Recent controversies over the safety of breast implants,¹ electrical power transmission lines,² and even cellular phones³ portend yet another period of protracted litigation in which the courts will confront issues of what constitutes admissible and sufficient evidence⁴ of causation in toxic torts.⁵ Questions have surfaced regarding the safety of each product, but there is no clearly established causal link between chronic exposure to any of them and disease or injury. News reports indicate that while anecdotal reports abound regarding breast implants, little if any systematic testing has been done to confirm suspicions of harmful effects.⁶ Concerns about cellular phones were prompted by an even sparser array of anecdotal reports and studies.⁷ Electromagnetic radiation from electrical power lines has been studied more extensively, but many scientists remain unconvinced of the purported link between such exposure and disease.⁸ Nonetheless, all three exposures are the subject of recently filed, and in some cases adjudicated, lawsuits.⁹

The current scare over cellular phones is instructive. The primary "evidence" of a causal link between the phones and brain cancer is the fact that a number of cellular phone users have been diagnosed with brain cancer, several with the cancer located near the location of the phone's antenna in use. Using newspaper estimates of over three million users of hand held portable cellular phones in the United States¹⁰ and 11,000 expected deaths from brain cancer this year,¹¹ it is hardly surprising that several cases of brain cancer in cellular phone users have been reported. One reported laboratory study which reported that radio-frequency radiation increased the growth rate of tumor cells is consistent with the possibility that such radiation could increase the growth rate of preexisting cancers,¹² but it does not prove that there is any effect in humans from cellular phone use.¹³

Despite the obvious lack of evidence to prove that cellular phone use causes brain cancer given the current state of knowledge, the evidence available today on cellular phones does not differ substantially in quantity or quality from the evidence that courts have found admissible and sufficient in other recent toxic tort cases. Those problematic cases are likely to be supported only by a combination of anecdotal evidence that amounts to no more than coincidence, speculation in the guise of scientific explanation, and testing based on unvalidated methodology or studies that have limited predictive value for human disease. Sometimes, as in the Bendectin litigation, such evidence is urged upon and accepted by courts in the face of overwhelming scientific consensus, supported by evidence, that a substance is unlikely to be a cause of injury. In other cases, very tenuous evidence is deemed sufficient where more probative positive or negative evidence is unavailable. Such unprobative and insufficient evidence and testimony, termed "junk science" by some observers,¹⁴ has been the subject of increasing commentary and criticism.¹⁵

Erroneous plaintiffs' verdicts and the corresponding overcompensation and overdeterrence are not just academic concerns. The prospect of useful products being driven from the market or of economic resources being diverted from productive uses is real, as the cases of vaccines¹⁶ and Bendectin¹⁷ illustrate. Submission of a case to the jury may result in a plaintiff's verdict where even the most cursory examination of the evidence reveals its deficiencies.¹⁸ Verdicts may be very large,¹⁹ and an occasional plaintiff's verdict may even encourage other suits and increase the settlement value of other cases.²⁰ The social and economic significance of breast implants, electrical transmission lines and cellular phones varies considerably, but clearly the costs to society of an erroneous conclusion that any of them causes harm are significant, potentially even catastrophic.

To deal with the problems of junk science in court, several commentators have suggested that courts regularize the standard for admissibility of scientific evidence. One frequent suggestion is that courts reinstate or continue to apply the standard announced in Frye v. United States,²¹ which requires that novel scientific evidence have general acceptance within the relevant scientific discipline,²² an issue that the United States Supreme Court is expected to address this year in Daubert v. Merrell Dow Pharmaceuticals, Inc.²³ As will be demonstrated in this article, however, many of the issues that arise are more properly viewed as questions about the sufficiency of relevant evidence to meet the more probable than not standard of proof. Thus, solutions that depend on tightening the criteria for admissibility will either require distortion of the admissibility inquiry to encompass sufficiency issues, or will address only part of the problem. Similar concerns are raised by proposals to change the rules of evidence to limit the use of expert testimony.²⁴

The problem of determining the sufficiency of evidence of causation is more directly addressed by proposals that courts use science boards, science panels or court-appointed experts to assist in resolving scientific issues.²⁵ Such proposals, however, except for the use of court-appointed experts, depart substantially from existing notions of civil jurisprudence because they involve delegation to experts of the traditional fact-finding functions of the lay trier of fact.

