 |
| Block 1: Epidemiology |
| EPI11-1: ARE ASSOCIATIONS CAUSAL? APPROACHES TO EXAMINING THIS QUESTION |
|
| Block 1: Epidemiology |
| EPI11-1: ARE ASSOCIATIONS CAUSAL? APPROACHES TO EXAMINING THIS QUESTION |
 |
 |
SESSION OBJECTIVE |
|
At the end of this session you should understand how to approach assessing whether an exposure-response relationship is causal and not just a simple association. You will be able to apply the Bradford Hill criteria to the scientific evidence in the literature in order to make a complex judgement as to whether as association is causal or not. |
These are useful guidelines for assessing whether associations found between exposures and health outcomes (effects) are truly causal (the effect is due to the exposure).
The example of benzene exposure and leukemia as an outcome will be used.
A relative risk (IRR or IPR or OR or PR) < 2 is typically not considered to be a meaningfully elevated in observational studies due to the effect of unmeasured confounding. These confounders if adjusted for could bring the relative risk down to 1.
Of course unmeasured confounders might also dilute a real causal effect and make the relative risk seem to be lower than it is. Misclassification, particularly non-differential misclassification also tends to dilute the relative risk estimate.
The main point is that with low relative risk point estimates near 2 one is less confident that there is a real and causal association.
Relative risks for benzene and leukemia reported in the literature include the values 3.8, 5.6, 20, 21 & 25.
It can be seen that these are all quite large relative to 2. This adds to our belief that these associations found in the literature are causal.
If a significantly increased relative risk is found repeatedly and in different settings - different workplaces, countries, times and populations - this would indicate that it is a real effect.
For benzene and leukemia studies have reported this effect from 1928 to recent times, from many countries, different industries and by different investigators. This similarly adds to our belief that this relationship is causal.
This means that the outcome in question, leukemia, is specific to benzene toxicity. This is not too useful a criterion in most cases, because of multiple causation of most health conditions and also because of the contribution of occupational and environmental exposures to health outcomes that are not intrinsically occupational in causation, eg asthma, cancer or musculoskeletal disorders. Even in the case of the rare and specific condition of mesothelioma which has been shown to be caused by asbestos there is another documented (and even rarer) cause of a similar nature to asbestos viz. fibrous zeolites.
However in this case there are many causes of leukemia, and this criterion is not helpful in coming to a decision.
This is a very important criterion and the "cause" must precede the outcome. And the latent period between first exposure and the manifestation of the outcome must be biologically plausible.
Often in cross-sectional or prevalence studies this is difficult to know as both exposure and outcome are measured at the same time and it is not always clear what came first. This may also be the case in case-control studies.
In the case of benzene and leukemia the studies have been able to document that the exposure preceded the disease. This criterion works rather well in reverse. If the exposure followed the disease then there could not have been causation. Similarly it is unlikely that a short period of exposure to benzene in a couple of years before developing leukemia cancer is the cause.
This refers to the existence of an exposure-response relationship usually whereby an increase in exposure will result in an increase in the frequency of the adverse outcome.
For tobacco smoking and lung cancer incidence in the Doll and Hill study of medical doctors this is more or less a straight line and the association between exposure to cigarettes and lung cancer incidence has a correlation coefficient near 1, providing strong evidence for causation.
For some exposures eg alcohol consumption and mortality there is a U-shaped exposure-response relationship with non-drinkers and those who drink more than 2 drinks per day having increased risk of mortality. This is also evidence for a (more complex) causal mechanism.
Often there is poor exposure characterisation. This is particularly the case for historical levels of exposure. Industry is not very good in performing these measurements and keeping records that can be used to attribute exposure to any individual at work across the duration of their years of service while exposed. Often measurements are crude and therefore exposure is considerably misclassified.Evidence of an exposure-response relationship beyond a simple difference in the level of the outcome between the group of all those exposed and all those not exposed is strong evidence for causality.
For benzene there is not particularly good evidence of an exposure-response relationship.
This involves an interpretation as to whether there is some biologic mechanism that could explain how the exposure causes the outcome. On the one hand this criterion is limited by current scientific knowledge and on the other it has been pointed out that there is no limit to the imagination of scientists. The mechanisms explaining cholera in the mid 1800s and that explaining lung cancer and tobacco in the mid 1900s were unknown. Yet the presumption of cause was very strong indeed and could be acted successfully upon to prevent both diseases. Hence one cannot attach too much weight to this criterion.
In the case of benzene, we know that similar chemicals can initiate or promote cancer of various types, so this is biologically plausible.
As is so often the case in epidemiology, it is not often ethical nor is it possible to experiment with human subjects, by say conducting a randomised controlled trial. Natural experiments may be helpful. Animal experiments are often performed but these do not allow confident extrapolation to humans.
There is evidence of carcinogenicity in laboratory animals where benzene has been observed to cause similar cancers, but chemicals and other exposures cause different outcomes in different species. There was a natural experiment in that the benzene standard was lowered in some countries substantially (from 100 to 35) around 1947/8 and studies showed that there was a significant occurrence of leukemia amongst employees who had their exposures before this time, while there were no new cases in men hired after 1948.
If similar associations are observed for related exposures and outcomes this will strengthen the case for causation. It could mean that sturcturally similar exposures (chemicals) cause the outcome of interest, or that similar outcomes have been caused by the exposure of interest.
For instance, other aromatic hydrocarbons such as xylene and toluene have been shown to produce cancers in laboratory animals. There is however laboratory evidence that Benzene is mutagenic and carcinogenic in different biological systems.
This relates to the overall coherence of all the evidence from the application of these causal criteria. Having thought about it all, does all the evidence cohere? Or is there just too much in the way of inconsistency, findings that don't fit and discrepancy? What is the summary position in your mind about the magnitude and direction of the causal association?
In the case of benzene, most scientists and authorities would judge it to be human carcinogen, which in fact the IARC has indeed done, and have regulated accordingly.
