Dr. Yoga Nathan
Senior Lecturer in Public Health
GEMS UL
If exposure X is
associated with
outcome Y…..then
how do we decide
if X is a cause of Y
Applying guidelines
for
causal inference
If exposure X is
associated with
outcome Y…..then
how do we decide
if X is a cause of Y

Two-stage process:

Stage I:
◦ Consider alternative “non-causal explanations” for the
association

In Stage I, we ask ourselves could the association be
due to:
◦ Bias?
◦ Confounding?
◦ Chance?

Stage II: If the association is unlikely to be due to bias,
confounding or chance…
◦ ….we apply ‘guidelines’ for causal inference
Assessing a reported association between an
exposure and an outcome in an epidemiological study
Could the observed
association be due to:
Selection or
measurement bias
No
Stage I
Confounding
No
Chance
Apply Guidelines
for Causal Inference
Probably Not
Could it be causal?
Stage II
“In what circumstances can we pass from an
observed association to a verdict of causation?
Upon what basis should we proceed to do so?”
Nine ‘aspects of an association’ should be
considered before deciding that the most likely
interpretation is causation

Strength

Consistency

Specificity

Plausibility

Coherence


Temporality

Dose-response

Experimental
evidence
Analogy
Chapter 5 pp 83 - 96


Repeated observation of an association in studies
conducted on different populations under different
circumstances
If studies conducted by….
◦ different researchers
◦ at different times
◦ in different settings
◦ on different populations
◦ using different study designs
……all produce consistent results,
this strengthens the argument for causation
Epidemiological studies (1 - 14)
e.g.
The association between cigarette smoking and
lung cancer has been consistently demonstrated in a
number of different types of epidemiological study
(ecological, case-control, cohort)
study


Repeated observation of an association in studies
conducted on different populations under different
circumstances
If studies conducted by….
◦ different researchers
◦ at different times
◦ in different settings
◦ on different populations
◦ using different study designs
……all produce consistent results, this strengthens the
argument for causation

e.g. The association between cigarette smoking and
lung cancer has been consistently demonstrated in a
number of different types of epidemiological study
(ecological, case-control, cohort)

18 studies have investigated the association
between hip fractures (outcome) and water
fluoride level (exposure)
◦ 30 separate statistical analyses

14 analyses produced a ‘positive association’

13 analyses produced a ‘negative association’

3 ‘no association’
The inconsistency of these results casts
doubt on the hypothesis that there is a causal
relationship between fluoride in water and
bone fractures

18 studies have investigated the association
between hip fractures (outcome) and water
fluoride level (exposure)
◦ 30 separate statistical analyses

14 analyses produced a ‘positive association’

13 analyses produced a ‘negative association’

3 ‘no association’

The inconsistency of these results casts
doubt on the hypothesised causal
relationship between fluoride in water and
bone fractures
Community-Based
Case-Control
Cohort
Hospital-Based
Case-Control
Oral Contraceptive Use and
Ovarian Cancer
Hildreth et al,
Rosenberg et al,
La Vecchia et al,
Tzonou et al,
Booth et al,
Hartge et al,
WHO,
Wu et al,
Prazzini et al,
Newhouse et al,
Casagrande et al,
Cramer et al,
Willet et al,
Weiss,
Risch et al,
CASH,
Harlow et al,
Shu et al,
Walnut Creek,
Vessey et al,
Beral et al,
1981
1982
1984
1984
1989
1989
1989
1988
1991
1977
1979
1982
1981
1981
1983
1987
1988
1989
1981
1987
1988
+ ve Association
-ve Association
0.0
0.5
1.0
1.5
2.0
2.5
Relative Risk or Odds Ratio
3.0
3.5
Hankinson SE et al. Obstet Gynecol. 1991;80:708-714.
www.contraceptiononline.
Community-Based
Case-Control
Cohort
Hospital-Based
Case-Control
Oral Contraceptive Use and
Ovarian Cancer
Hildreth et al,
Rosenberg et al,
La Vecchia et al,
Tzonou et al,
Booth et al,
Hartge et al,
WHO,
Wu et al,
Prazzini et al,
Newhouse et al,
Casagrande et al,
Cramer et al,
Willet et al,
Weiss,
Risch et al,
CASH,
Harlow et al,
Shu et al,
Walnut Creek,
Vessey et al,
Beral et al,
1981
1982
1984
1984
1989
1989
1989
1988
1991
1977
1979
1982
1981
1981
1983
1987
1988
1989
1981
1987
1988
+ ve Association
-ve Association
0.0
0.5
1.0
1.5
2.0
2.5
Relative Risk or Odds Ratio
3.0
3.5
Hankinson SE et al. Obstet Gynecol. 1991;80:708-714.
www.contraceptiononline.
“….to our knowledge no other data on the
association between preschool diet
and breast cancer are available”
(Michels et al., 2006: 751)