The thesis of this article is that measures such as the return to the Frye rule, or the use of science panels or science courts are unnecessary, because common law courts already possess the authority under the existing rules to "actively review"²⁶ scientific evidence by eliciting and scrutinizing the reasoning underlying scientific evidence and expert testimony and determining its validity and probative worth. As this article will demonstrate, much of the junk science that appears in toxic tort cases is readily apparent or easily uncovered by inquiry of which courts are quite capable.

If active review under the existing rules can uncover bad science, why do a significant number of courts take a lenient posture toward scientific evidence? There appear to be two major reasons for the deferential approach. First, some courts are philosophically indisposed to examine scientific reasoning or methodology, fearing that they are ill-equipped to delve into scientific disciplines. As will be described below, however, scientific reasoning and legal factfinding employ the same rules of logic. Thus, lay judges need not fear that examination of scientific evidence to determine whether it is soundly reasoned and reliable is beyond their capabilities.

Moreover, the reasons for judicial control of evidence are more compelling where technical evidence is concerned than for non-technical evidence. Judges exhibit no hesitation in barring non-expert testimony based on hearsay and otherwise lacking in foundation even though juries could readily identify the flaws in such testimony with skilled cross-examination and argument by opposing counsel. Juries are less likely to identify the weaknesses in testimony cloaked in technical jargon from an expert with a lengthy list of credentials than in testimony on ordinary factual issues.²⁷ Thus, it is more important for the judge, who understands the legal requirements of proof, to discriminate between reliable and unreliable scientific evidence than between well founded and unfounded evidence on matters within the understanding of ordinary people.²⁸

A second reason for the lenient treatment of scientific evidence in some courts is the apparent desire to compensate for perceived inequities and deficiencies of the tort system. Much of the movement toward the adoption of lenient standards of admissibility and proof of causation in toxic torts has been prompted by the recognition of the difficulties faced by plaintiffs in meeting the traditional requirement that they prove, by a preponderance of the evidence, that their injuries were caused by chronic, low-level chemical or radiation exposures that were remote in time from the manifestation of injury. The paucity of scientific evidence on the causation of diseases such as cancer and birth defects, and the difficulty of distinguishing other identified or background risk factors for the disease, decrease the likelihood that deserving plaintiffs will be compensated. The level of concern about those difficulties was heightened by increasing scientific knowledge of the role of chemicals and radiation in diseases such as cancer and birth defects, as well as scientific speculation about potential effects of the greatly accelerated dissemination of untested new chemicals in consumer products and the environment.²⁹ Taking their cue from the scientists,³⁰ legal scholars began to address the difficulties faced by plaintiffs in proving that exposure to toxic substances or chemicals caused their diseases or injuries,³¹ difficulties that can result in uncompensated injuries and the failure to adequately deter harmful activity.³² Lenient standards of admissibility and proof certainly facilitate plaintiffs' recoveries; further, they are consistent with courts' suspicions that mainstream scientists are too demanding in their requirements of proof, and that the unconventional scientists who testify that an exposure caused a plaintiff's disease may be correct.

More than a decade of scientific research into cancer incidence and causation, however, has failed to bear out the fears that prompted deferential review of causation evidence. Many of the assumptions that underlay the shift to more lenient standards for causation evidence in toxic torts are still unproven or are even contrary to current scientific thinking. The contribution of toxic synthetic chemicals and other hazards of the industrial age to cancer and other diseases and injuries is still an open question, but it appears unlikely that such substances cause anything approaching a majority of human cancer and birth defects.

