Difference between revisions of "Viewpoints for Causality (Bradford Hill)"
Bosmana fem (talk | contribs) m |
Bosmana fem (talk | contribs) m (→Experimental evidence) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 53: | Line 53: | ||
==Experimental evidence== | ==Experimental evidence== | ||
| − | Ideally experimental evidence should be obtained if at all possible. Robert Koch proposed four postulates that establish an micro-organism as a cause of a disease. | + | Ideally, experimental evidence should be obtained if at all possible. Robert Koch proposed four postulates that establish an micro-organism as a cause of a disease. |
| − | Summary of Koch's | + | Summary of Koch's Postulates |
| − | + | (i) the micro-organism must be consistently present in the diseased and not in the healthy individual | |
(ii) the micro-organism must be isolated and grown | (ii) the micro-organism must be isolated and grown | ||
| Line 65: | Line 65: | ||
(iv) the micro-organism must be re-isolated and shown to be the same as in (i) | (iv) the micro-organism must be re-isolated and shown to be the same as in (i) | ||
| − | These four postulates are regarded as sufficient but not necessary to establish causation. Certain types of study designs may provide more convincing evidence than others. In current systems for | + | These four postulates are regarded as sufficient but not necessary to establish causation. Certain types of study designs may provide more convincing evidence than others. In current systems for classifying evidence (e.g. the Scottish Intercollegiate Guidelines Network), randomised controlled trials are considered to provide strong evidence of cause and effect <Ref>SIGN 50: A guideline developers' handbook. Section 6: Forming guideline recommendations. http://www.sign.ac.uk/guidelines/fulltext/50/section6.html (4 August 2006).</ref> . The highest level is provided by systematic reviews of randomised controlled trials. |
| − | Since it is sometimes unethical and/or impractical to conduct controlled trials, a possible alternative is to remove the exposure and see if the disease decreases, unless the causal process is | + | Since it is sometimes unethical and/or impractical to conduct controlled trials, a possible alternative is to remove the exposure and see if the disease decreases, unless the causal process is considered irreversible. This is observed in “natural experiments” when intervention leads to change in one direction, and the removal of that intervention reverses the direction in the outcome. The introduction of pertussis vaccination in the UK, for instance, led to a fall in the incidence of whooping cough. Unsubstantiated concerns about adverse effects later led to a fall in its uptake. This was followed by a rise in the incidence of the disease. Uptake then rose again and incidence fell correspondingly <Ref>Gangarosa EJ et al. Impact of anti-vaccine movements on pertussis control: the untold story. Lancet 1998;351:356-61. </ref>. |
==Analogy== | ==Analogy== | ||
| Line 76: | Line 76: | ||
| − | = | + | <div style="display: inline-block; width: 25%; vertical-align: top; border: 1px solid #000; background-color: #d7effc; padding: 10px; margin: 5px;"> |
| + | '''FEM PAGE CONTRIBUTORS 2007''' | ||
;Editor | ;Editor | ||
:James Stuart | :James Stuart | ||
| Line 87: | Line 88: | ||
:Aileen Kitching | :Aileen Kitching | ||
:James Stuart | :James Stuart | ||
| + | </div> | ||
[[Category:Causal Inference]] | [[Category:Causal Inference]] | ||
Latest revision as of 18:42, 10 April 2023
Nine "viewpoints" for causality were set out by Bradford Hill [1].
Contents
Strength of Association
Strong associations are more likely to be causal than weak ones. In common source outbreaks, we look for one (or two) vehicles of infection, so we expect the true source's risk ratio to be very high. They often are. In food-borne disease outbreaks, the association between illness and exposure to a common source may lead to odds ratios exceeding 50 [2], [3]. On the other hand, not all strong associations are causal and may be due to confounding. This can be illustrated by the association observed between multiple births and Down syndrome, which is actually confounded by the mother's age [4]. The use of the strength of the association as a causal criterion can be misleading where confounding factors are not known.
Weak associations do not rule out causality and may still have public health importance. This can apply when exposure is common in a population. e.g. passive smoking and lung cancer (Risk Ratio = 1.3) [5].
