Evaluating the Role of

Confounding
Confounding - Definition and Impact

• An alternate explanation for observed association

between an exposure and disease.
• A mixing of effects. The association between
exposure and disease is distorted because it is
mixed with the effect of another factor that is
associated with the disease.
• Result of confounding is to distort the true
association toward the null (negative confounding)
or away from the null (positive confounding).

Confounding slide 2
 An example: Who can run faster,
men or women?
• Exposure = gender Outcome = speed
• Null Hypothesis: average speed of men = average speed
of women
• All men and women in one town invited to participate in a
road race. On race day, both men and women come and
race. The average running time for the men is faster than
the women.
• CONCLUSION: Men run faster than women because of
their gender.
Confounding slide 3
 An example: Who can run faster,
men or women?

• But Wait! Someone notices that women with

young children did not race. In fact, women who
ran the race were, on average, older than men
who ran. For example, the average age of
women was 50 years while the average age of
men was 25 years.

• CONCLUSION: Perhaps men were faster not

because of their gender, but because they were
younger.
Confounding slide 4
 An example: Who can run faster,
men or women?

• So another race is held, this time making

sure ages in the two compared groups
(men and women) are comparable. That is,
the men and women have same
distribution of ages.
• Race result: Once again, men are faster.
• CONCLUSION: Controlling for age, men
are still faster than women. Confounding slide 5
• BUT WAIT! Someone points out that the
men are, on average, taller than the women.

• CONCLUSION: Perhaps men were faster

not due to their gender, but because their
legs are longer.

• So another race is held, this time making

sure heights and ages in the two groups
(men and women) are comparable.

• Race result: Once again, men are faster.

Confounding slide 6
• BUT WAIT! Someone points out that 50%
of the women had hair longer than their
shoulders, and only 5% of the men did!

• CONCLUSION??? Long hair made the

women run slower. Is this a reasonable
conclusion?

Confounding slide 7
Lessons from the road race: Criteria
for a characteristic to be a confounder

• In general, for a characteristic to be a

potential confounder, it must be associated
with both the disease (outcome) and the
exposure under study. (Why are age and
height competing explanations, but not
hair length?)

Confounding slide 8
Lesson 2:

Review meaning of association: If characteristic

is associated with disease, then risk of disease
is different among people with the characteristic
compared to those without. If characteristic is
associated with exposure, then the distribution
of the characteristic is different among people
with the exposure compared to people without
exposure (unbalanced between groups).

Confounding slide 9
Criteria for a characteristic to be a confounder

More on variable being associated with disease

– Can be a risk factor, preventive, or correlate
(or marker) for a cause of disease (need not be
cause itself, for example, socioeconomic
status)
– Must be associated with disease independently
of exposure

Example: Age and height are associated with speed

regardless of gender. Taller people (both men and women)
have greater speed. Younger people (both men and women)
have greater speed. Confounding slide 10
Criteria for a characteristic to be a confounder

• More on variable being associated with

exposure. We have talked about the need for
comparability between the exposed and
unexposed groups in experimental and cohort
studies. Recall purpose of randomization in an
experimental study, and selection of the
comparison group in a cohort study.

Confounding slide 11
• A variable cannot be a confounder if it is a step
in the causal chain or pathway.

• Example of step in causal chain: Moderate

alcohol consumption increases serum HDL
levels which, in turn, decrease the risk of heart
disease. HDL level is a step in this causal chain,
not a confounder that needs to be controlled.
Rather, it is something interesting that helps us
understand the disease mechanism.

Confounding slide 12
• Examples of confounding

– Smoking is confounder of effect of

occupational exposures (to dyes) on
bladder cancer

– Age is confounder of effect of DDT

pesticide exposure and breast cancer

Confounding slide 13
Controlling for confounding, in general

• Why control? Confounding factors are

nuisance variables. They get in the way of
the relation you want to study. You want to
remove their effect. We have done this
already in age-standardization

Confounding slide 14
Controlling for confounding, in general

• Confounding can be controlled in the design

and the analysis phase: To control for
confounding you must have information on the
variables that are potential confounders. This is
true for all methods of controlling confounding
except randomization.

Confounding slide 15
Controlling for confounding, in general

• Which variables are potential confounders?

Usually, risk factors for disease.

• Is a variable a confounder in your data or

not? Compare crude and adjusted
measures of association. If they differ
appreciably, the answer is “yes.”
Confounding slide 16
 Controlling for confounding in the
design phase

• Randomization - with sufficient sample

size, randomization is likely to control for
both known and unknown confounders.
Remember “likely to control” means just
that. It’s not a guarantee.

Confounding slide 17
 Controlling for confounding in the
design phase

• Restriction - restrict admissibility criteria for

study subjects and limit entrance to
individuals who fall within a specified
category of the confounder

• Example: In the road race, you can

restrict the race to people in a certain age
range (say, 25-30) or to people in given
height range.
Confounding slide 18
 Controlling for confounding in the
design phase

• Goal is to eliminate or reduce variation in the

level of the confounding factor between
compared groups

• Remember, a variable can only be a

confounder if it is different between compared
groups. If you restrict the study to one
category of a confounder, then compared
groups cannot differ.
Confounding slide 19
 Controlling for confounding in the
design phase

• Advantages of restriction: straightforward,

convenient, inexpensive

• Drawback of restriction: limits generalizability

Confounding slide 20
 Controlling for confounding in the
design phase

• Matching -- select study subjects so that

the potential confounders are distributed in
an identical manner among the exposed
and unexposed groups (cohort study) or
among the cases and controls (case
control study)

Confounding slide 21
• Example: matching in cohort study of exercise
and heart attack.
• Two groups: exercisers and non-exercisers
• Confounders to be matched: age, sex, smoking
• Exposed subject is a 45 year old female who
doesn’t smoke
• Thus, you need to find an unexposed subject
who is a 45 year old female who doesn’t smoke.
(Can loosen the age match to 45 + or – a
couple of years)
Confounding slide 22
Controlling for confounding in the design phase

• In a cohort study, if the exposed and

unexposed groups do not differ on the
distribution of a variable, then that variable
cannot be a confounder. Thus, the crude
measure of association from matched
cohort study is unconfounded by matching
variable.

