From a scientific standpoint, toxic tort cases, are about attributing a specific injury to a specific exposure. How can plaintiffs appropriately identify those individuals whose exposure was responsible for an injury? Or, alternatively, how can the defense identify those individuals whose exposure was not responsible for an injury. In epidemiology, this concept is assessed using a metric known as “attributable risk.”
In this post, I will review same basic epidemiological principles related to the concept of attributable risk, including when it is important in assessing alternative causation.
Risk Ratios Are Determined from Epidemiological Studies
Sometimes, plaintiffs attempt to make causal attributions by relying on uncontrolled, anecdotal data, which can be fraught with peril and often met with resistance by judges appropriately serving as gatekeepers. More often (and more appropriately), however, plaintiffs base causal attributions on controlled observational epidemiological data (e.g., analytical cohort and/or analytical case control studies). These epidemiological studies typically produce risk ratios that compare the risk of an event (i.e., an injury) in an exposed population to the risk of that same event in a similar, unexposed population. Assuming the risk ratio does not suffer from a biased study, a confounded study, and is sufficiently unlikely to have arisen by chance, the next step can then be taken – we can take what we learned about the population and attempt to apply it to an individual case.
Relevance of Risk Ratios to Individual Cases
The application of risk ratios to an individual case can be confusing to judges and juries because a risk ratio is calculated from populations not from individuals (technically, they are calculated from samples of populations). Therefore it seems as if a risk ratio should not apply to an individual – i.e., statistics coming from large samples should not be used to infer a causal attribution to an individual.
Yet, we use population-based metrics all the time to infer causal attributions in specific situations. For example, physicians routinely use data based on samples of people from clinical trials to make decisions about prescribing drugs to their individual patients. Or they utilize risk factor epidemiology to make lifestyle recommendations (e.g., the health benefits of diet or exercise). Similarly, we use statistical probabilities about weather patterns generally to make inferences about the weather on a given day, in a given geographical region. The fact is, statistical, population-based metrics provide valuable information about causation in individual instances and these population-based metrics can be used to make causal attributions in individual cases.
In the sections below, I describe some broad examples where risk ratios based on epidemiological studies preclude causal attribution and other instances where they have been appropriately used to make causal attributions in individual legal cases.
Lack of Causal Attribution
Sometimes well performed epidemiological studies can be used to conclude that we cannot attribute risk from a given exposure to a particular injury. A good example was the alleged link between silicone breast implants and systemic autoimmune disease. Epidemiological studies clearly demonstrated that the rate of systemic autoimmune disease was no different in individuals exposed to silicone implants as compared to the rate of systemic autoimmune disease in unexposed women (these data were nicely summarized in a 1999 report by the Institute of Medicine).
From these epidemiological studies, the medical and legal communities concluded that systemic autoimmune disease in any individual could not be reasonably attributed to her silicone breast implants. Without general causation (i.e., silicone breast implants are not generally capable of causing autoimmune disease in any person), we can’t have specific causation (i.e., silicone breast implants cannot have cause autoimmune disease in an individual).
Similar analyses have been performed on other exposure-disease relationships and have been used by the medical and legal communities to come to similar conclusions about causal attribution in individuals (e.g., cellular telephone use and brain cancer; welding rods and Parkinson’s disease; thimerosal in vaccines and autism).
Causal Attribution from Large Risk Ratios
In direct contrast to the examples described above, over the last five decades there have been instances where epidemiological studies have linked an exposure to an injury with a high degree of strength. For example, the risk of lung cancer in smokers has been estimated to be as high as 20- or 30-fold. Similarly, the risk of mesothelioma in individuals occupationally exposed to amosite asbestos has been estimated to be in a similar range (depending, of course, on the fiber burden and duration of exposure).
These extraordinarily high risk ratios are no surprise– everyone knows that populations of cigarette smokers have more lung cancers than non-smokers or that workers occupational exposed to amosite asbestos are far more likely to get mesothelioma. But the operational question for litigators is whether findings from these epidemiological studies can be used to make a causal attribution in an individual case. Once again, the answer is clear: of course. A lung cancer in an individual who smoked three packs of cigarettes a day for 20 years should be attributed to his tobacco exposure.