Consistency
Repeated observations of an association in different populations under different circumstances showing the same or similar results suggest that the results of a single study are not due to chance. Over 100 studies into the association between smoking and lung cancer demonstrate an increased disease risk. Note that the weighted results of similar studies, if statistically homogeneous, can be combined together in a meta-analysis.
However, consistency can also be misleading. Five studies compared the risk of dying from meningococcal disease after administration of oral antibiotics before admission to hospital versus no administration. All 5 studies showed lower mortality in those given antibiotics [6]. The results were consistent and statistically homogeneous, thus in favour of a protective role for oral antibiotics before hospitalisation. However, doctors are more likely to prescribe oral antibiotics if they do not suspect meningococcal disease i.e. in patients with milder illness. We can infer that the association observed is confounded by disease severity. Consistency of results in observational studies may simply be due to the presence of the same confounding factors.
Epidemiologists should be cautious not to mistake statistical significance for consistency. Different studies may show similar effect measures but with different levels of significance (including significant and not significant) and still be consistent.
Specificity
The criterion of specificity requires that a cause leads to a single effect, not multiple effects.
It is not very helpful in establishing causality. The fact that one agent contributes to multiple diseases should not be evidence against its role in any one disease. Smoking, for example, can lead to many ill effects in the smoker. The specificity criterion has repeatedly been used by those against ‘smoking as a cause of lung cancer’ as their main argument.
Temporality
Exposure must precede the disease.
This is the only criterion that is fundamental to postulating a cause and effect relationship and fits our intuitive understanding of causality.
However the example of Mumps Measles Rubella (MMR) vaccine and autism illustrates that our self-taught causal inference can be erroneous. Some parents observe the beginning of autism shortly after an injection of MMR. Quite naturally, they attribute this illness to the vaccine. But autism often begins at the age when MMR is given. Studies have shown that autism is just as likely to occur before as after an MMR injection, and that children who have not been given MMR are just as likely to get autism as those who have [7].
It is sometimes difficult to document sequence, especially if there is a long lag between the exposure and the disease, subclinical disease, exposure (e.g., a treatment) brought on by an early manifestation of the disease.
Biological gradient (dose response)
A biological gradient exists when the risk of disease/outcome increases with increasing exposure to the suspected risk factor.
A linear relationship between dose and response supports causality. For example, the higher the number of cigarettes smoked, the greater the risk of lung cancer.
On the other hand, a lack of dose response does not exclude a causal link. Causal associations showing a single jump (threshold, saturation effects) rather than a monotonic trend have been described.
Sometimes associated non-causal factors may also increase in a similar way to the causal factor. As discussed earlier, the risk of Downs syndrome seems to increase with birth rank, while birth rank increases with age of the mother. Although a gradient is observed, the cause of the increased risk of Down's syndrome is linked to age of the mother not birth rank [4].
Plausibility
This refers to the biological plausibility of the hypothesis i.e. its consistency with current biological knowledge about the disease (for example, oral contraceptivs and *** cancer). Being largely based on prior beliefs, it remains a subjective judgement.
During an outbreak of psittacosis in Australia, reported in the Lancet ,16 cases had spent 17.5 hrs in the garden compared with controls who had only spent 5.2 hrs [8]. Cases were more likely to mow the lawn than their controls (OR 8.8, 95%CI 1.2 – 389). It was quite plausible that this was a causal relationship (and still is). However, the authors in their study, had not taken account of the gender of controls. Controls were evenly distributed between males and females yet nearly all the cases were male. Stratification by gender reduced the strength of association, as measured by Mantel Haenszel Weighted Odds Ratio, to 5.5 and its lower limit of the confidence interval to 0.6 (p=0.19). Although the OR was still raised increased after stratification, it was misleading to present the results without taking gender into account.
In this way, plausibility may sometimes mislead in drawing conclusions.
Coherence
We speak of coherence when the interpretation of cause-effect relationship does not conflict with what is known of the natural history and biology of disease. This is similar to plausibility. All observations are expected to fit with a hypothesized model to form a coherent picture.
Absence of coherence cannot be taken as evidence against causality, and vice versa. Many studies have shown that prevalence of meningococcal carriage in teenagers rises with age. This has been explained biologically as the result of changes in mucosal characteristics that occur with age. However, when prevalence was adjusted for social factors such as going to pubs and clubs, kissing and smoking, the increasing trend disappeared [9].
