APPENDIX III-B

CALCULATING AND INTERPRETING

ATTRIBUTABLE RISK AND
POPULATION ATTRIBUTABLE RISK

RISK

Risk as defined for public health planning is the probability of the occurrence of a disease or
other health outcome of interest during a specified period, usually one year. Risk is calculated
by dividing the number who got the disease during the defined period by the total population
of interest during that period. For example if there were 1000 births in a health jurisdiction in
one year and 72 of those babies weighed less that 2500 grams, the risk of low birth weight
(LBW) in the community would be 72/1000 = 0.072 or 7.2%.

However, we seldom rely on a single risk calculation. Most often we are interested in
comparing the occurrence or prevalence of a health condition or problem in two groups, one
who experiences a risk factor or condition (referred to as the exposed) and one who does not
(referred to as the unexposed). This allows us to determine whether a particular exposure does
represent a risk, and if so how big a risk. When conducting a risk analysis, epidemiologists
generally begin by constructing a 2x2 table as illustrated below where ‘a’ represents those in the
exposed population who experienced the outcome of interest and ‘b’ those is the exposed
population who did not experience that outcome. In this case, the risk of exposure is expressed
as ‘a/a+b’. Conversely the risk for those not exposed to the risk factor would be ‘c/c+d’.

Disease or Other Health Outcome

Yes No

Exposure or a b
Yes a + b (n1)
Person,
Place, or Time No c d c + d (n2)

a+c b+d
(m1) (m2)

The risk can vary from 0 to 1 in the case of a risk factor and from –1 to 0 in the case of a
protective factor.

For the rest of this document, we will assume that the data in the 2x2 table represent either the
entire population or a random representative sample of the population since those are the types
of data used by most public health departments in their surveillance activities. The rest of this
discussion will focus on methods for comparing two independent proportions, those exposed to
a risk factor or condition and those not exposed. We have used reference material from
Handler & Rosenberg, “Analytic Methods in Maternal and Child Health” in developing this
resource.i

FHOP Planning Guide 155

RISK RATIO OR RELATIVE RISK

Relative risk is the calculated ratio of incidence rates of a health condition or outcome in two
groups of people, those exposed to a factor of interest and those not exposed. It is used to
determine if exposure to a specific risk factor is associated with an increase, decrease, or no
change in the disease or outcome rate when compared to those without the exposure. Relative
risk is a statistical measure of the strength of the association between a risk factor and an
outcome.

Relative risk can be calculated from a simple 2 X 2 table such as the one above. The formula for
calculating relative risk is:
a

(a + b)
Relative risk =
c

(c + d)

Example: In a particular year in your health jurisdiction, there were 1000 births. In reviewing
their birth certificates, you found that 72 had low birthweights (<2500 gms), and 158 had
mothers who smoked during pregnancy. Of the mothers who smoked during pregnancy, you
found that 19 had low birth weight babies. In order to explore the relationship between
maternal smoking and the occurrence of low birth weight, you would construct the following
table.

Low Birth Weight

Smoked During Pregnancy Yes No
Yes 19 139 158
No 53 789 842
72 928 1000

19
19

(19 + 139) 158

Relative risk = =
53 53

(53 + 789) 842

.120
Relative risk = = 1.9
.063

FHOP Planning Guide 156

In this example, the risk of LBW in smokers is 12% and in non-smokers 6% and the relative risk
of low birthweight associated with cigarette smoking while pregnant is 1.9. In other words,
women who smoke while they are pregnant are about twice as likely as those who do not to
have a low birthweight infant.

When the relative risk associated with a factor is more than 1, then the factor is called a risk
factor. When the relative risk associated with a factor is less than 1, then the factor is called a
protective factor.

If the relative risk equals 1, then factor is not associated with the outcome.

Another statistic similar to relative risk and often used by epidemiologists is the odds ratio:

a/b
c/d

When the difference in incidence or prevalence rate of an outcome is very small, the odds ratio
and the relative risk yield similar results, but when the differences are significant the difference
is much greater with the odds ratio being higher. For routine public health use, the relative risk
and relative prevalence are considered to be preferable to the odds ratio because they are
directly related to the probability of developing or having a health outcome.

Confidence Intervals Around Relative Risk

To calculate the 95% confidence intervals for relative risk, we use the following formula:
CI = (r1/r2) plus or minus 1.96 x square root of {(1/a x b/n1) + (1/c x d//n2)}

Where r1 = a/(a+b) and r2 = c/(c+d)

n1 = total number of births in group 1, those with the risk factor. In this example, it is the
total number of births to smokers, which is 158.

n2 = total number of births in group 2, those without the risk factor. In this example, it is
the total number of non-smokers who gave birth, which is 842.

