Measure of association
Tarekegn Solomon (MPH, PhD)
Assistant professor of Epidemiology
1
Learning objective:
At the end of this session the student will be able to:
• Describe common measures of association and
measures of public health impact
• Calculate and interpret relative risk and odds ratio
and describe their use
• Calculate and interpret Attributable Risk (AR), AR
% , Population AR and Population AR%
2
Measure of association
• Association means correlation or relationship
between two or more variables.
• It summarize the frequency measures of two
variables into a single summary parameter that
estimates the degree of association
• It quantifies the relationship between exposure
and disease among two or more groups.
3
Measure of association…
• Example of measures of association
• chi-square
• relative risk
• odds ratio.
• Chi-square measures the presence or absence of statistically
significant association between exposure and outcome variables
but not strength of association.
• Relative risk and odds ratio measures both the presence and
strength of association.
4
Measure of association…
• Epidemiologic data are often presented in the form
of two by two (contingency) table
• Generally we can have r-by-c table
Disease
Yes(+) No(-) Total
Yes(+) a b a+b
Exposure
No(-) c d c+d
Total a+c b+d a+b+c+d
5
Measure of association…
a = number of individuals who are exposed and
have the disease
b = number of individuals who are exposed and do not
have the disease
c = number of individuals who are not exposed and
have the disease
d = number of individuals who are both non-exposed
and non-diseased
6
Measure of association…
a + b = the total number of individuals exposed
c + d = the total number of unexposed
a + c = the total number with the disease
b + d = the total number without the disease
a + b + c + d = sum of all four cells and the total sample
size for the study
7
Measure of association…
In cohort studies with variable lengths of
follow-up, a special type of two-by-two table is
used for data presentation, since the numbers
of person-time units for exposed and non-
exposed subjects are provided rather than the
total numbers of individuals in each group
8
Measure of association…
Disease Person-time units
Yes No
Exposure Yes a - PY1
No c - PYo
a+c - PY1 + PYo
9
1. Relative risk
• Indicate the likelihood of developing the disease in the
exposed group relative to those who are not exposed
• For a cohort study with count data in the denominator,
the RR is calculated as the ratio of the cumulative
incidence among those exposed compared with those
not exposed
Incidence rate among exposed Ie a/(a+b)
RR= = =
Incidence rate among unexposed Io c/(c+d)
10
RR= risk in exposed
risk in non exposed
• We use RR for cohort or follow up study design
• If RR calculated from cumulative incidence it is called Risk
ratio where as if it is calculated from incidence rate
(person units of time) it is called Rate ratio
11
Example 1 (cohort study)
Breast No Breast
Cancer Cancer Total
Alcohol 70 2,930 3,000
No alcohol 50 2,950 3,000
RR using Cumulative Incidence (CI):
= a/(a + b) = 70 / 3,000 = 1.4
c/(c + d) 50 / 3,000
Interpretation: Women who drink alcohol were 1.4 times at
higher risk of developing breast cancer than who not
drink alcohol
12
Example 2
Table 1: data from a cohort study of oral contraceptive (OC) use and
bacteruria among women aged 16-49 years
____________ ______
Bacteruria
Yes No Total
Current OC use
Yes 27 455 482
No 77 1831 1908
Total 104 2286 2390__
RR using Cumulative Incidence (CI):
RR = a/(a+b) =27/482 =1.4
c/(c+d) 77/1908
Interpretation: Interpretation – OC users had 1.4 times the risk to develop
bacteruria than non-users or
OC users were 40 percent (i.e 1.4 minus the null value of 1.0) more likely
to develop bacteruria than non-users
13
Analyze the following data
We conducted a study to assess the association between
taking anti-malarial pills (E) and the development of
malaria (D) among a group of volunteers in Kenya in
1997
malaria no malaria
Pills 10 990
No pills 60 940
total 70 1930
How would you analyze these data?
Interpret your results.
14
Example 3
Data from a cohort study of postmenopausal hormone use and
coronary heart disease among female nurses
Coronary heart disease
Yes No Person-years
Postmenopausal
hormone use
Yes 30 - 54,308.7
No 60 - 51,477.5
Total 90 105,786.2
15
Example 3 cont…
RR = Ie =IDe = a/PY1 = 30/54,308.7 = 0.5
Io IDo c/PYo 60/51,477.5
Interpretation: women who used postmenopausal
hormones had 0.5 times, or only half, the risk of
developing coronary heart disease compared with
nonusers.
16
Example 4:
Maternal alcohol consumption (E) and development of fetal
alcohol syndrome (D) in babies born in Oslo, Norway from 1990 -
1993.
Exposed Un Exposed Total
Incident cases 60 30 90
Person-yrs 6,000 15,000 21,000
IDE = 60 / 6,000 = 1 / 100 person years
IDUE = 30 / 15,000 = .2 / 100 person years
IDR = 60 / 6000 ÷ 30 / 15000 = 5.0
17
Interpretation of Relative Risk
The strength of association can be considered
• A RR of 1 indicates no association
• RR > 1 indicates a positive association, or an increased risk
among those exposed to a factor.
• RR < 1 indicates negative association, which means the
exposure is preventive
• In general the strength of association can be considered:
High - if the RR is 3.0 or more
Moderate – if the RR is from 1.5 to 2.9
Weak – if the RR is from 1.2 to 1.4
18
2. Odds ratio
• In case control study it is usually not possible to
calculate the rate of development of disease given
the presence or absence of exposure.