Experimental evidence
Ideally, experimental evidence should be obtained if at all possible. Robert Koch proposed four postulates that establish an micro-organism as a cause of a disease.
Summary of Koch's Postulates
(i) the micro-organism must be consistently present in the diseased and not in the healthy individual
(ii) the micro-organism must be isolated and grown
(iii) pure culture of the micro-organism should induce disease
(iv) the micro-organism must be re-isolated and shown to be the same as in (i)
These four postulates are regarded as sufficient but not necessary to establish causation. Certain types of study designs may provide more convincing evidence than others. In current systems for classifying evidence (e.g. the Scottish Intercollegiate Guidelines Network), randomised controlled trials are considered to provide strong evidence of cause and effect [10] . The highest level is provided by systematic reviews of randomised controlled trials.
Since it is sometimes unethical and/or impractical to conduct controlled trials, a possible alternative is to remove the exposure and see if the disease decreases, unless the causal process is considered irreversible. This is observed in “natural experiments” when intervention leads to change in one direction, and the removal of that intervention reverses the direction in the outcome. The introduction of pertussis vaccination in the UK, for instance, led to a fall in the incidence of whooping cough. Unsubstantiated concerns about adverse effects later led to a fall in its uptake. This was followed by a rise in the incidence of the disease. Uptake then rose again and incidence fell correspondingly [11].
Analogy
The existence of other cause-effect relationships analogous to the one under study supports a causal interpretation. This is a weak criterion for causality, but can be useful for speculating how a risk factor may operate in a different context.
References
- ↑ Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965; 58:295-300.
- ↑ Killalea D, Ward LR, Roberts D, de Louvois J, Sufi F et al. International epidemiological and microbiological study of outbreak of Salmonella agona infection from a ready to eat savoury snack - I: England and Wales and the United States. BMJ 1996; 313:1105-7.
- ↑ Lewis DA, Paramathasan R, White DG, Neil LS, Tanner AC et al. Marshmallows cause an outbreak of infection with Salmonella enteritidis phage type 4. Commun Dis Rep Review 1996; 6: 183-8.
- ↑ 4.0 4.1 Chan A, McCaul KA, Keane RJ, et al. Effect of parity, gravidity,previous miscarriage, and age on risk of Down’s syndrome:population based study. BMJ 1998;317:923–4.
- ↑ Taylor R, Najafi R, Dobson A. Meta-analysis of studies of passive smoking and lung cancer: effects of study type and continent. Int J Epidemiol 2007; 36: 1048-59
- ↑ Purcell B, Samuelsson S, Hahne S, Ehrhard I, Heuberger S, Camaroni I, Charlett A, Stuart JM. Effectiveness of antibiotics in preventing meningococcal disease after a case: systematic review. BMJ 2004;328: 1339-42.
- ↑ Taylor B et al. Autism and MMR vaccine: no epidemilogical evidence for a causal association. Lancet 1999; 353: 2026-9.
- ↑ Williams J, Tallis G, Dalton C, Ng S, Beaton S et al. Community outbreak of psittacosis in a rural Australian town. Lancet 1998; 351: 1697-9.
- ↑ MacLennan J et al. Social behaviour and meningococcal carriage in British teenagers. Emerg Infect Dis 2006; 12: 950-7.
- ↑ SIGN 50: A guideline developers' handbook. Section 6: Forming guideline recommendations. http://www.sign.ac.uk/guidelines/fulltext/50/section6.html (4 August 2006).
- ↑ Gangarosa EJ et al. Impact of anti-vaccine movements on pertussis control: the untold story. Lancet 1998;351:356-61.
FEM PAGE CONTRIBUTORS 2007
- Editor
- James Stuart
- Original Authors
- Alain Moren
- James Stuart
- Contributors
- Lisa Lazareck
- Arnold Bosman
- Aileen Kitching
- James Stuart
Root > Assessing the burden of disease and risk assessment > Field Epidemiology > Methods in Field Epidemiology > Scientific (evidence base for prevention) > Causal Inference