“Measures of association”
◦ used to quantify the strength of the association
between an exposure and outcome
◦ e.g. Relative risk, odds ratio

Strong associations are more likely to be
causal than weak associations
◦ The larger the relative risk (RR) or odds ratio (OR),
the greater the likelihood that the relationship is
causal

Weak associations are more likely to be
explained by undetected biases or
confounders

How large must a relative risk or odds ratio
be to be considered ‘strong’:
◦ 2 ? 4 ? 20 ? …..?

No universal agreement regarding what
constitutes a ‘strong’ or ‘weak’ association
◦ An OR or RR > 2.0 is ‘moderately strong’
◦ An OR or RR > 5.0 is ‘strong’

The relationship between smoking and lung
cancer is an excellent example of a ‘strong
association’
◦ odds ratios and relative risks in different studies
are in the 4 to 20 range
“For one additional serving of French Fries
per week, the odds ratio for breast cancer
was 1.27” (Michels et al., 2006)
i.e. a “weak association”

This refers to the necessity for the exposure to
precede the outcome (effect) in time

Any claim of causation must involve the cause
preceding in time the presumed effect

Easier to establish in certain study designs
◦ Prospective cohort study
Easiest to establish in a cohort study
Lack of temporality rules out causality
Normal
Exposure
lung
TIME
Cancer
Outcome


Lung Ca.
This refers to the necessity for the exposure to
precede the outcome (effect)Smokers
in time
no
Lung Ca.
Any claim of causation must involve the cause
preceding in time the presumed effect
Lung Ca.
Population
 Easier to
Ex
40,634 British
establish
study designs
Doctorsin certain Smokers
◦ Prospective cohort study
Lack of temporality rules out causality
Non
Smokers
Time
Exposure
TIME
no
Lung Ca.
Lung Ca.
no
Lung Ca.
Outcome

This refers to the necessity for the exposure to
precede the outcome (effect) in time

Any claim of causation must involve the cause
preceding in time the presumed effect

Easier to establish in certain study designs
◦ Prospective cohort study

Lack of temporality rules out causality
Exposure
TIME
Outcome


Dose-response (‘biological gradient’)
◦ the relationship between the amount of exposure
(dose) to a substance and the resulting changes in
outcome (response)
If an increase in the level of exposure increases
the risk of the outcome
◦ this strengthens the argument for causality
R
I
S
K
0 cigs/day
R
I
S
K
< 5 cigs/day
R
I
S
K
5 - 20 cigs/day
R
I
S
K
> 20 cigs/day
Dose-response relationship
Dose-Response
Percentage of people with hearing loss
relative to workplace noise exposure
Average noise level
during an 8-hour
working day
(decibels)
<80
85
90
95
100
105
110
115
Exposure time (years)
5
10
40
0
1
4
7
12
18
26
36
0
3
10
17
29
42
55
71
0
10
21
29
41
54
62
64
 Plausibility
refers to the
biological plausibility of the
hypothesised causal
relationship between the
exposure and the outcome
◦ Is there a logical and plausible biological
mechanism to explain the relationship?
“A high dose of caffeine could constrict a
mother’s blood vessels reducing the blood flow
to the placenta” (Biological Plausibility)
< 200 mg caffeine/day
“There is no accepted biological
mechanism to explain the
epidemiological results; indeed the
relation may be due to chance
or confounding”
(Draper et al., 2005)
But other researchers have argued that there
is a biologically plausible explanation……..



