Association and cause
Aims of this resource
- To enable an understanding of the important concepts in determining causes of ill-health with emphasis on epidemiology and the environmental and occupational aspects of public health
- To enable a distinction to be made between associations that are likely to be causal and those which probably have other explanations
Background
For the purposes of maintaining health by preventing ill-health, it is clearly essential to have knowledge of the environmental causes and other determinants of ill-health. Relevant information is also important for the purposes of reaching a correct and complete diagnosis and to assist in decisions on treatment and prognosis. When exposure to an agent appears associated to a health effect, what criteria can one use to determine whether the former is really the cause of the latter? The main criteria are summarised below. Stop after reading each one and consider in turn relevant examples and then other non-causal explanations (especially bias, random variation and confounding) for the associations.
Learning objectives
Bear the following objectives in mind:
You should:
- Accept the importance of determining causation for the purposes of prevention of ill-health and protection of the public health
- Understand and apply criteria for determining cause of ill-health
- Be able to discuss the principles of the interactions contributing to associations between various factors and ill-health, and illustrate these principles with reference to environmental and occupational factors
Criteria for determining causation
These are based on the work of Austin Bradford-Hill.
Temporality
Does the presumed cause precede the effect? Obviously, a cause must precede its effect. However, that is as far as can be said with any degree of certainty. It does not follow that if exposure to a postulated causative agent precedes an effect, then the latter is the direct consequence of the former.
Reversibility
Does removal of a presumed cause lead to a reduction in the risk of ill-health? Reduction in a particular exposure if followed by a reduced risk of a particular disease may strengthen the presumption of a real cause-effect relationship. This reversibility of association may suffer from similar fallacies as temporality.
Strength of association
Is the exposure associated with a high relative risk of acquiring the disease? The concept of "risk" and its measurement also features elsewhere. How does the strength of association between a risk and a possible causal factor influence the weight of evidence for a causal association?
Exposure-response
Is increased exposure to the possible cause associated with an increased response (i.e. an increased likelihood of an effect)?
There are a number of illustrations of "exposure-response" relationships. However, first ensure that you understand the distinction and the link between exposure and dose:
- Exposure is usually quantified as a product of duration and intensity - (such as airborne concentration of a relevant inhalable agent)
- Dose is a measure (of energy, mass or number of infective particles as the case may be) that is actively taken up by the human body
The demonstration of an "exposure-response" relationship (provided it is not the result of confounding) has two important implications:
- It is good evidence of a true causal relationship between exposure to a particular agent and a health effect
- It may permit a measure of "risk" of a particular health effect to be related to a given exposure and it may even suggest levels of exposure to that causal agent below which its specific health effect is a most unlikely or even impossible consequence
Stop and consider possible examples of exposure response relationships. How many can you think of?
Examples of exposure response relationship include:
- The demonstration that the higher the concentration of particulate air pollution in cities the greater the death rate especially from lung disease
- The longer the duration of occupational exposure to noise and the greater its intensity (loudness), the greater the likelihood of subsequent deafness among the exposed workers (noise induced hearing loss)
- The longer the duration of occupational exposure to asbestos and the higher the concentration of asbestos fibres in air, the greater the likelihood that the exposed workers will develop a fibrotic lung disease (asbestosis)
Consistency
Have similar results been shown in other studies? Elsewhere you can learn how to critically appraise literature. It follows that if a number of good studies using different approaches lead to the same interpretation of a cause-effect relationship, it is more likely to be a valid one
Biologic plausibility: Is there a reasonable postulated biologic mechanism linking the possible cause and the effect?
Analogy
Can parallels be drawn with examples of other well established cause-effect relationships?
Specificity
Does the cause lead to a specific effect? (i.e. one cause - one effect) Many diseases and symptoms can be the result of a number of causes. Similarly many causes of ill-health can have different effects on the body. Only rarely is specificity demonstrable in environmental cause-effect relationships (other than in infectious diseases). Thus, for example, mesothelioma of the pleura (or peritoneum) is a relatively specific consequence of asbestos exposure. (However, this criterion has to be treated with some caution: for example, we know that tobacco smoking can cause many diseases ranging from lung cancer to chronic bronchitis to bladder cancer and that asthma can be caused by many occupational causes, i.e. a single cause does not necessarily equal a single effect).
A similar logical thought process is applied when taking and interpreting a medical history.
