals to kill mosquito larvae.
Let’s pause now so you can answer some more questions on what you have just
learned.

Practice: Calculating Relative Risk (part 1)

Practice: Calculating Relative Risk (part 2)

Exercise 2

Attributable Risk (AR)

Remember that relative risk is the likelihood of an event
(or of developing a disease) in persons who are exposed to
some factor, as compared to another group which is not.
Attributable risk is more concrete. It is expressed as
an absolute rate of disease, rather than as a ratio. It’s the
rate, or incidence, of disease in an exposed group that
is attributable to the exposure. Attributable risk in an
exposed group gives us the absolute amount of disease
(or the number of cases) resulting from the exposure. We
express it as cases per multiple of the population. We can

Northwest Center for Public Health Practice 7

Measuring Risk in Epidemiology
Transcript a b
c d

use attributable risk to determine the amount of disease that could theoretically be
eliminated or prevented by removing the exposure in the exposed population.

Calculating Attributable Risk

We calculate attributable risk by subtracting the incidence
in the unexposed from the incidence in the exposed.
Let’s look at the example of factory workers exposed
to a toxic substance and who develop lung disease to see
how this works.
Remember, in our fictitious cohort study, 800 of the
exposed workers developed some type of lung disease
during the study period, and 40 of the non-exposed work-
ers developed a lung disease. So the incidence of lung
disease among exposed workers was 800/1000 and the
incidence in non-exposed workers was 40 per 2000. Since
we have to use incidence rates with the same denominators, we’ll convert this to
20 over 1000. Plugging these numbers into the attributable risk equation, we get an
attributable risk of 780 cases of lung disease per 1000 factory workers exposed to
the toxic substance attributable to their exposure.
Attributable risk can be used to uncover unsuspected features of associations
between exposures and diseases. Let’s look at an example to clarify this important
idea.

Usefulness of Attributable Risk

In 1956, a study of lung cancer and smoking in British
physicians reported these results. The study showed that
although heavy smokers face a relative risk of dying from
lung cancer 24 times higher than non-smokers, they face a
relative risk of only 1.4 times of dying of heart disease, as
compared to non-smokers.
Remember, to calculate the relative risk of death from
lung cancer, you divide the death rate from lung cancer in
the heavy smokers by the death rate from lung cancer in
the non-smokers; and to calculate the relative risk of dying
from heart disease, you divide the death rate from heart
disease in the heavy smokers by that in the non-smokers.
By the way, we use death rates here in the same way as we’d use incidence rates
because they are a form of incidence rate.
The point I want to make here is that if we examine only relative risk, we’ll miss a
crucial point.

Northwest Center for Public Health Practice 8

Measuring Risk in Epidemiology
Transcript a b
c d

By calculating the attributable risk for lung cancer deaths in heavy smokers as
compared to non-smokers (that is, subtracting the lung cancer death rate in the
non-smokers from that in the heavy smokers), and the attributable risk for deaths
from heart disease, we can see that over all, heavy smokers are more likely to die of
heart disease than of lung cancer, because the death rate from heart disease is much
higher than the death rate from lung cancer in both groups.
Heart disease presumably has many other causes besides heavy smoking.
However, if one were to eliminate smoking, the attributable risk tells us that more
people would be saved from death by heart disease than from death by lung cancer.
This information would be important in arguing for funding a big anti-smok-
ing campaign, since it would provide evidence that many additional lives could be
saved by removing smoking as a risk factor for heart disease death as well as for
death from lung cancer.

Attributable Risk Review

Let’s review attributable risk before we go on. Attributable
risk is an absolute rate of disease, rather than a ratio. It’s
the incidence of disease in an exposed group that is attrib-
utable to, or thought to be caused by, the exposure. The
calculation is simple. We subtract the incidence rate in
the unexposed from the incidence rate in the exposed.
In the St. Louis encephalitis outbreak in Houston that I
mentioned earlier, we would have calculated the inci-
dence rate of encephalitis that theoretically could have
been reduced by spraying mosquito larvicide in the high-
incidence neighborhoods and discovered that 71 cases per
100,000 people living in the high risk areas would theoretically have been prevented
by the spraying if standing water were the only cause of the difference in incidence.
Let’s pause now so you can answer some questions on what you have just
learned.