• Odds – the ratio of the probability of occurrence of
an event to that of non occurrence
Odds = Probability of the event occurring (P)
Prob of the event not occurring (1-P)
19
OR cont’d…
• Odds ratio for exposure is simply the odds a÷c divided by
the odds b÷d.
Indicates the likelihood of having been exposed
among cases relative to controls
OR =odds in favor of disease when the factor is present
odds in favor of disease when the factor is absent
20
OR cont’d…
• If a and c are very much smaller than b and d then RR is
approximated by OR = ad/bc
• OR can be used as a valid estimate of the RR when:
1. The disease is rare.
2. The controls are representative of the general population
3. The selected cases are representative of all cases
21
OR cont’d…
Example 5 (case control study)
Breast No Breast
Cancer Cancer
Alcohol 70 100
No alcohol 50 140
OR= = = 2.0
axd (70) (140)
bxc (50) (100)
Interpretation: women who were drinking alcohol had a risk of
breast cancer 2 times that of non drinkers
NB: OR used for case control studies because persons are
selected based on disease status so you can’t calculate risk of
getting disease
22
OR cont’d…
• If there is no association, the OR =1
• If a disease is associated with an exposure, then
OR>1
– the odds of exposure in the diseased group will be
higher than the corresponding odds in the non-
diseased group.
• If the exposure is protective against disease, the
OR<1
23
3. Attributable Risk (AR) or Risk
Difference (RD)
• Provides information about the absolute effect of the
exposure or the excess risk of disease in those exposed
compared with those non exposed
• Answer questions such as: how much of the disease that
occurs can be attributed to the exposure under study
• How much disease could be prevented by eliminating the
exposure under study
• Theoretically, the excess incidence that would be prevented
by eliminating the exposure.
24
AR…
• In a cohort study, AR is calculated as the difference of
cumulative incidences (risk difference) or incidence density
(rate difference) depending on the study design
• Is important in clinical practice and public health, useful for
public health planning
• It is calculated as risk difference (RD)
ARE = Incidence(exp) – Incidence(unexp) = [a/(a+b) – c/(c+d)]
AR=Ie-Io
25
AR…
26
27
AR…
Example 6, Refer Example 1 and calculate AR
AR=Ie-Io = 70 / 3,000 - 50 / 3,000
0.0233-0.0166 = 0.0067 = 67/10,000
Interpretation: the excess occurrence of breast cancer among
women who drink alcohol attributable to their
alcohol drink is 67 per 10,000 alcohol drunker
28
AR…
A study was conducted to determine the association
between blood pressure and heart attack. The
following data were obtained:
D non-D
30 970
E
15 985
UE
Calculate and Interpret the Attributable Risk
29
Attributable Risk percent (AR%)
• AR% is an attributable risk expressed as a percentage of
risk in exposed
• What is the proportion of disease among the exposed
which can be attributed to the exposure?
• Estimate the proportion of cases among the exposed
that is attributable to the exposure, or
• the proportion of the disease that could be prevented
by eliminating the exposure among the exposed.
30
AR%...
AR% = =
• Measures public health impact of an exposure
• Synonyms
– Attributable proportion
– Attributable fraction
– Etiologic fraction (EF)
Example 5, Refer table 1 and calculate AR
AR%=Ie-Io = 70 / 3,000 - 50 / 3,000 x 100 = 28.57%
Ie 70/3,000
Interpretation: if the drinking alcohol causes breast cancer, about
28.57% of breast cancer among women who drink alcohol
attributed to their alcohol drink and can be eliminated if they did
not drink alcohol.
31
Population Attributable Risk (PAR)
• We want to study health related problem in a community, But
we do the study in a sample of a population
• What is the excess risk among the general population that is
due to exposure of interest?
• Calculated by:
– Risk in population - Risk in unexposed
– I population- I un exposed
– AR * Proportion of exposed in the population
32
PAR cont’d…
• Measures public health impact of an exposure
• Estimate the excess rate of disease in the total study
population that is attributable to the exposure.
• Indicates the number of cases of the disease in the
total population that can be attributed to the
exposure itself
33
34
Population Attributable Risk Percent
(PAR%)
• Proportion of cases in the population attributable to the
exposure
• PAR expressed as a percentage of total risk in Population
• PAR%= =
P, is proportion of population exposed
35
PAR%
• What proportion of cases is attributed to the
actual exposure among the general population?
• Estimate the proportion of disease in the study
population that is attributable to the exposure
and thus could be eliminated if the exposure
were eliminated.
• Measures public health impact of an exposure
36
37
38
39
40
Example:
From the information below, calculate:
RR, OR, AR, AR%, PAR, PAR%
Smoking Lung cancer No lung Total
cancer
Yes 9 837 846
No 6 1755 1761
Total 15 2592 2607
41
Answer
• I exposed=10.6 per1000
• I unexposed= 3.4 per 1000
• I Population= 15/2607= 5.8 per 1000
• RR= 10.6/3.4= 3.11
• OR= ad/bc=3.15
• AR= 10.6-3.4 = 7.2 per 1000
• AR% =7.2/10.6= 67.9%
• PAR= I Population-I unexposed= 5.8- 3.4 = 2.4 per
1000
• PAR%= 2.4/5.8 = 41%
42
Answer….
AR%: 67%
– Among Smokers, 67% of cases of lung cancer are due to
smoking
PAR= 2.4 per 1000
– Among every 1000 a population, there are 2 excess cases
of lung cancer attributable to Smoking:
– if the entire population stops smoking, then 2 cancer cases
can be averted for every 1000 population
PAR%= 41%
– Making all population stop smoking would eliminate 41%
of all cases of lung cancer in the population.
43
