The relative risk argument is simple. A relative risk of 1.0 means that the rate of disease incidence or mortality is the same among the exposed and control populations. A relative risk of 2.0 means that the incidence rate in the exposed population is twice that in the controls. The existence of an observed rate among the non-exposed controls suggests that we are dealing with a disease of “ordinary life,” for which there is an expected rate of occurrence. Most chronic diseases, such as cancer, autoimmune disease, cardiovascular diseases, fall into this category of diseases of ordinary life.
If a study of a disease that is prevalent in the general population, say colon cancer, is conducted in an exposed cohort of workers, say asbestos insulators, and the study finds a relative risk of 1.5, we would have to take several steps to assess the finding’s relevance in litigation. First, this positive association would have to be evaluated for causality. Bias and confounding would have to be ruled out as explaining the apparent increase in risk. Furthermore, the association would have to be evaluated for various indicia of causality, such as consistency with other studies, dose-response relationship between exposure and outcome, biological plausibility and coherence, and support from experimental studies. In the case of asbestos and colon cancer, the causal hypothesis has repeated failed to be supported by such evaluations, but even if we were to assume general causation, arguendo, we would be left without a way to infer causation in a given case. If plaintiff supported his case with evidence or a relative risk of 1.5, we would have 50% more observed cases than expected. So if the observed population was expected to experience 100 colon cancer cases over the observation period, a relative risk of 1.5 means that 150 such cases were observed, or 100 expected cases and 50 putative excess cases. Alas, there is no principled way to tell an excess case from an expected case, and the odds favor the defense two to one that any given case arose from the expected population as opposed to the excess group. As a probability, the probability that plaintiff’s case arose from the excess portion is 33%, well below what is needed to support a sustainable claim. Again, this assumes many facts in plaintiff’s favor, such as a perfect epidemiologic study, without bias or confounding, and with consistency among the findings of similar studies. (None of these assumptions is even close to satisfied for asbestos and colon cancer.)
In the Agent Orange litigation, Judge Weinstein implicitly recognized the problem that very large relative risks suggested that an individual case was likely to have been related to its antecedent risks. Small relative risks suggested that any inference of specific causation from the antecedent risk was largely speculative, in the absence of some reliable marker of exposure-related causation. See In re Agent Orange Product Liab. Litig., 597 F. Supp. 740, 785, 817 (E.D.N.Y. 1984)(plaintiffs must prove at least a two-fold increase in rate of disease allegedly caused by the exposure), aff’d, 818 F.2d 145, 150-51 (2d Cir. 1987)(approving district court’s analysis), cert. denied sub nom. Pinkney v. Dow Chemical Co., 484 U.S. 1004 (1988); see also In re “Agent Orange” Prod. Liab. Litig., 611 F. Supp. 1223, 1240, 1262 (E.D.N.Y. 1985)(excluding plaintiffs’ expert witnesses), aff’d, 818 F.2d 187 (2d Cir. 1987), cert. denied, 487 U.S. 1234 (1988).
Ever since Judge Weinstein embraced the relative risk of two, as an important benchmark to be exceeded if plaintiffs hoped to show specific causation, scientists who practice medicine for the redistribution of wealth have attacked the concept. The challengers have urged that small relative risks, including relative risks of two or less, could suffice to support causal attribution in a given case, especially in the presence of relevant clinical findings. The challengers, however been vague and evasive when it comes to identifying what are the relevant clinical findings and how they operate to show that the risk has actually operated to become part of the causal pathway that has led to the individual’s injury or disease.
Among the most vociferous of the challengers has been Professor Sander Greenland, of the University of California Los Angeles School of Public Health. Greenland has published his criticisms of the inference of a probability of individual causation from the relative risk on many occasions. See, e.g., Sander Greenland & James Robins, “Conceptual Problems in the Definition and Interpretation of Attributable Fractions,” 128 Am. J. Epidem. 1185 (1988); James Robins & Sander Greenland, “The Probability of Causation Under a Stochastic Model for Individual Risk,” 45 Biometrics 1125 (1989); James Robins & Sander Greenland, “Estimability and Estimation of Excess and Etiologic Fractions,” 8 Statistics in Medicine 845 (1989); James Robins & Sander Greenland, “Estimability and Estimation of Expected Years of Life Lost Due to a Hazardous Exposure,” 10 Statistics in Medicine 79 (1991); Jan Beyea & Sander Greenland, “The Importance of Specifying the Underyling Biologic Model in Estimating the Probability of Causation,” 76 Health Physics 269 (1999; Sander Greenland, “Relation of Probability of Causation to Relative Risk and Doubling Dose: A Methodologic Error That Has Become a Social Problem,” 89 Am. J. Pub. Health 1166 (1999); Sander Greenland & James Robins, “Epidemiology, Justice, and the Probability of Causation,” 40 Jurimetrics 321 (2000).
Greenland’s criticisms turn on various assumptions such as the risk may not be evenly distributed within the sampled population, or the causal mechanism may accelerate onset of disease in such a way as to leave the relative risk unchanged in the study under consideration. Greenland is correct that it is important to have a clear causal model in mind when evaluating the possibility of causal attributions in the light of population studies and their measures of relative risk. He is also correct that his clever assumptions, if true, could affect the reasonableness of claiming that a relative risk of two or less supports the defense position in many toxic tort cases. Unfortunately, Greenland’s clever assumptions and his arguments prove too much, because in many, if not most, cases the causal model is not defined. There is often no evidence to support the plaintiffs’ claims of acceleration, or of sequestration of risk within the sampled population, and certainly no basis for claiming that the plaintiff belongs to a subset of “vulnerable” exposed persons with a higher than average risk that is reflected in the study relative risk. Without evidence to support Greenland’s various assumptions, even higher relative risks than 2.0, say risks in the range of 2.0 to 20.0, would be unhelpful to support a plaintiffs’ case. We would be thrown back to the early case law that held that risk can never support individual attributions, and Judge Weinstein’s rather pragmatic pronouncement in Agent Orange would be thrown aside, to the benefit of defendants in toxic tort cases.