Exercise 3

Calculating Attributable Risk Percent

Sometimes it is more useful to express attributable risk in terms of a percent, or
proportion. Attributable risk can be calculated as a percentage by subtracting the risk
in the unexposed from the risk in the exposed and dividing the result by the risk in
the exposed.
Here’s an example of that calculation using the attributable risks from the British
study of deaths due to lung cancer and smoking that we just looked at. For lung

Northwest Center for Public Health Practice 9

Measuring Risk in Epidemiology
Transcript a b
c d

cancer deaths, the attributable risk percent is 96%. This

figure is known as the attributable risk percent in the
exposed.
For heart disease deaths, the attributable risk percent
is 30%.
When we compare the attributable risk percents for
deaths due to lung cancer and to heart disease, it appears
that eliminating smoking would reduce a larger percent-
age of deaths due to lung cancer than of deaths due to
heart disease in smokers, but as I said before, since the
frequency of heart disease deaths in the general popula-
tion is so much greater than the frequency of lung cancer
deaths, more heart disease deaths would be prevented,
even though the relative risk of dying of heart disease in smokers, as compared to
non-smokers, is smaller.

Calculating Population Attributable Risk

If we know the death rate due to lung cancer in the
general population and in non-smokers, we can also
calculate the risk in the general population of lung cancer
deaths attributable to smoking. To do this we take the lung
cancer death rate in the general population and subtract
the lung cancer death rate in the non-smokers. The result
is the population attributable risk (or PAR). This rate tells
us the number of lung cancer deaths that would be elimi-
nated from the general population if the exposure, in this
case, smoking, were eliminated.
Suppose we know that the lung cancer death rate in the
general population is 62 per 100,000 persons and in non-smokers is 7 per 100,000
persons. The rate of lung cancer deaths in the general population attributable to
smoking is 62 minus 7 or 55 per 100,000 persons per year.
Population attributable risk is helpful for guiding public health decisions about
where to focus resources most effectively to protect the public’s health.
As with attributable risk, population attributable risk can be expressed as a
percent.
We calculate it in a way similar to calculating attributable risk percent: the inci-
dence in the general population minus the incidence in the unexposed population
divided by the incidence in the general population.
Turning our smoking example into a population attributable risk percent, we can
say that of the 62 per 100,000 deaths per year due to lung cancer in the general
population, 62 minus 7 divided by 62, or 89%, can be attributed to smoking.

Northwest Center for Public Health Practice 10

Measuring Risk in Epidemiology
Transcript a b
c d

In this example, I have assumed that all persons are either smokers or non-smok-
ers, and I have not taken secondary smoke exposure into consideration.

Uses of RR and PAR in PH Practice

As I mentioned before, relative risk is important for infer-
ring causation. For example, public health researchers
looking at the cause or source of a disease may use it to
evaluate the role of an exposure in causing the disease.
Population attributable risk, on the other hand, is
more important to public health practitioners and policy
makers since they want to know the numbers of deaths or
illnesses prevented in order to use their resources to great-
est advantage to control or prevent disease in populations
in their jurisdictions. The size of the problem in actual
numbers counts for much when allocating limited public
health resources.
Let’s look at our factory workers again to clarify this point.
Suppose the factory is located in a county of 100,000 people. We know that the
relative risk of developing lung disease for factory workers exposed to this toxic
substance compared to other factory workers is 40, as we showed earlier. This high
relative risk is very suggestive that the exposure is causal. But how big is this problem
in the rest of the county?
When we calculate the attributable risk percent of the exposed factory workers,
we get 98 percent. So we can say that 98 percent of the lung disease in the exposed
group is attributable to their exposure.
If we assume that the incidence of lung disease in the rest of the county is the
same as that in the non-exposed factory workers, we find that the overall incidence
in the community is 27.8 per thousand persons per year. Keep in mind that this
figure includes the factory workers because they’re part of the county population.
The population attributable risk is the incidence in the entire population (27.8
per thousand per year) minus the incidence in the non-exposed (20 per thousand
per year), or 7.8 cases per thousand persons per year. The population attributable
risk percent is the PAR divided by the incidence in the total population (2.8/27.8),
or 28%. We can say that in this community, 28% of the cases of lung disease are
attributable to the exposure in the factory. Expressed another way, if we eliminate
the exposure to toxic substances in the factory, we can reduce lung disease in the
county by 28%.
Clearly, exposed workers in the plant have a high risk for developing lung disease,
but if we look at the county as a whole, we can see that this risk does not constitute
as large a public health problem as the exposed factory workers’ attributable risk
percent of 98 would suggest.