EMF can induce currents that might alter the
voltages across cell membranes
Magnetic fields might cause the movement of
ferromagnetic particles within cells
EMF fields might also influence free radicals
Power lines might deflect and concentrate
cosmic rays on people living within their
vicinity


It is generally easy to ‘manufacture’
biologically plausible explanations for
the findings from epidemiological
research
Biological plausibility is not a
particularly useful viewpoint for
assessing a causal relationship
NB: Assuming study well-designed & conducted & bias etc. minimised
Type of Study
Ability to ‘prove’
causation
1) Randomised
Controlled Trial
2) Cohort Study
STRONG
3) Case-control study
Moderate
4) Cross-sectional study
WEAK
5) Ecological study
WEAK
Moderate
Does consumption of French fries by preschool
children cause breast cancer?
Strength
Consistency
Temporality
Dose response
Biological plausibility
Study design
Does consumption of French fries by preschool
children cause breast cancer?
Strength
Weak: OR = 1.27
Consistency
No
Temporality
Yes
Dose response
No
Biological plausibility
Yes
Study design
Case Control
Is this association causal?
Does consumption of French fries by preschool
children cause breast cancer?
Strength
Weak: OR = 1.27
Consistency
No
Temporality
Yes
Dose response
No
Biological plausibility
Yes
Study design
Case Control
Is this association causal?
Does cigarette smoking cause lung cancer?
Strength
Strong: OR, RR = 4 - 20
Consistency
Yes
Temporality
Yes
Dose response
Yes
Biological plausibility
Yes
Study design
Ecological, C/S, CC, Cohort
Is this association causal?
Does cigarette smoking cause lung cancer?
Strength
Strong: OR, RR = 4 - 20
Consistency
Yes
Temporality
Yes
Dose response
Yes
Biological plausibility
Yes
Study design
Ecological, C/S, CC, Cohort
Is this association causal?
If exposure X is
associated with
outcome Y…..then
how do we decide
if X is a cause of Y
Applying guidelines
for
causal inference
If exposure X is
associated with
outcome Y…..then
how do we decide
if X is a cause of Y




Strength of the association. How large is the
effect?
The consistency of the association. Has the
same association been observed by others, in
different populations, using a different method?
Specificity. Does altering only the cause alter
the effect?
Temporal relationship. Does the cause precede
the effect?





Biological gradient. Is there a dose response?
Biological plausibility. Does it make sense?
Coherence. Does the evidence fit with what is
known regarding the natural history and
biology of the outcome?
Experimental evidence. Are there any clinical
studies supporting the association?
Reasoning by analogy. Is the observed
association supported by similar associations?

: Strength of Association. “The lung
cancer rate for smokers was quite a bit
higher than for non-smokers (e.g., one
study estimated that smokers are about
35% more likely than non-smokers to get
lung cancer)”.
 2: Temporality. Smoking in the vast
majority of cases preceded the onset of
lung cancer
Different methods
(e.g., prospective and retrospective
studies) produced the same result.
The relationship also appeared for
different kinds of people (e.g., males
and females)
 Theoretical Plausibility. Biological
theory of smoking causing tissue
damage which over time results in
cancer in the cells was a highly
plausible explanation
 Consistency.
The conclusion
(that smoking causes lung
cancer) “made sense” given
the current knowledge about
the biology and history of the
disease
 Specificity in the causes. Lung
cancer is best predicted from
the incidence of smoking
 Coherence.
 Dose
Response Relationship. Data
showed a positive, linear
relationship between the amount
smoked and the incidence of lung
cancer.
 Experimental Evidence. Tar painted
on laboratory rabbits’ ears was
shown to produce cancer in the ear
tissue over time. Hence, it was clear
that carcinogens were present in
tobacco tar.

Analogy. Induced smoking with laboratory rats

References






showed a causal relationship. It, therefore, was
not a great jump for scientists to apply this to
humans
Doll, R. (1991). Sir Austin Bradford Hill and the progress of medical
science. British Medical Journal, 305, 1521-1526.
Hill, B.A. (1965). The environment and disease: Association or causation?
Proceedings of the Royal Society of Medicine, 58, 295-300.
Susser, M. (1977). Judgement and causal inference: Criteria in
epidemiologic studies. American Journal of Epidemiology, 105, 1-15
Bradford-Hill A. The environment and disease: Assocation or causation?
Proc R Soc Med 1965;58:295-300.
Grimes DA. Cause and effect - or coincidence? Contemporary OB/GYN Jan
1984;109-15.
Peterson HB, Kleinbaum DG. Interpreting the literature in Obstetrics and
Gynecology: I. Key concepts in epidemiology and biostatistics. Obstet
Gynecol 1991;78(4):710-17.

“None of these nine viewpoints can
bring indisputable evidence for or
against a cause and effect hypothesis
…. What they can do, with greater or
less strength, is to help answer the
fundamental question—is there any
other way of explaining the set of facts
before us, is there any other answer
equally, or more, likely than cause and
effect?” (Cited in Doll, 1991).