Friday, December 28, 2007

TWO-OH (2.0) OR NOT TWO-OH: The Proper Role of Statistical Thresholds in Toxic Tort Litigation

This guest post was written by David Venderbush. Mr. Venderbush is counsel, resident in the New York office of Dechert LLP. This post is entirely his work. It represents only his views and not the views of his clients or firm.

The quiet period between Christmas and New Year’s Day seems like the perfect time to think about everybody’s favorite toxic tort topic: the use of epidemiologic relative risks as a substantive legal threshold. [NB: “epidemiologic,” not “epidemiological”] Of course, because the horde of corporate counsel, outside defense counsel, and justice lawyers (f/k/a trial lawyers (f/k/a plaintiffs’ lawyers)) who slavishly follow this site are probably away from the office, this post may be the blog equivalent of one hand clapping. (Cue: sound effect). With that in mind, this pedantic little nugget will be short and sweet and will assume a certain familiarity with principles of epidemiology and standards causation.

The take-home point here is that the fabled doctrine of 2.0 applies to proof of specific causation (did the particular plaintiff’s actual exposure cause the plaintiff’s alleged injury?) and not to proof of general causation (is the substance at issue at the dose alleged by plaintiffs capable of causing the particular injury?). See In re Silicone Gel Breast Implants Prod. Liab. Litig., 318 F. Supp. 2d 879, 893 (C.D. Cal. 2004) (“In re SBIs”) (the “‘doubling of the risk” requirement applies to statistical evidence proffered to prove specific, not general causation”). Getting this right matters because, when defendants erroneously urge 2.0 in the general causation context, they risk embarrassment, see id. (characterizing defendant’s general causation 2.0 argument as “based on a misunderstanding of relative risk, a mis-reading of Ninth Circuit precedent and a lapse in basic logical reasoning”), or worse, the loss of hard-won victories and years of work. See In re Hanford Nuclear Reservation Litig., 292 F.3d 1124 (reversing 762-page district court order granting defendants’ motion for summary judgment).

Characterizing 2.0 as an individualized causation doctrine may at first blush seem unscientific –and it actually is. The Federal Judicial Center's Reference Manual on Scientific Evidence states quite clearly that “[e]pidemiology is concerned with the incidence of disease in population and does not address the question of the cause of an individual’s disease." Reference Manual (2000) at 381 (citing inter alia, DeLuca v. Merrell Dow Pharms., 911 F.2d 941, 945 and n.6). But, as the Manual goes on to explain, “a number of courts have confronted the legal question of what is acceptable proof of specific causation and the role that epidemiologic evidence plays in answering that question.” Id. at 382. The Manual contains several pages on the 2.0 doctrine, with the caveat that that discussion “should be understood as an explanation of judicial opinions, not as epidemiology.” Id.

The 2.0 doctrine is thus not a matter of science, but a judicially created legal fiction that permits plaintiffs to attempt to prove specific causation when the lack of direct proof (because of the limits of scientific knowledge) makes proving causation impossible as a practical matter. Rather than toss plaintiffs out of court in the many cases involving injuries for which science cannot tell the precise cause (e.g., cancer, heart attack, birth defects), courts have permitted, as a matter of policy, proof of specific causation based on statistical inference. Still the best explanation of this legal doctrine comes from the Texas Supreme Court in Merrell Dow Pharms., Inc., 953 S.W.2d 706 (Tex. 1997), which I recommend reading in full to anyone who wants to understand this subject better. In summary, the Havner court explained the underpinnings of the doctrine:

In the absence of direct, scientifically reliable proof of causation, claimants
may attempt to demonstrate that exposure to the substance at issue increases the
risk of their particular injury. The finder of fact is asked to
infer that because the risk is demonstrably greater in the general population
due to exposure to the substance, the claimant's injury was more likely than not
caused by that substance. Such a theory concedes that science cannot
tell us what caused a particular plaintiff's injury. It is based on
a policy determination that when the incidence of a disease or injury is
sufficiently elevated due to exposure to a substance, someone who was exposed to
that substance and exhibits the disease or injury can raise a fact question on

Id. at 715 (citing Daubert v. Merrell Dow Pharms., Inc., 43 F.3d 1311, 1320 n. 13 (9th Cir. 1995) (“Daubert II”)). Thus, using epidemiology to prove individual causation (1) assumes that there is no direct proof of specific causation; (2) is a policy decision; and (3) relies on a logical inference.

The logical inference to be drawn from the number 2.0 arises from the more-likely-than-not standard of proof in civil cases. When an epidemiologic relative risk reaches 2.0, the agent is associated with an equal number of cases of disease as all other background causes. So the 2.0 legal fiction goes something like this:

Plaintiffs have the burden of proving specific causation under the more likely than not standard;
More likely than not implies greater than 50% probability;

An epidemiologic relative risk of 2.0 implies a 50% probability that the agent at issue was responsible for a particular individual’s disease;

Thus, a relative risk that is greater than 2.0 may permit the conclusion that the agent was more likely than not responsible for a particular individual’s disease.

