Session 7: Causation

Association

An observable relationship or connection between certain exposure and a certain disease or health event

E.g., many social factors are clearly associated with health outcomes

Chance (random variation)

p-value

95% CI

Bias

Any trend in the collection, analysis, interpretation, publication or review of data that can lead to conclusions that are systematically different from the truth.

It arises from any trend in the collection, analysis, interpretation, publication or review of data that can lead to conclusions that are systematically different from the truth.

Confounding factor

Factor that is independently associated with the exposure and the disease or outcome but is not on the causal pathway between exposure and outcome

Disease risks would be different even if the exposure were absent in both populations

Reverse causality

When a cause-effect relationship exists in the opposite direction

Outcome causes exposure

Causation

A factor that plays about a role in bringing about a disease

A causal factor must be NECESSARY, SUFFICIENT - both of these or neither at all.

Necessary = a cause is termed necessary when it ALWAYS precedes a disease

Sufficient = a cause is termed sufficient when it can cause the disease ON ITS OWN

Bradford Hill's Criteria for Causation (1965)

Bradford Hill's Criteria for Causation (1965)

1) Strength of Association

A strong association is more likely to be causal because strong associations are less likely to be explained by confounding/bias

Defined by the size of risk, as measured by appropriate statistical tests (RR, OR etc.)

Bradford Hill's Criteria for Causation (1965)

2) Dose-Response Relationship

An association in which varying amounts of exposure to the factor leads to varying strengths of association with outcome of interest – is more likely to be causal

Bradford Hill's Criteria for Causation (1965)

3) Lack of Temporal Ambiguity

An association in which the putative factor has been demonstrated to precede the outcome of interest – is more likely to be causal

Exposure always precedes outcome

Bradford Hill's Criteria for Causation (1965)

4) Consistency of Findings

An association that has been demonstrated by different studies on different groups of people in different places at different times – is more likely to be causal

Bradford Hill's Criteria for Causation (1965)

5) Biological Plausability

An association for which there is a biologically plausible mechanism – is more likely to be causal

The association should be compatible with existing theory/knowledge

Bradford Hill's Criteria for Causation (1965)

Coherence of Evidence

An association that conforms with current knowledge and theory is more likely to be causal

Bradford Hill's Criteria for Causation (1965)

Specificity of Association

An association that is specific to the exposure-outcome association under investigation is more likely to be causal

Single putative cause produces a specific effect

When planning or implementing research, what phenomena should you be aware of which may give rise to apparent associations?

- Bias

- Chance

- Confounding

Three main types of bias - what are they?

1) Information bias

2) Selection bias

3) Confounding bias

Information bias

Measurement error of exposure or disease

Selection bias

Does the selection of the control/reference group depend on the outcome and the exposure of interest?

Confounding bias

Lack of comparability (lack of exchangeability) between exposed and non-exposed populations

Two types of error

Random and differential (systematic error)

Differential (systematic) error

Use of an invalid measure that misclassifies cases in one direction and misclassifies controls in another

Random error

Use of invalid outcome measure that equally misclassifies cases and controls

CHANCE and its relation to error

Chance is caused by RANDOM error

Chance leads to IMPRECISE results

Error from chance will be cancelled out by e.g., large sample size

BIAS and its relation to error

Bias is caused by SYSTEMATIC error

Bias leads to INACCURATE results

Error from bias will NOT cancel eachother out - whatever the sample size might be