Northwest Center for Public Health Practice 11

Measuring Risk in Epidemiology
Transcript a b
c d

Measures of Association Review

Let’s take a moment to review the measures of association
we’ve covered.
Attributable risk, or AR, is the absolute rate of a health
event in people who have been exposed to a risk factor
that is attributable to the exposure.
We can express attributable risk as a percentage of the
overall risk by subtracting the incidence rate in the unex-
posed from the incidence rate in the exposed, which gives
us the AR, and then dividing the resulting AR by the inci-
dence rate in the exposed.
Population attributable risk, or PAR, on the other
hand, tells us the amount of a disease that would be eliminated from the general
population if the exposure to that disease were eliminated.
We calculate the PAR by subtracting the incidence in the non-exposed from the
incidence in the entire population.
Why would we want to know the population attributable risk? Because it can help
guide public health decisions about where to focus resources most effectively. In
Houston, for example, the incidence of St. Louis encephalitis, or SLE, in the entire
city was 34 per 100,000. We can calculate the risk of getting SLE for the general
population by subtracting the incidence rate in the people who weren’t exposed
to standing water, which was 6 per 100,000, from the incidence rate for the entire
city. And we get a PAR of 28 cases per 100,000 people that could theoretically be
eliminated if standing water were eliminated, and if exposure to standing water was
responsible for the entire excess in SLE incidence.
As with the AR percent, to turn the PAR into a population attributable risk percent,
we divide the PAR by the incidence rate in the exposed population.
Let’s pause now so you can answer some questions on what you have just
learned.

Practice: Measures of Association

Exercise 4

Odds Ratio (OR)

So far, the calculations we’ve discussed are useful when
incidence rates (perhaps from cohort studies) are available,
since you can calculate relative risk directly.
However, since cohort studies are time-consuming and
expensive, epidemiologists often investigate association

Northwest Center for Public Health Practice 12

Measuring Risk in Epidemiology
Transcript a b
c d

and risk using data from case-control studies. Case-control studies, though, do not
produce incidence rates, so we cannot calculate a true relative risk.
Fortunately, case-control studies can produce odds ratios, which approximate rela-
tive risks in certain situations: when the cases are from the same populations as the
controls, and when the disease is relatively rare in the population (say, less than 5%).
Let’s use an example to see how to calculate odds ratios.

Calculating Odds Ratio

When we examine the rates of death by suicide among
men and women living in Washington State between
2003 and 2005, we find that men had a higher suicide
rate. Suppose we have an idea that the higher rate in men
might have something to do with whether they were living
alone. We decide to do a case-control study and interview
100 family members of persons who died by suicide in
2003 and later, and 100 family members of persons of the
same age who did not die from suicide.
To calculate the odds ratio, we multiply the number in
cell a (the suicides who lived alone) times the number
in cell d (the non-suicides who didn’t live alone) and the numbers in cells b (non-
suicides living alone) and c (suicides not living alone). We then divide the product
ad by the product bc. Here is the calculation.
This ratio is our estimation of the relative risk for suicide for men living alone as
compared to men not living alone. We could say that living alone increases the risk
for suicide in men by a factor of 6.8.

Odds Ratio Reviewed

Odds ratios approximate, or allow you to estimate, relative
risk when incidence data in exposed and non-exposed
populations are not available. We also need to make two
assumptions: the cases are from the same populations
as the controls and the disease is relatively rare in the
population.
Odds ratios are calculated from case-control stud-
ies. We usually set up the data in a 2x2 table and do the
calculations by cross-multiplying the data in the corner
cells, and dividing the products. The number of controls
usually equals the number of cases and are expressed in
the column totals.
In public health practice, one of the most common uses of an odds ratio is in
determining a possibly responsible food item in a restaurant-based food-borne

Northwest Center for Public Health Practice 13

Measuring Risk in Epidemiology
Transcript a b
c d

outbreak. A health department investigator questions both ill and well people who
ate at the restaurant before the beginning of the outbreak. The investigator then
compares the food consumption history for each food in people who are and who
are not ill to try to identify the food that might be responsible for the illness.
Let’s pause now so you can answer some questions on what you have just
learned.

Practice: Calculating Odds Ratio (part 1)

Practice: Calculating Odds Ratio (part 2)

Summary
To summarize, in this module we’ve talked about risk and
association as used by epidemiologists. Risk is the proba-
bility of an outcome (often a negative outcome) in a speci-
fied period of time.
Relative risk is the ratio of the incidence in the exposed
to the incidence in the non-exposed groups. Relative risk
measures the strength of an association and is useful when
looking at possible causes of disease. A larger relative risk
provides stronger evidence for a causal role for the expo-
sure but does not prove causation.
Attributable risk and population attributable risk
measure the absolute amount of disease attributable to
an exposure in either the exposed or total population. Both can be calculated as
percents. Public health policy makers use these measures to help decide where to
allocate scarce resources.
An odds ratio is an estimate of relative risk, using data from case-control studies.
Odds ratios are used when we don’t have actual incidence rates. As with relative
risk, the larger the odds ratio, the stronger the evidence
for a causal relationship between an exposure and the
outcome of interest.

Resources
If you would like to learn more about the concepts in this
module, you might want to explore some of the resources
listed here.
Now, if you’re ready, please go on to the final
assessment.

Northwest Center for Public Health Practice 14