Measures of Association

Dr Gail Davey
Epidemiology 1 (COMH 603)
Tuesday 10th May 2005
 Measures of Association
Objectives
By the end of this session, students should
be able to:
 Define, calculate and interpret relative risk

and excess risk from simple data;

 Define and interpret population

attributable risk (fraction);

 Outline the advantages and disadvantages

of each of these measures.

 Measures of Association
Why do we need them?

 Move from descriptive to analytical

epidemiology.
 Comparisons within and between
populations.
 Risk factor identification.
 Measures of Association

Ratio Measures: relative risk

Difference Measures: excess risk

These measures may be applied at:

 Individual Level

 Population Level
 Measures of Association
Relative Risk

Ratio Measures, which include:

 Rate ratio;

 Risk ratio;

 Incidence odds ratio; and

 Prevalence odds ratio

(depending on the type of study data

arise from)
 Measures of Association
Longitudinal Studies
Risk ratio =
cumulative incidence in exposed group
cumulative incidence in unexposed gp

Rate ratio =
incidence rate in exposed group
incidence rate in unexposed group
 Measures of Association
Longitudinal Studies
Risk ratio in the Jimma Birth Cohort.

Maternal Number Number of IMR per Risk Ratio

education of infants deaths 1000

Illiterate 5021 579 115.3 1.59

Junior + 1368 99 72.4 1.00

 Measures of Association
Longitudinal Studies
Another risk ratio in the Jimma Birth Cohort

ANC Number of Number of IMR per Risk Ratio

attendance infants deaths 1000
Yes 4262 354

No 4011 493
 Measures of Association
Longitudinal Studies
Another risk ratio in the Jimma Birth Cohort

ANC Number of Number of IMR per Risk Ratio

attendance infants deaths 1000
Yes 4262 354 83.1

No 4011 493 122.9

 Measures of Association
Longitudinal Studies
Another risk ratio in the Jimma Birth Cohort

ANC Number of Number of IMR per Risk Ratio

attendance infants deaths 1000
Yes 4262 354 83.1 1.00

No 4011 493 122.9 1.48

 Measures of Association
Longitudinal Studies
Rate ratio in ENARP cohort
HIV status Number PY Incidence Rate ratio
dev TB observed Rate/1000
PYO
Negative 14 2054

Positive 10 222
 Measures of Association
Longitudinal Studies
Rate ratio in ENARP cohort
HIV status Number PY Incidence Rate ratio
dev TB observed Rate/1000
PYO
Negative 14 2054 6.8

Positive 10 222 45.1

 Measures of Association
Longitudinal Studies
Rate ratio in ENARP cohort
HIV status Number PY Incidence Rate ratio
dev TB observed Rate/1000
PYO
Negative 14 2054 6.8 1.00

Positive 10 222 45.1 6.63

 Measures of Association
Cross-sectional Studies

Prevalence odds ratio

= prevalence odds in exposed group
prevalence odds in unexposed group

= exposure odds in diseased group

exposure odds in undiseased group
 Measures of Association
Cross-sectional Studies
Prevalence odds in Butajira Women’s
Health & Life Events Study
Physical Health Health Odds Ratio
Violence poor good
No 566 594 1.00

Yes 677 424 1.68

 Measures of Association
Cross-sectional Studies
Prevalence odds in Butajira Women’s
Health & Life Events Study
Physical Depressive No dep. Odds Ratio
Violence disorder disorder

No 38 1122 1.00

Yes 81 1020
 Measures of Association
Cross-sectional Studies
Prevalence odds in Butajira Women’s
Health & Life Events Study
Physical Depressive No dep. Odds Ratio
Violence disorder disorder

No 38 1122 1.00

Yes 81 1020 2.34

 Measures of Association
Cross-sectional Studies
Prevalence odds in HIV behavioural
survey, Asosa.
Education level VCT willing VCT unwilling Odds Ratio

Secondary 193 238 1.00

Elementary 132 198

Not literate 38 84
 Measures of Association
Cross-sectional Studies
Prevalence odds in HIV behavioural
survey, Asosa.
Education level VCT willing VCT unwilling Odds Ratio

Secondary 193 238 1.00

Elementary 132 198 0.82

Not literate 38 84 0.56

 Measures of Association
Note on ‘Odds Ratios’

Relative risk (RR) is calculated as the

cross-product ratio in cross-sectional
and case-control studies.
The exact interpretation of the RR in a
case-control study depends on how
the cases and controls are selected.
So, be careful what you call the cross-
product ratio in case-control studies.
 Measures of Association
Interpretation
A. RR = 0.87
B. RR = 4.25
C. RR = 1.86

