Causal Inference
Causal Inference:Introduction
Why be concerned with cause?
– so that we can intervene and prevent disease
– Presence of valid statistical association in no way
implies causality
Basic definition of “cause”:
– exposure that leads to new cases of disease
– remove exposure and some cases do not occur
(disease rate drops)
do not need to understand all casual factors to
prevent disease
A Cause?
The first and difficult question is what is a
cause?
A cause is something that has effect
In epidemiology a cause may be
considered to be something that alters the
frequency of disease, health status or
associated factors in a population
Philosophers have grappled with the nature
of causality for thousands of years
Cause and Effect
Causal knowledge permits rational plans
and actions to break the links between the
factors that cause disease and the disease
itself
Quote Hippocrates “To know the cause of
a disease and to understand the use of
various methods by which the disease may
be prevented amounts to the same as being
able to cure the disease”
Epidemiologic Triad
Agent Host
Environment
What causes Heart Attack?
– Cigarette smoking
– Cholesterol
– Elevated blood pressure
– Stress
– Family history
– Obesity
Which contributes the most risk?
What are the relationships among risk
factors?
Necessary and Sufficient
Necessary Cause
– A causal factor whose presence is required for
the occurrence of the outcome
Sufficient Cause
– A minimum number of conditions, factors or
events needed to produce a given outcome
Association vs. Causation
Does exposure A cause disease B?
First, find out if variables associated
Then use causal inference methods to
assist
Association vs. Causation
Association in simply an identifiable relationship
between an exposure an exposure and disease
e.g. Coronavirus is isolated more frequently from
individuals with diarrhea than those without
Implies that exposure might cause disease
Exposures associated with a difference in disease
risk are often called “risk factors
Association vs. Causation
Causation implies that there is a true mechanism
that leads from exposure to disease
– e.g., long-term heavy smoking causes myocardial
infarction
Finding an association does not make it causal
– e.g., hospital stays are associated with an increased
mortality rate, but this does not mean they cause
death
– E.g Because thunder follows lightning does not mean
that thunder is caused by lightning
Bradford Hill’s
Criteria for Causal Inference
Consistency of findings
Strength of association
Biological gradient (dose-response)
Temporal sequence
Biological plausibility
Coherence with established facts
Specificity of association
Experimental Evidence
Analogy
Strength of Association
Strong associations are less likely to be
caused by chance or by bias
– A strong association is one in which the
relative risk is
A very high (>>1), or
Very low (<<1)
Consistency of Effect
Relationships that are demonstrated in
multiple studies are more likely to be
causal,
i.e., consistent results are found
– In different populations,
– In different circumstances, and
– With different study designs.
Biological Gradient
“Dose- response”
Changes in exposure are related to a
trend in relative risk
Standardized Mortality Ratios (lung Cancer)
Dose (cigs/day) Study 1 Study 2 Study 3
<10 1.3 1.8 1.4
10-20 2.8 2.3 2.4
>40 4.7 3.7 6.3
Temporal Sequence
Exposure must precede disease
In disease with latency periods, exposures
must precede the latent period
In chronic disease, often long-term
exposure for disease induction
Does the cock crowing make the sun rise?
thanks to
Tom Grein
Plausibility & Coherence
The proposed causal mechanism should be
biologically (etiologically) plausible
Causal mechanism proposed must not contradict
what is known about the natural history and
biology of the disease, but
– The causal relationship may be indirect
– Data may not be available to directly support the
proposed mechanism
– Must be prepared to reinterpret existing understanding
of disease processes in the face of new findings
Specificity of the Association
An exposure leads to a single or
characteristic effect, or affects people with
a specific susceptibility
– Easier to support causation when associations
are specific
– But, obviously not always true
– Many exposures cause multiple diseases,
e.g., smoking causes many diseases including
heart disease, lung and other cancers, emphysema
Experimental evidence
human experiments
animal experiments
does removing exposure
lead to fall in outcome?
Analogy
Existence of other cause-effect relationships
analogous to the one studied supports a causal
interpretation.
If particular drug is a potent carcinogen, then different
drugs of same class may also be carcinogens
Weak criterion for causality. Useful for speculating
how risk factor may operate in different context
Causal Inference: Cautions
No single study is sufficient for causal
inference
Causal inference is not a simple process
– Consider “weight of evidence”, using Bradford Hill’s
criteria
– Always requires judgment and interpretation, no
cookbook method
– Some consider causal inference to be in the public
policy domain, rather than the scientific domain
No way to prove causal associations
Some notes from Bradford Hill
A caveat:
“None of my … [criteria] can bring indisputable
evidence for or against the cause-and-effect hypothesis
and none can be required as a sine qua non”
But also a need to act without “proof”:
“All scientific work is liable to be upset or modified by
advancing knowledge. That does not confer upon us a
freedom to ignore the knowledge we already have, or
to postpone the action that it appears to demand at a
given time.”
Summary
Cause and effect understanding is the
highest form of scientific knowledge
An association between disease and
postulated causal factors lies at the core of
epidemiology
Demonstrating causality is difficult of the
complexity and long natural history of
many human diseases and the ethical
restraints on human experimentation
Take Home Message
Epidemiological approaches to and
criteria for causality are not a checklist and
therefore conclusions must be carefully
judged and tentative.
There is need to synthesize data from
epidemiological studies and other
disciplines before reaching conclusions