Confounding slide 23
Controlling for confounding in the design phase

• In a case control study, matching does not in

itself control confounding. Matched analysis
must be also performed also. The crude
measure of association from a matched case-
control study will be biased. An adjusted
estimate (using matched analysis) is OK. Point
of matching in case-control study: to make sure
there are enough subjects in each category of
confounding variable to do an adjusted
analysis.
Confounding slide 24
 Controlling for confounding in the
analysis: stratification

• Definition: evaluate the association within

homogeneous categories (strata) of the
confounding variable

Example: Case control study of oral

contraceptive use and risk of heart attack.
Age is a confounder.
Confounding slide 25
Controlling for confounding in the
analysis: stratification

TOTAL DATA (One 2 x 2 Table)

Case Control
OC Use Yes 39 24

No 114 154

Crude OR = 2.2
Confounding slide 26
 Stratified Data (Two 2 x 2 Tables)

Age< 40 Age 40 and over

Case Control Case Control

OC Yes 21 17 OC Yes 18 7
Use Use
No 26 59 No 88 95

Stratum-specific Stratum-specific
OR = 2.8 OR = 2.8
Note each stratum is like a restricted analysis. There is a narrow
range of the confounder.The stratum specific ORs (2.8) differ from
the crude OR (2.2) by about 25%. This difference indicates that
there is confounding by age. Confounding slide 27
• Pooling stratum-specific estimate into one estimate:
various methods including the Mantel-Haenszel method.

• What if stratum-specific estimates are appreciably

different? This is effect measure modification (see
Chapter 14, for more details).

– Definition of effect measure modification: when

magnitude of effect of an exposure varies by levels of
a third variable.

Example: The relation between body mass index (a

measure of obesity) and breast cancer varies according
to menopausal status. Among pre-menopausal women,
higher BMI decreases risk. Among post-menopausal
women, higher BMI increases (or does not affect) risk.
Confounding slide 28
• It is unusual for stratum-specific measures
of association to be exactly the same,
even when there is no effect measure
modification. How do you tell when the
difference between the stratum-specific
measures of association are big enough to
suggest effect modification? You often
make a subjective judgement, although
there are statistical tests to help.

Confounding slide 29
Summary of stratified analysis:

• Purpose: To control confounding and to

assess effect modification.

• If stratum-specific measures of association are

roughly similar, they can be combined into one
pooled estimate. This is adjusted for the
variable on which the data are stratified.

Confounding slide 30
Summary of stratified analysis (cont’d):

• Next step is to compare the adjusted estimate

with the crude estimate. If they are different,
then confounding has occurred, and the
adjusted estimate is correct. If crude and
adjusted estimates are roughly the same, then
no confounding has occurred.

Confounding slide 31
Summary of stratified analysis (cont’d):

• If the stratum-specific estimates differ

appreciably from one another, then effect
measure modification is present. This should
be described by reporting all stratum-specific
estimates separately.

Confounding slide 32
 Controlling for confounding in the
analysis: multivariate analysis

• The limitations of stratification: It is difficult

to control for many variables
simultaneously because a large number of
strata will be generated relative to the
number of study subjects.

Confounding slide 33
 Controlling for confounding in the
analysis: multivariate analysis

Example: Case-control study of physical activity and

heart disease. If you stratify simultaneously by
gender (two categories: male and female), age
(five categories: 40-49, 50-59, 60-69, 70-79, 80
and over), and smoking status (three categories:
never smoked, light smoker, heavy smoker) you
will end up with 30 strata.
If you have a study with a few hundred cases and
controls, you will end up with small numbers or
even zeros in many cells.
Confounding slide 34
Solution: multivariate analysis
– This is an analysis technique that
simultaneously adjusts for several variables
– Involves construction of mathematical model
that describes the association between
exposure, disease, and confounders

Examples: multiple linear regression for

continuous variables, logistic regression for
case-control data, Cox proportional hazards
model for cohort data
Confounding slide 35
 Residual Confounding

• There may be residual or remaining confounding

even after many confounding variables have
been controlled.

• Sources: confounders for which no data were

collected, use of broad categories of a
confounder in your analysis, inaccurate data on a
confounder.

• Residual confounding should be addressed in the

discussion section of a published paper.
Confounding slide 36
 Summary of Confounding

• Mixing of effect between and exposure,

an outcome, and a third variable known
as a confounder

• Considered a nuisance

• Studies may have a small, moderate or

large degree of confounding
Confounding slide 37
 Summary of Confounding

• Confounding can either exaggerate or

minimize the true association

• Epidemiologists have developed many

methods to control confounding in the
design and analysis

Confounding slide 38