Which RR suggests the strongest

association? Why?
Which RR suggests a protective effect?
Why?
 Advantages and Disadvantages of
Relative Risks

Advantages Disadvantages

Can be estimated from all

common study designs
Can be compared between
studies
Used as a measure of
strength of association
 Advantages and Disadvantages of
Relative Risks

Advantages Disadvantages

Estimated from all Poor guide for public

common study designs health policy
Can be compared
between studies
Used as a measure of
strength of association
 Measures of Association
Excess Risk
Difference Measures which include:
 Risk Difference

 Rate Difference

(derived from longitudinal studies)

 Prevalence Difference

(derived from cross-sectional

studies, rarely used)
 Measures of Association
Rate and risk difference
Rate difference =
Inc. rate in exposed - inc. rate in unexposed

Risk difference =
Cumulative inc. in exposed – cumulative inc.
in unexposed

Units as per original rates or risks.

 Measures of Association
Rate difference
Rate difference in ENARP cohort=
Rate TB in HIV pos - Rate of TB in HIV neg
= 45.1/1000 PYO – 6.8/1000 PYO
= 38.3/1000 PYO

Interpretation: 38.3 cases of TB per 1000

person years observed are attributable to
underlying HIV infection in the ENARP
cohort.
 Measures of Association
Risk difference
Risk difference in Jimma Birth Cohort
= IMR in uneducated mothers – IMR in
educated mothers
= 115.3 per 1000 – 72.4 per 1000
= 42.9 per 1000

Interpretation: 42.9 infant deaths per 1000

live births are attributable to lack of
maternal education in Jimma.
 Measures of Association
Population Level Measures

Measures of public health impact must

take into account –
 Rates of disease in exposed groups

 Rates of disease in unexposed

groups, AND
 Proportion of population exposed.
 Measures of Association
Population Attributable Risk

The amount of disease that would be

averted if the whole population were
unexposed.

PAR = rate or risk diff x proportion exposed

= inc. (population) – inc. (unexposed)
 Measures of Association
Population Attributable Risk
PAR in ENARP cohort
= rate difference x proportion exposed
= 38.3/1000 PYO x 0.118
= 4.5/1000 PYO

Interpretation:
 Measures of Association
Population Attributable Risk
PAR in ENARP cohort
= rate difference x proportion exposed
= 38.3/1000 PYO x 0.118
= 4.5/1000 PYO

Interpretation: if none of ENARP cohort

were ever exposed to HIV, then 4.5
cases of TB disease per 1000 PYO
would be averted.
 Measures of Association
Population Attributable Risk
PAR in Jimma Birth Cohort
= risk difference x proportion exposed
= 42.9/1000 x 0.608
= 26.1/1000

Interpretation:
 Measures of Association
Population Attributable Risk
PAR in Jimma Birth Cohort
= risk difference x proportion exposed
= 42.9/1000 x 0.608
= 26.1/1000

Interpretation: if all mothers in Jimma

were educated, 26.1 infant deaths
per 1000 live births would be averted
 Measures of Association
Population Attributable Risk Fraction
PARF = PAR x 100%
Inc. in popn

= p(RR – 1) x 100%
p(RR – 1) + 1

So is derivable from case-control

studies as well as cohort studies
 Measures of Association
Population Attributable Risk Fraction
PARF in Jimma Birth Cohort
= 26.2/1000 x 100%
106.2/1000
= 24.7%

Interpretation:
 Measures of Association
Population Attributable Risk Fraction
PARF in Jimma Birth Cohort
= 26.2/1000 x 100%
106.2/1000
= 24.7%

Interpretation: If all mothers in Jimma

were educated, one quarter of the
IMR would be avoided.
 Measures of Association
Confidence Intervals
Derived from
 the sum of the variances of the

component measures (difference

measures);
 the sum of the variance of the

natural log of the component

measures (ratio measures)
But for now, it is enough to know that
they can be calculated!
 Measures of Association
Sources.
Modern Epidemiology (Rothman)
A birth cohort study in South-West Ethiopia to identify factors
associated with infant mortality that are amenable to
intervention (Assefa & al, EJHD 2002, vol 16, special issue)
Behavioural survey for HIV/AIDS in Asosa among the general
population and sex workers. (Eshetu & al, EJHD 2004, vol
18: 75-81)
Low CD4+ count and high HIV viral load precede the
development of tuberculosis disease in a cohort of HIV-
positive Ethiopians. (Wolday & al, EMJ 2003, vol 41, special
issue)
Women’s Health and Life Events Study in Rural Ethiopia.
(Gossaye & al, EJHD 2003, vol 17, special issues)