For more details about some causes of ill-health, see:
Chance, bias and confounding
There are various factors which may explain why an apparent association is not in fact causal. The following brief account contains supplementary information on these three important factors, namely chance, bias and confounding, which need to be borne in mind when drawing conclusions about cause and association.
These are considered below:
Chance
Imagine that you want to determine the frequency of back pain among employees in a particular workplace. Rather than questioning all the employees, it would be easier to administer questionnaires to only a sample of this population and from them, estimate the frequency of back pain in the workers. However, you would have to bear in mind that CHANCE may have affected your results because of random variation in the population - it could be that, by chance, the sample you chose were a particularly fit and healthy group and you would therefore underestimate the frequency of back pain in your workplace.
The larger the size of your sample, the smaller the effect that chance will have on your results. To quantify the degree to which chance may account for the results observed, a test of statistical significance would need to be performed.
Chance can also operate in a different direction, especially if multiple testing is undertaken without specific prior hypotheses. Assume that you wished to determine whether air pollutants in the home caused asthma. You could identify a number of children with asthma (obviously preferably children who have lived in that particular home since before they had symptoms of the disease) and compare their homes with a control group without asthma. You could undertake a range of measurements of air pollutants in the home, for example butane, pentane and other aliphatic hydrocarbons, benzene, toluene, xylene and other aromatic hydrocarbons, formaldehyde and other aldehydes, acetone and other ketones and so and so forth - let us assume that you measure the concentrations of forty chemical pollutants. Then you could compare the concentrations of the pollutants in the homes of the asthmatics with those in the homes of the non-asthmatics. There is a likelihood of 1 in 20 that by chance alone there will appear to be 'statistically significant' differences at the conventional level (P=<0.05) between the two sets of homes in the concentrations of two of the pollutants.
Bias
A further important factor to consider is whether some aspect of the design, or conduct of the study has introduced a systematic error or BIAS into the results. Bias is most easily understood if you think in terms of the danger of not comparing 'like with like'.
The main types of bias are:
-
Selection and participation bias
This occurs if the study populations being compared are not strictly comparable. For example, in a study to determine the effect of a Workplace Health Promotion (WHP) programme on 'sickness absence', the rate of subsequent sickness absence might have been compared between those who participated in the WHP programme and those who did not. What if the results appeared to show that the WHP group had lower rates of sickness absence?
Would one be entitled to conclude that WHP resulted in less sickness absence?
Bias may well have been present in this study because those who took part in the WHP may, for other reasons such as their smoking habits, diet or psychological factors, have been at a lower likelihood of sickness absence even before joining the WHP. In other words, bias would have arisen through 'not comparing like with like'.
-
Observation bias
This occurs if non-comparable information is obtained from each study group.
For example, if one was conducting a case-control study to determine whether scleroderma (systemic sclerosis) was associated with occupational exposure to certain hazards (such as trichloroethylene or silica), bias could arise if the interviewer knew (or could tell) which people were the cases and which were the controls. He/she might seek more detail about exposures from the cases than the controls (referents) who did not have the disease. Thus, observer bias would have influenced the results.
Recall bias might arise if the cases (suffering from the disease) having previously pondered about possible causes of their misfortune, were to recollect more detail about their past exposures, than the controls (who may have no real motivation to reflect at length on their past occupations).
Note: 'Bias' in an epidemiologic context has a specific meaning, which does not necessarily imply bad faith. In other words, a criticism of a study as having been biased does not necessarily nor usually mean that the investigators set out with the intent of swaying the results or interpretation one way or the other.
Confounding
A third possibility which has to be entertained is CONFOUNDING - this results from multiple associations between the exposure, the disease and some third factor (the 'confounding variable') which is associated with both the exposure and independently affects the risk of developing the disease.
An example of confounding is the observed association between air pollution and cardiac or pulmonary disease. There now appears to be little doubt that a causal association exists between say particulate air pollution and respiratory morbidity and mortality. However, an unsophisticated study simply relating air pollution to ill healths and deaths might lead to the conclusion that the association is much stronger than it really is. Why?
Because of confounding variables such as temperature.
Low temperatures in winter may contribute to increased mortality. In addition, low temperatures (in meteorologic conditions of inversion) may favour increased pollution levels. If the confounding caused by temperature is taken into account (i.e. it is resolved), then the association between air pollution and health becomes weaker.
