Week 4 From Association to Causation Flashcards
How can causation be inferred?
1) Through demonstration of statistical association between a factor and a disease
2) By examining all available evidence to ascertain the meaning of the association - i.e. Is the factor causal?
What is the definition of causality?
Any event, condition, or characteristic that precedes disease, without which the disease would not have occurred, or would not have occurred till much later
What are some approaches to observing aetiology in humans?
Studies of group characteristics: Aggregate risk studies
Studies of individual characteristics: Case-control studies, cohort studies, experimental studies
Systematic review, meta analyses, randomised control trials
What is an Aggregate Risk Study?
What are the uses and shortcomings?
Looking at a population’s average exposure, and then the incidence of the disease
Shortcomings:
1) there is no way to infer whether or not the individuals with higher exposures are the ones with the disease or not
2) Not useful for causation
Uses:
1) can be useful for highlighting areas for future studies
What are the various types of associations?
Artefactual
Interrelated but Not Causal
Uncontrolled Confounding
Effect Modification
True Association
Define Artifactual Association
Associations that occur due to statistical artefact
I.e. due to poor study design: selection biases, biases in obtaining information, etc.
What are Interrelated but not Causal Associations?
Factors that may occur with disease, but that are not causal of disease.
I.e. Low socioeconomical status often occurs with lung cancer, but is not causal of lung cancer. To infer that it was, would be a false association
What are Uncontrolled Confounding Associations?
Confounding = confusion variables, produce mixing of effects
Occurs when the effect of the true exposure (the true causal agent) is mixed together with another variable which does not actually cause the disease - this leads to bias
E.g. smoking and drinking alcohol often occur together, but only smoking causes alcohol. If we inferred that alcohol caused lung cancer, that would be a confounding bias.
How can we differentiate between confounding and true causal associations?
If the relationship is truly causal, then incidence of disease should decline with removal of the causal agent
If the relationship is confounding, then the occurrence of disease will be unaffected by the removal of the confounder
**This can still be difficult if the confounder and causal agent are very closely related/intwined
What is an ‘Effect Modification’ association?
Effect modification occurs when the strength of the relationship between two variables is altered by the presence of a third variable.
E.g. smoking is causal of lung cancer. However, smokers who also work in coal mines have even higher rates of lung cancer.
Thus, coal mining works as an effect modifier in that is acts in synergy with smoking and magnifies the occurrence of the disease.
Define ‘necessary’ and ‘sufficient’ exposures
Necessary exposures = the exposure is necessary for the disease to occur. It always precedes disease
Sufficient exposure = the exposure is sufficient to cause disease - i.e. makes disease inevitable
This gives rise to 4 types of causal relationships:
Necessary but not sufficient
Unnecessary but sufficient
Necessary and sufficient
Neither necessary or sufficient
Necessary but not sufficient
The exposure is necessary to cause the disease (the disease will not occur without it) but it does not cause the disease by itself
Necessary and Sufficient
The exposure alone will inevitably cause the disease. The disease will always follow the exposure, and will not occur without the exposure
Sufficient but not necessary
The exposure will always cause the disease, but it is not the only cause of the disease - other things can cause the disease
Neither sufficient or necessary
The exposure is not needed for disease, and it cannot cause the disease by itself.
E.g. MI has a multifactorial aetiology. There are many potential causes: none of them alone are sufficient to cause MI, and not all of them have to be present for MI to occur
Define Causal Complements
The effect of an exposure on a population will depend on the prevalence of ‘causal complements’ - i.e. factors that ‘enable’ the exposure to cause the disease.
E.g. the effect of mycobacterium on a fully immunised and this immune population may be nil
However, the effect of mycobacterium on a non-immunised, susceptible population may be as high as 100%.
In this case, absence of immunisation is the causal complement
How do we assess causality?
A postulated cause-and -effect relationship should be examined in as many ways as possible.
Should attempt to isolate the proposed cause as much as possible - make every effort to eliminate confounding variables and biases from the study design
Causality is inferred based on all available evidence
What is the Bradford Hill criteria?
Used to determine is an association is likely causal or not…
1) Temporality
2) Strength
3) Dose-relationship
4) Reversibility
5) Consistency
6) Biological Plausibility
7) Specificity (not always so important given prevalence of multi-factoral aetiologies for most diseases)
8) Analogy
Temporality
The exposure must precede the disease
Strength
There must be a large calculated relative risk (for cohort studies), or large Odds ratio for case control*
Dose-relationship
The incidence of disease must be greater with higher dosages of exposure
Reversibility
The incidence of disease must reduce with reduction in exposure (e.g. reduction in risk of developing LC in ex-smokers)
Consistency
The relationship must be demonstrated in several studies, using different study designs, in different populations, different context, etc
Biological Plausibility
The causality must be logical in light of biological understandings of the time
Specificity
One cause leads to one effect - not so relevant in health as so many diseases are multifactorial in aetiology
Analogy
Examples exist of well established causes to the disease that are analogous to the one in question
E.g. that smoking causes LC is more plausible in light of longstanding evidence that LC is also caused by other inhaled toxins such as asbestos etc.
Things to consider when inferring causality
Consider all available evidence
Examine the quality of the study
Consider evidence both for and against the causality in question