Using the example above:

r1 = a/(a+b) = 19 / (19 + 139) = .120

r2 = c/(c+d) = 53 / (53 + 789) = . 063

r1 / r2 = .120/.063
= 1.9

(1/a x b/n1) = 1/19 x 139/158

= .052 x .880
= .819

(1/c x d//n2) = 1/52 x 789/842

FHOP Planning Guide 157
 = .019 x .937
= .018

square root of (1/a x b/n1) + (1/c x d//n2)

= square root of .819 + .018
= square root of .837
= .915

CI = (r1/r2) plus or minus 1.96 x square root of (1/a x b/n1) + (1/c x d//n2)
= 1.9 +/- 1.96 x .915
= 1.9 +/- 1.79

Relative Risk = 1.9 (95% CI .11, 3.69)

ATTRIBUTABLE RISK AND

POPULATION ATTRIBUTABLE RISK

As discussed above, relative risk helps you determine whether and how strongly a precursor is
associated with a particular outcome. Attributable risk helps you determine how much of an
outcome may be attributable to a particular risk factor (i.e. an estimate of the excess risk) in a
population exposed to that factor. This is a valuable measure, since it provides estimates of the
relative impact of the poor outcome that could be achieved if the risk factor were reduced or
eliminated. On the other hand, one might want to know the proportion of all cases of an
outcome in the total population that could be attributed to the exposure to the risk factor. This is
called the population attributable risk and when expressed as a percent, the population
attributable risk percent. Calculating the population attributable risk percent allows you to
determine what percent of an outcome could possibly be prevented if a risk factor were to be
removed from the population.

Attributable Risk and Attributable Risk Percent

To calculate the attributable risk, one simply subtracts the risk for the non-exposed group
from the risk for the exposed group. Thus, attributable risk is sometimes called the Risk
Difference, or Excess Risk. The excess risk is “attributed” to the exposure.

Attributable risk (AR) = p1 – p2

Where:

p1 = a/a+b, the proportion of smokers with LBW infants and

p2 = c/c+d, the proportion on nonsmokers with LBW

p1 = a/a+b = 19 /(19 + 139) = .120

p2 = c/(c+d) = 53 /(53 + 789) = .063

AR = .120 - .063

FHOP Planning Guide 158

 = .057

Calculating Confidence Intervals for Attributable Risk

To calculate the 95% confidence intervals for attributable risk, we use the following
formula:

Confidence Interval = Attributable risk +/- 1.96 x Square Root of [p x q (1/n1+ 1/n2)]

p = the risk, which is the number of adverse outcomes divided by the total number
of events. In the example on low birth weight, low birth weight is the adverse
outcome and the total number of events is the total number of births. From the 2x2
tables, we see that there were 72 low birth weight births out of a total of 1000 birth,
so the risk is 72/1000 or .072.

q = 1 – p. In the birth weight example, 1 - .072 = .928

n1 = total number of births in group 1, those with the risk factor. In this example, it is
the total number of births to smokers, which is 158.

n2 = total number of births in group 2, those without the risk factor. In this example,
it is the total number of non-smokers who gave birth, which is 842

(1/n1+ 1/n2) = 1/158 + 1/842

= .006 + .001
= .007

[p * q (1/n1+ 1/n2)] = 0.072 x 0.928 x .007

= .00047, which we can round to .0005

Square Root of [p * q (1/n1+ 1/n2)] = square root of .0005

= .022

Confidence Interval = Attributable risk +/- 1.96 x Square Root of [p * q (1/n1+ 1/n2)]
= .057 +/- 1.96 * .022
= .057 +/- .04

Attributable risk = .057 (95% CI = .014, .100)

Attributable risk is often measured as a percent. The formula for attributable risk
percent (ARP) is:

ARP = (p1 – p2 ) / p1 x 100

Where:

p1 = a/a+b, the proportion of smokers with LBW infants and

p2 = c/c+d, the proportion on nonsmokers with LBW

FHOP Planning Guide 159

 Refer to the data presented in the 2 X 2 table above for smoking during pregnancy and low
birth weight.

Assuming that we are measuring incidence as a percentage, one can calculate the incidence
of low birth weight attributable to smoking during pregnancy:

AR% = (.120 - .063)/.120 x 100

= .475 x 100
= 47.5%

This means that, among newborns with mothers who smoke and had low birth weight
babies, almost half (48%) of those low birth weights are attributable to smoking. Keep in
mind that 6.3% of mothers who don’t smoke will have low birth weight babies, while 12% of
mothers who smoke will have low birth weight babies, indicating that smoking almost
doubles the risk of low birth weight babies.

A fairly simple calculation gives us the number of low birthweight babies where the
condition is attributable to smoking. We know that 158 babies had mothers who smoked
during pregnancy and that 5.7% of those had low birthweights as a consequence of
smoking. Thus, we can estimate that about nine babies had low birthweights as a
consequence of maternal smoking.