Last year, the Vermont Supreme Court reaffirmed the continuing vitality of the relative risk argument, on the original pragmatic justification offered by Judge Weinstein in the Agent Orange cases. George v. Vermont League of Cities and Towns, 2010 Vt. 1, 993 A.2d 367 (Vt. 2010). Indeed, George may well have been one of the best, and the least unheralded, decisions of 2010.
Mr. George had been a fireman before he died of non-Hodgkin’s lymphoma (NHL). In administrative workman’s compensation proceedings, the Commissioner ruled that widow failed to show a causal connection between firefighting and NHL, although there was an “association.” His widow appealed the denial of benefits. On de novo review, the trial court excluded plaintiffs’ expert witnesses on Rule 702 grounds. (Vermont law follows federal law on requiring relevance and reliability of expert witnesses’ opinions.) The case ended up before the Vermont Supreme Court, which had to review the trial court’s handling of the Rule 702 issues.
Several issues were at play. The plaintiff had presented multiple expert witnesses, Drs. Tee Guidotti and James Lockey, who had presented general and/or specific causation opinions on firefighting and NHL. These witnesses relied upon epidemiologic studies, some of which had been incorporated into a meta-analysis, and a so-called “weight of the evidence” methodology.
The Vermont Supreme Court recognized the limits of using epidemiology to resolve the specific causation question in George. The Court found the Texas Supreme Court’s treatment of this issue to be persuasive:
“epidemiological studies can assist in demonstrating a general association between a substance and a disease or condition, but they cannot prove that a substance actually caused a disease or condition in a particular individual.”
Id. at 374 (relying upon and quoting from Merrell Dow Pharms., Inc. v. Havner, 953 S.W.2d 706, 715 (Tex.1997)).
The Court also quoted from, and relied upon, the pronouncement of the Federal Judicial Center’s Reference Manual, which explains that ‘‘epidemiology is concerned with the incidence of disease in populations and does not address the question of the cause of an individual’s disease. This question, sometimes referred to as specific causation, is beyond the domain of the science of epidemiology.’’ Id. at 375 (quoting from M. Green et al., “Reference Guide on Epidemiology,” in Reference Manual on Scientific Evidence 333, 381 (2d ed. 2000); footnote omitted in court’s quotation of this source).
Faced with the academic and judicial criticisms of using the relative risk (which is sometimes referred to as “effect size”), the Court recognized the pragmatic compromise between science and the needs of the legal system, embraced by using the relative risk as a benchmark showing for plaintiffs to make in toxic tort litigation:
“The trial court here adopted a relative risk factor of 2.0 as a benchmark, finding that it easily tied into Vermont’s ‘more likely than not’ civil standard and that such a benchmark was helpful in this case because the eight epidemiological studies relied upon by claimant’s experts reflected widely varying degrees of relative risk.”
Id. at 375.
“Given claimant’s burden of proof, however, and the inherent limitations of epidemiological data in addressing specific causation, the trial court reasonably found the 2.0 standard to be a helpful benchmark in evaluating the epidemiological evidence underlying Dr. Guidotti’s opinion.”
Id. at 377.
“Mindful of this balance, we conclude that the trial court did not abuse its discretion in considering a relative risk greater than 2.0 as a reasonable and helpful benchmark under the circumstances presented here.”
Id. at 378.
The Vermont Supreme Court was also clearly worried about how and why plaintiff’s expert witnesses selected some studies to include in their “weight of evidence” methodology. Without an adequate explanation of selection and weighting criteria, the choices seemed like arbitrary “cherry picking.” Id. at 389. This worry is amply justified. Weight of the evidence methodology is notoriously vague and indeterminate; unless the criteria for weighting are pre-specified and rigorously followed, claims based upon this methodology may be little more than subjective preferences. See, e.g., Douglas L.Weed, “Weight of Evidence: A Review of Concept and Methods,” 25 Risk Analysis 1545 (2005).
In part, plaintiff’s expert witnesses also relied upon a meta-analysis of observational studies that looked at NHL risk among firefighters. The Court was concerned about the plaintiffs’ expert witnesses’ failure to explain selection and weighting of studies in the meta-analysis methodology. This criticism may well be simply plaintiff’s witnesses’ failure to explain the methodology of a published study, which in turn may have properly used an acceptable methodology to provide a summary estimate of risk of NHL among firefighters. The meta-analysis in question, however, appears to have found a summary risk estimate of 1.51, with a 95% confidence interval, 1.31-1.73. G.K. LeMasters, et al., “Cancer risk among firefighters: a review and meta-analysis of 32 studies,” 48 J. Occup. Envt’l Med. 1189 (2006). The plaintiff’s expert witnesses were thus relying upon a study that quantifying the increased risk at 51%, with an upper bound from sampling variability, at 73%. To the extent that the plaintiff had succeeded in providing reliable evidence of increased risk, she had also succeeded in showing that a doubling, or more, of the risk for NHL was statistically unlikely. This is hardly a propitious way to win a lawsuit.