See In re SBIs, 318 F. Supp. 2d at 893 (giving simple numerical examples to explain why the inference is logical).

It is important to remember, though, that a relative risk of more than 2.0 is not a “litmus test” and that a single epidemiologic study should not be considered legally sufficient evidence of causation; instead “[o]ther factors must be considered.” Havner, 953 S.W.2d at 718. Those factors include the Bradford-Hill criteria (which are beyond the scope of this allegedly short blog post), examining the design and execution of the studies relied on, and a showing that the individual plaintiff is similar to the those in the epidemiologic studies relied on such that the study results can be applied to the particular plaintiff. Id. at 718-20; see also In re SBIs, 318 F. Supp. 2d at 894. The context for such further analyses is usually the pretrial admissibility review for expert testimony under Federal Rule of Evidence 702 and state law equivalents.

The last point that I want to make is that the Ninth Circuit got this topic exactly right early on in Daubert II (which I also recommend reading in full for an intuitive explanation of the doctrine), but got it very wrong in In re Hanford, which I admit is a bit of a personal pet peeve. In Hanford, the district court bifurcated discovery to address first the general causation question of whether ionizing radiation at the levels to which plaintiffs were exposed were capable of causing the cancers and other injuries alleged. The district court got confused, however, and applied a 2.0 requirement at the general causation stage, granting summary judgment against any plaintiff who could not show that he or she had been exposed to a “doubling dose,” i.e., a dose of radiation that epidemiology showed more than doubled the risk of contracting that particular cancer.

On appeal, the Ninth Circuit correctly reversed the summary judgments on procedural grounds that the district court had violated its own discovery plan and had given plaintiffs inadequate notice of what they would have to prove in the ostensible general causation phase. In re Hanford, 292 F.3d at 1134. The Court should have stopped there.

It went on in dictum, however, to make a terrible 2.0 mistake as it attempted to distinguish Judge Kozinski’s correct treatment of 2.0 in Daubert II. The In re Hanford court stated, “It is critical to stress that the plaintiffs in Daubert II had no scientific evidence that Bendectin was capable of causing birth defects (generic causation), and therefore were required to produce epidemiological studies to prove that Bendectin more likely than not caused their own particularized injuries (individual causation).” Id. at 1136-37. Knowing what we now know about 2.0, we can see that that is an entirely nonsensical statement. If there is no scientific evidence of general causation, then there certainly wouldn’t be epidemiologic results showing a greater than 2.0 relative risk. Courts have repeatedly held in the Daubert context that without reliable proof of general causation, the inquiry never gets to specific causation. See, e.g., Kelley v. American Heyer-Schulte Corp., 957 F. Supp. 873 (W.D. Tex. 1997) (excluding specific causation testimony “[i]n the absence of any evidence regarding general causation”); Hall v. Baxter Healthcare Corp., 947 F. Supp. 1387, 1413 (D. Or. 1996)(specific causation proof “assumes that general causation has been proven”). Thus, the Hanford’s court statements on 2.0 should be viewed as mistaken dictum. All that was necessary to say was that the legal fiction of 2.0 should have played no role in the general causation phase.

We hope that makes your new year happy.

1 comment:

Nathan said...

On this peaceful New Year's day, I am reluctant to disturb the clarity and certitude of David Venderbush's helpful, "epidemio-logical" essay on the proper role for legal arguments based upon a relative risk (RR) of two in epidemiologic studies. But a couple of points are in order. First, the post is perhaps too kind in not pointing the finger at who was responsible for the Hanford Court's inappropriate use of the RR argument in a bifurcated procedure that dealt solely with general causation. Was this self-inflicted, invited error by the defense? Second, the post ignores that there is some semblance or relevance of the RR in a general causation analysis. If the RR < 2, especially in observational studies, many epidemiologists will have an increased concern that the estimate of risk is the result of bias or confounding. Obviously, that is a qualitative assessment, which does not affect whether or not the RR can be used for the quantitative inferences discussed. Third, the post is exactly right that the RR argument started as a way for plaintiffs to circumvent the problem that epidemiologic proofs do not provide information about individual causation. The argument was embraced by the tobacco industry, even in the face of very high RRs (> 20), and Courts eventually lost patience with the argument. In Agent Orange, Judge Weinstein articulated the RR > 2 argument as a way to suggest that individual causation might be satisfied, and also as a way to justify summary judgment when RR = 2, or < 2. With some trepidation, I must disagree with the post because the Weinstein approach is not illogical if the risk is stochastically distributed throughout the sampled population. Furthermore, the stochastic assumption is a reasonable, parsimonious default for the judicial system in dealing with uncertainty in the face of an established burden of proof. The assumption is, of course, a policy decision, but no more so than choosing a level of statistical significance, or the appropriate statistical tests to assess differences between two groups, exposed and unexposed.

Nathan Schachtman
Phillips Lytle LLP
New York, NY