Calculating Confidence Intervals for Attributable Risk Percent

The formula for calculating the 95% confidence intervals for the attributable risk percent is:

95% CI = ARP +/- ARP x (C.I. range from the attributable risk / the attributable risk)

Using the numbers from the examples presented so far, we have the following:

ARP = 47.5%
C.I. range from the attributable risk = .04
Attributable risk = .057

Thus, the formula becomes:

95% C.I. = 47.5% +/- 47.5% x (.04/.057)
= 47.5% +/- 47.5% x .7
= 47.5% +/- 33.3

Attributable Risk Percent = 47.5% (95% CI = 14.2%, 80.8%)

POPULATION ATTRIBUTABLE RISK AND

POPULATION ATTRIBUTABLE RISK PERCENT

While attributable risk helps us estimate the excess risk among the exposed that can be
attributed to the risk factor, from a public health perspective it is often more useful to re-define
the attributable risk in terms of the whole population, and thus to know the proportion of cases
in the total population that can be attributed to the risk factor. For this calculation, we use the
FHOP Planning Guide 160
population attributable risk (PAR). Population attributable risk depends not only on the excess
risk imposed by the exposure, but also on the share of the total population that is exposed. Two
formulas can be used to calculate the PAR:

PAR = p0 - p2 or an alternate formula, PAR = (p1 – p2) x n1/N

Where:

p0 = the proportion of ALL births with the outcome of interest (a + c / a + b + c + d)

p1 = the proportion of births with the outcome of interest to mothers WITH the risk factor
(a / a + b)
p2 = the proportion of births with the outcome of interest to mothers WITHOUT the risk
factor (c / c + d)
N = the total number of births (a + b + c + d)
n1= the number of births exposed to the risk factor (a + b)

Using the same data,

PAR = p0 - p2
= 72/1000 – 53/842
= .072 -.063
= .009

or using the alternate formula, we have:

PAR = (p1 – p2) x n1/N

= (.120 - .063) x 158/1000
= .057 x .158
= .009

This means that the overall risk of low birth weight for the total population imposed by
smoking is about nine per 1,000 births. This is a fairly small added risk, mainly because smokers
are a relatively small share of the study overall population.

Another way of looking at the numbers is population attributable risk percent (PAR%), which
tells us the percent of cases in the total population that can be attributed to smoking.

PAR = p0 - p2/p0 x 100

= .072 - .063 / .072 x 100
= 12.5%

This means that out of 72 low birthweight cases, including those born to both smoking and non-
smoking mothers, 9 cases or 12.5% can be attributed to smoking. The calculation helps estimate
the percent of cases in the total population that might be prevented by removing the exposure.

SUMMARY

FHOP Planning Guide 161

To summarize, both relative risk and attributable risk are valuable tools for determining the
contribution of risk factors to an adverse outcome. Relative risk (RR) is a measure of the
strength of the association or causal link between a risk factor and an outcome. Attributable risk
(AR) helps measure the excess risk associated with the risk factor. Population attributable risk
(PAR) gives the added risk in relation to the total population. Population attributable risk
percent (PAR%), gives the percent of cases in the total population that can be attributed to the
risk factor.

The PAR% is an especially useful, and underutilized tool in program planning. It can be used to
predict the impact of public health interventions on adverse outcomes, since it considers both
the excess risk associated with the exposure and the proportion of the population that is
exposed. A risk factor with a large excess risk and widespread exposure poses the most severe
public health risk. One with a relatively small excess risk and relatively rare exposure poses the
lowest public health risk. Factors with small excess risk and wide exposure, or large excess risk
and relatively rare exposure form an intermediate group of public health risks.

The PAR% quantifies the contribution of the risk factor to the outcome and can thus help direct
interventions. The higher the PAR%, the greater the proportion of the outcome that is
attributable to the risk factor. One can compare the values of population attributable risk
percents for selected risk factors to identify those risk factors that are most important for
planning interventions.

Most of the time when we examine risk factors, we look at behaviors, medical conditions, and
environmental factors. It makes sense to talk of "eliminating" these kinds of risk factors. There
are other risk factors, such as race or age that cannot be changed. Instead, identifying people at
risk for an adverse outcome by race or age groups provides populations for targeting
interventions.

It is important to note, however, that population attributable risk percents calculated from a 2 x
2 table are crude measures of attributable risk. Because the outcome is compared to only one
risk factor at a time, there is no way to know if other risk factors may underlie or explain the
associations found in a 2 x 2 table. More advanced statistical methods, such as multivariate
analysis, can be used to calculate attributable risks for individual risk factors that adjust for the
influence of other potential risk factors. Another option is to review the literature for studies on
the outcomes of interest that use multivariate analyses to assess the impact of the risk factors of
interest.

Rosenberg, D. & Handler A., Measures of Association and Hypotheses Testing, In: Analytic Methods in Maternal and Child Health,
i

Handler A., Rosenberg D, Monahan C., & Kennelly J. (eds), Maternal and Child Health Bureau, HRSA, 1998.

FHOP Planning Guide 162