As for the possibility that the unconventional expert may be right, even a superficial examination of much of the disputed evidence reveals that it amounts to speculation about possibilities that have not been tested or that fall far short of meeting the more probable than not standard of proof. Speculation about possibilities forms the beginning, not the endpoint, of factual inquiry, in either the scientific or legal realm. A causal explanation of disease or injury can be said to be probable only when it is supported by observations or data that distinguish between it and other possible explanations. When courts authorize or approve plaintiffs' verdicts without a factual basis for causal inference, they undermine traditional tort requirements for rational factfinding and the "more probable than not" standard of proof. The case for the abrogation of those standards has not been made, nor have courts given full consideration to the implications of such a radical change in the law.

The purpose of this Article is to demonstrate that courts can and should actively review scientific evidence of causation in toxic tort cases. The next Part describes how courts have loosened the standards for expert testimony in an effort to compensate for the perceived problems faced by toxic tort plaintiffs. Part III then discusses active review and its relation to the rules of evidence and civil procedure and attempts to allay courts' fears that they are ill equipped to evaluate the basis of scientific opinion testimony. Part IV then describes the criteria against which the reliability of scientific evidence can be evaluated and then applies those criteria to the kinds of evidence offered on causation in toxic tort suits. Part V examines a sampling of recent cases that illustrate inadequate judicial scrutiny of scientific evidence, as well as cases that skillfully distinguish probative from nonprobative or insufficient evidence. Lastly, Part VI discusses in depth the factors that underlie courts' failure to examine adequately scientific evidence and shows that many of those concerns are unjustified or that, even where justified, the remedy of authorizing plaintiffs' verdicts that are unsupported by a factual foundation goes too far.

As numerous commentators have explained, proof of causation³⁴ has been the biggest stumbling block to recovery in toxic torts cases.³⁵ Both negligence and strict liability require the plaintiff to prove that the substance in question³⁶ caused the plaintiff's disease or injury.³⁷ That inquiry often involves a number of subissues,³⁸ including whether: (1) the toxic substance is capable of causing the harm complained of³⁹; (2) the plaintiff was exposed to the toxic substance in quantity sufficient to cause disease,⁴⁰ and (3) the toxic substance exposure caused the particular plaintiff's injury or disease.⁴¹ Proof of any of these propositions is likely to require expert testimony on scientific evidence.⁴²

Several characteristics of the typical toxic tort case diminish the prospects of recovery by deserving plaintiffs.⁴³ The long latency period between exposure and disease manifestation⁴⁴ decreases the likelihood that the plaintiff will even suspect the causal connection, as well as decreasing the likelihood that the plaintiff will be able to marshal the facts on issues such as exposure necessary to prove her case.⁴⁵ Typically there is no clinical evidence capable of linking the substance to the disease.⁴⁶ The situation is further complicated by the fact that exposure to the toxic substance, even at relatively high levels, may not result in disease in most persons.⁴⁷ Moreover, many of the diseases caused by toxic chemicals, particularly cancers and birth defects, occur in the general population.⁴⁸ The absence of any unequivocal linkage between the disease and the toxin, together with the absence of clinical tests that could establish a linkage, means that proof of causation, if it can be made out at all, must be made indirectly, from comparisons between exposed and unexposed groups, or from studies where surrogates such as animals or single-celled organisms are used. Further, there may be other known risk factors for the claimed injury, whose role in the disease process must be considered.⁴⁹

The obvious difficulties of proof in toxic tort cases provoked a flood of commentary and proposals for reform.⁵⁰ A number of commentators have focused specifically on limitations placed by courts on the kinds of evidence deemed admissible or sufficient to prove causation. One set of problems has been courts' reluctance to accept statistical evidence, such as epidemiologic studies, because statistical evidence does not provide mechanistic explanations of cause and because statistics do not provide a basis for distinguishing between persons in an exposed group whose disease was caused by the exposure from those whose disease was caused by background or other risk factors.⁵¹ Recognizing that epidemiologic evidence is often the best if not the only evidence linking a toxic substance exposure to disease, however, recent cases have been more accepting of epidemiologic evidence,⁵² in some cases evidencing quite a sophisticated understanding of epidemiologic evidence.⁵³

