Causality Flashcards
What is a cause in epidemiological terms?
A cause is an exposure or factor that increases the probability of disease
What is a counfounding factor?
Variable associated with the exposure being investigated
Variable that independently associated with the risk of developing the outcome of interest
Possible explanations for observed associations
An observed association between X and Y may be:
a true causal association really does exist where X > Y
due to an unknown confounding factorN.B. – usually have partial rather than full confounding
due to a common causee.g. lung cancer is associated with chronic bronchitis but both are caused by tobacco smoking
reverse causality, i.e. you believe X>Y but in fact Y<x></x>
<p>
<br></br> </p>
</x>
How do you evaluate the strength of evidence in favour of a cause-effect relationship?
Bradford-Hill criteria:
Strength of association
Specificity of association
Consistency of association
Temporal sequence
Dose response
Reversibility
Coherence of theory
Biological plausibility
Analogy
Why does strength of assocaition help determine causality?
A strong association (e.g. a high incidence rate ratio or odds ratio) is more likely to be causal. This is because strong associations are less likely to be explained by undetected confounding or bias.
How does specificity of association help to determine causality?
One exposure leads to one outcome
Specificity of association strengthens the case for a causal link
What is consistency of association?
When a causal link is observed in different studies and in different sub-groups/populations.
Consistency of association between studies or groups is unlikely to be due to the same confounding or bias. Causal link is more likely
Why is temporal sequence important in determining causality?
A causal link is more likely if exposure to the putative factor has been shown to precede the outcome
A causal link cannot exist if the outcome preceded exposure to the putative factor
What is dose response? How does it show causality?
Dose response is a biological gradient where the level of exposure is related to the risk of the outcome.
Unknown confounding factors or bias are unlikely to operate to the same degree in various levels of exposure with the outcome.
Note: biological gradients can be ‘J’ shaped (exponential) or U shaped (extremes increase risk)
The RCT of the effect of cholesterol-lowering therapy on cardiovascular outcomes is an example of reversibility. Explain how this shows causality
A causal link is very likely if removal or prevention of the putative factor leads to a reduced or non-existent risk of acquiring the outcome
Difficult to demonstrate in progressive conditions (e.g. cancers) and because public health programmes to remove or prevent an exposure requires action by society.
Why is coherence of theory a weaker indication of causality.
Coherence theory states that coherence with current paradigms / constructs / theories strengthens the case for a causal link however,
this leads to inappropriate rejection of ‘unfavoured’ associations, i.e. ‘publication bias’ towards studies that support favoured theories or demonstrate that drug / interventions work
lack of coherence does not rule out a causal link, e.g. Helicobacter pylori and peptic ulcers
What is biological plausibility?
demonstration of a biologically plausible mechanism strengthens the case for a causal link
How are analogies used to help determine causality?
A causal link is more likely if an analogy exists with other diseases, species or settings. An analogy is easier to infer than a biologically plausible mechanism
Example of an analogy:The epidemiology of Hepatitis B virus was successfully used to predict how HIV virus would spread
however, analogies can be inappropriate, e.g. sheep scrapie’s non-transmission to humans led scientists to believe that cow BSE would not cross the species barrier to humans
How can you be sure that an observed association is causal?
Apparent associations can result from: chance, bias or confounding. These explanations need to be excluded when validating associations
- if the 95% CI of a result excludes the null value, the association is unlikelt to be due to chance. p-value of <0.05 also decreases the likelihood that the result is due to chance.
- selection bias or information bias can affect teh results, leading to conclusions that are inaccurate
- confounding factors can influence the outcome under study and can result in asssociations which are non-causal