But why is it untenable to make the argument that a given lung cancer in a long-term, three-pack-a-day smoker occurred as part of the background rate of lung cancer? Because the risk ratios are so incredibly high (40-fold or more), the probability that the cause was cigarette smoking outweighs any other reasonable explanation. In other words, the background rate of lung cancer in non-smokers is dwarfed by the rate of lung cancer in chronic cigarette smokers. Thus, the attributable risk of lung cancer in a smoker is so high as to make any other exposure untenable.
This extreme example highlights the importance of an epidemiological term known as attributable risk. (Technically, attributable risk is a population-based concept, but for purposes of this post, I am loosely using it synonymously with “probability of causation” as the two concepts are clearly linked.)
In the section below, I briefly describe the calculation of attributable risk and then move on to a much more recent and relevant concept: causal attribution from small risk ratios and low attributable risks.
Calculating Attributable Risk
Attributable risk is the difference in the rate of a condition between an exposed population and an unexposed population. Unlike a risk ratio – which gives us the increased risk in an exposed population – attributable risk gives us an indication of the amount of risk that occurred because of the exposure. To do this, we must separate out the risk from the unexposed population and then express the difference as a percent.
There are many ways to calculate the attributable risk, but the simplest approach is to subtract 1 from the risk ratio for the exposure (RR) and to divide by the RR.When you perform this calculation on a RR of 20, you get a very high attributable risk, approaching 100%:
This tells us that in a broad population of smokers, the risk attributable to smoking is 95%. One would be hard pressed to find any other factor with a higher attributable risk than that. Based on this calculation, it is reasonable to conclude that chronic smoking is the crucial factor in an individual lung cancer case. And to a very high probably, smoking is “a cause” of any lung cancer case in a group of patients who engaged in this form of chronic self-abuse. Stated another way, as risk ratios get very high (20- or 30-fold), the attributable risk approaches 100%, and it becomes more and more reasonable to attribute the exposure of interest to the injury in every case.
Causal Attribution From Small Risk Ratios and Low Attributable Risks
But what happens when the attributable risk is much lower?
For example, what if we are confronted with a risk ratio of 1.2 as in the case of environmental tobacco smoke and lung cancer or with talc and ovarian cancers? In this example, the attributable risk is far lower:
Just as a very high attributable risk gives us confidence that a given injury should be attributed to a given exposure, a low attributable risk makes that attribution far less likely. As the attributable risk approaches zero, the background rate becomes more and more relevant and it becomes more and more appropriate to consider the role of other factors in attributing causation in an individual case. Whereas it is unreasonable to search for alternative causes when an attributable risk is extremely high (as in the case of tobacco and lung cancer), it is extremely important to search for alternative causes when an attributable risk is low.
This type of analysis is particularly important for non-occupational and environmental exposures to different asbestos fiber types.
Genomics Will be the Ultimate Tool for Determining Alternative Causation
This assessment leads naturally to the question of how we can most appropriately identify those cases that are not attributable to an exposure; or where another factor explains the injury to a higher degree of probability as compared to the exposure of interest.
In our view, the most promising approach to identifying alternative causation factors will be the application of genomic science, such as the identification of mutations associated with specific injuries as well as the identification of gene expression data and genomic signatures that can be used as biomarkers of exposure and/or biomarkers of injury. These factors provide reliable, highly specific, measurable biological markers that can be used to explain the presence of specific injuries (e.g., cancer, birth defects, neurodegenerative disease, etc.). If the attributable risk of these genomic factors is higher than the attributable risk of an exposure, they provide important and meaningful information that can outweigh the causal role of an alleged toxicant.
It is important to point out this post is designed to illustrate some fundamental concepts related to the attribution of risk and a number of logical leaps are made in the stated approach. The most important assumption is the view that associations demonstrated in epidemiological studies are causal in nature (i.e., no assessment was made regarding the role of chance, confounding or bias in interpreting epidemiological findings as well as the appropriate implementation of systematic criteria to assess causal inference from observational studies). In addition, the approaches described assume appropriately conducted and interpreted meta-analyses. All of these concepts are critical and must be determined with appropriate rigor in order to make the causal attributions described.
Most importantly, however, this piece did not deal with the critical legal concept as to whether a risk ratio greater than 2.0 is necessary to establish specific causation in the courtroom. Much has been written on this topic, but the following posts by Nathan Schachtman are particularly insightful and packed with valuable resources.