Other commentators have urged courts to liberalize the standards for admissibility of scientific evidence in general.⁵⁴ They have suggested that the traditional requirement under United States v. Frye that limits the admission of scientific evidence to that generally accepted in the relevant scientific discipline⁵⁵ may preclude recovery by deserving plaintiffs who must rely on novel, yet valid and reliable evidence.⁵⁶ That line of reasoning was accepted in Ferebee v. Chevron Chemical Company,⁵⁷ in which the Court of Appeals for the District of Columbia Circuit upheld a jury verdict of liability based on expert opinion testimony on causation that did not enjoy general acceptance in the scientific community.⁵⁸

Ferebee coincided with a general move away from the Frye standard under the Federal Rules of Evidence toward the relevancy or reliability test articulated in United States v. Downing.⁵⁹ In Downing, the Third Circuit stated that the admissibility of scientific evidence should focus on the soundness and reliability of the expert's methodology, the strength of the connection between the evidence and the issues in the case, and the possibility of confusing or misleading the jury.⁶⁰ Acceptance of the expert's techniques or methodology in the relevant scientific community is evidence of soundness, but need not be the sole basis for that determination.

Although Frye has been justifiably criticized as too simplistic and inflexible,⁶¹ the Downing standard is equally problematic when it is used to justify such minimal scrutiny of the reliability of scientific evidence, particularly of expert opinion testimony, that it amounts to no standard at all. The troublesome, deferential application of the reliability standard adopts the approach that if "qualified" experts are willing to testify that a causal relationship exists, the court is willing to uphold a plaintiff's verdict without examining whether a reasoned basis exists for the expert's opinion.⁶² This approach is undoubtedly the result of some courts' reluctance to delve into the reasoning underlying scientific evidence, a reluctance that results in deference to the expert with seemingly impressive credentials. The crucial determination then becomes whether the expert is qualified, a particularly weak screening device given the lenient standards for determining expert qualifications.⁶³

Deferential review is the gateway for the admission of junk science into the courts. When courts do not examine the reasoning of expert testimony, they are likely to accept medical opinion based on the facts in the case at hand, or supported by perhaps a few other case reports, facts that cannot establish causation because the coincidence of exposure and disease may be the result of chance.⁶⁴ In some cases, courts accept as sufficient medical or similar opinions supplemented by reference to animal studies, chemical structure-activity analyses, mutagenicity testing, or other similar lines of reasoning that are subject to a large degree of uncertainty.⁶⁵ Affirmative epidemiologic evidence of a statistically significant association between the alleged causative agent and human disease is absent.⁶⁶ As a practical matter, only those cases based on studies in human populations of the association of suspected toxic substances and disease-e.g., epidemiologic studies or highly unusual disease clusters-have proven to be sound as new scientific information developed.⁶⁷

A reliability analysis should not result in uncritical acceptance of junk science.⁶⁸ Tort jurisprudence requires that there be a rational basis for judicial findings of fact.⁶⁹ The relevancy or reliability standard's "soundness and reliability" inquiries bear directly on whether there is a rational basis for findings of fact and whether the evidence is sufficient to meet the more probable than not standard of proof.⁷⁰ Active review facilitates the inquiries necessary to decide those issues, while deferential review avoids them. Courts cannot and should not avoid those responsibilities by deferring to "qualified" experts.

The second inquiry focuses on the sufficiency of the admissible evidence to meet the plaintiff's burden of proof. This inquiry goes to the reliability or accuracy of the evidence and requires that the plaintiff present admissible evidence from which a reasonable juror could find that it is more probable than not that the defendant caused the plaintiff's disease or injury. The same tools used to probe the underlying reasoning of the evidence can be used to inquire into its accuracy, but the question of whether the evidence is sufficiently accurate to satisfy legal standards is, of course, a legal question.

It is important to note that active review is not strict scrutiny.⁷² The plaintiff need not show that her evidence is stronger than the defendant's or that it meets some high level of certainty. The plaintiff's scientific evidence need only be such that a rational factfinder could conclude from the testimony that it is more likely than not that the defendant caused the plaintiff's injury.⁷³ Only when the factual basis and reasoning underlying the expert's opinion on causation do not meet that minimum level of rationality and accuracy should the evidence be excluded.

Active review is not tied to any particular formulation of the standards for admissibility of expert testimony. It is, however, more easily related to the "reliability" determination embraced by a number of courts⁷⁴ than it is to the general acceptance rule of United States v. Frye.⁷⁵ The Frye rule forecloses the occasion for the court to examine the reasoning underlying the expert's method; however, it leaves questions such as the applicability of a generally accepted method to a particular case, the way in which a generally accepted method was carried out in a particular case,⁷⁶ and the sufficiency of the evidence to be addressed under other criteria. Thus, even if the United States Supreme Court upholds the application of the Frye rule in Daubert v. Merrell Dow Pharmaceuticals, Inc.,⁷⁷ it will not eliminate the need for courts to actively review scientific expert testimony.⁷⁸

Judge Markey has succinctly stated an essential distinction between science and technology on the one hand, and law on the other:

Unfortunately, these institutional and methodological differences obscure the reality that factfinding, that is, science in its broadest sense, is a necessary part of legal decisionmaking. Legal decisionmaking has additional policy components beyond the purely factual, so that it may attach different consequences to the same facts than would a scientist. Thus, the starting point for the analysis of the relationship between science and law on the issue of causation is a delineation of the factual and nonfactual components of legal concepts of cause.

To be sure, causation issues in tort law have nonfactual, policy-laden elements, as exemplified by the legal concept of proximate cause.⁸⁵ All tort theories include some notion of "cause-in-fact" as a prerequisite to liability,⁸⁶ however, and where cause-in-fact is concerned, science and law are attempting to answer the same questions. Further, law, like science, accepts only rational or reasoned findings of fact.⁸⁷ Most importantly, scientific and legal factfinding employ the same logic.⁸⁸

Much of the early commentary about the differences between science and law in toxic torts concerned courts' discomfort with statistical evidence of causation. Commentators have attributed that discomfort in part to courts' preferences for mechanistic causal explanations and their reluctance to rely heavily or entirely on statistical evidence.⁸⁹ Courts and lay persons typically think about causal issues in terms of how things happen and statistical evidence does not explain how events occur.⁹⁰

When mechanistic thinking about cause is extended to the area of toxic substance disease causation, it immediately encounters a large, perhaps insurmountable, stumbling block. Scientists know very little about how, in a mechanistic sense, toxic substances cause diseases such as cancer or injuries such as birth defects.⁹¹ Nonetheless, they may know a considerable amount about whether toxic substances cause disease or injury through inferences drawn from statistical associations and other indirect means.⁹² Thus, the shift in thinking required for courts to come to grips with current scientific knowledge had more to do with abandoning a felt need for an explanatory process that increases comfort with the causal inference than it did with redefining causation.

Courts' discomfort with statistical evidence has gone beyond the absence of mechanistic explanations, however.⁹³ Statistical evidence by definition provides information only about the incidence of disease in groups. Where there are other possible causes of disease, statistical evidence cannot determine which individuals' diseases within the exposed group were caused by background or other factors.⁹⁴ It can only provide an estimate of the likelihood that an individual's disease was caused by the toxic substance in question.⁹⁵ Thus, courts' concerns are not unreasonable. The more likely than not standard of proof, however, implicitly contemplates the marshaling of facts that ultimately prove liability in terms of probabilities.

Uncomfortable with factual indeterminacy, some courts rejected statistical evidence entirely, demanding evidence that is particular to the plaintiff.⁹⁶ Other courts have accepted statistical evidence on issues such as whether a toxic substance is capable of causing harm, but not on the question of whether it caused the plaintiff's harm.⁹⁷ A number of recent cases, however, have recognized the necessarily statistical nature of proof at all levels in toxic torts, and accepted statistical evidence as probative of individual causation, at least where there is evidence indicating a greater than 50% likelihood that the toxic substance caused the plaintiff's disease.⁹⁸ A number of recent decisions evidence a sophisticated understanding of epidemiologic evidence and its relation to legal standards of proof.⁹⁹

The remaining areas where science seems to fit poorly with legal problems are largely the result of failure to distinguish legal standards of proof from factual issues. Courts are concerned that they must decide cases based on the information available, which may not be complete enough to satisfy the requirements of a particular scientific discipline.¹⁰⁰ Some courts perceive scientists as generally requiring higher levels of certainty than does the law.¹⁰¹ That perception may be correct in some instances, particularly in areas such as epidemiology, where standard protocols for statistical analysis of relative risk data typically require a 95% level of certainty that an observed increased in risk is not due to chance.¹⁰² Scientists do not require a high degree of certainty for all purposes, however. Risk assessment for purposes of regulation is based on highly uncertain risk estimates. Additionally, scientists often use highly tenuous or uncertain assumptions in making decisions about further research.¹⁰³

The issue of how much uncertainty is acceptable is a legal requirement to be applied to the evidence once the uncertainty attending the scientific evidence is established.¹⁰⁴ Where the law requires the plaintiff to prove her case by a preponderance of the evidence, current standards permit the plaintiff to win if sufficient evidence is available, but not prevail if it is not available.¹⁰⁵ Scientific evidence can be evaluated against those standards, irrespective of whether the scientific discipline would be satisfied or not with the available level of certainty.¹⁰⁶ Moreover, the fact that scientists may require a different level of certainty is not a good reason to dispense with science's requirement of a reasoned analysis, a requirement common to law and science. Unfortunately, some courts throw the baby out with the bathwater by rejecting scientific reasoning altogether when they perceive scientists' requirements for certainty to be too stringent.¹⁰⁷

Courts can and should evaluate the underlying reasoning of scientific evidence and measure its reliability or uncertainty against legal standards of sufficiency to meet the applicable burden of proof. The following part of this article attempts to facilitate that process by explicating the bases on which courts can recognize and reject invalid or unreliable evidence, matters on which the differences between science and law are a matter of degree, not kind. Thus, courts need not fear that delving into science and technology will be entirely a foray into alien territory.

Consider, for example, a diagnostic blood test for a viral blood disease. Without the blood test, the disease can be diagnosed only by elaborate procedures. A simple test is desired for screening large numbers of blood samples for the presence of the virus. A virologist might speculate about any number of parameters that might be indicative of the presence of the virus. None of the possible indicators could be used as a diagnostic test, however, until validated by testing that demonstrates a correspondence between the indicator (a "positive" test) and the presence of the virus. This example illustrates the more general principle that where the physical connections between observed and inferred facts are hidden from direct observation, it is necessary for the inferred connection (e.g., between the indicator and the virus) to be validated through trials or tests that independently measure the properties or characteristics that are ostensibly connected.¹¹⁴

A valid method may nonetheless be insufficiently reliable for evidentiary purposes; that is, the method may be incapable of producing the desired information to an acceptable level of certainty. Using again the example of a test for an asymptomatic virus, the test might have a high rate of false positives or false negatives, or both. Thus, although persons who are test positive are more likely than those who test negative to actually have the virus in their blood, the test may be too inaccurate or unreliable for the purpose for which it is administered.¹¹⁵ Similarly, if the question of whether someone is infected with the virus were a factual question in a legal setting, our hypothetical test might be insufficiently reliable to satisfy the legal standard of proof.¹¹⁶

Validity or reliability questions may arise when methodology that has proved valid and reliable is applied in new circumstances. Invalid application of valid methodology may result from extending a method or line of reasoning to purposes for which it has not been validated.¹¹⁷ In toxic torts, this question arises in connection with whether the conclusions derived from toxicological research on animals or single-celled organisms are applicable to humans.¹¹⁸

Uncertainty or reliability questions may also result from the improper application of valid and reliable methodology. Failure to properly calibrate an instrument such as a breathalyzer, or other concerns related to how a method is applied in a particular case, may increase the likelihood of erroneous results.¹¹⁹ Assume, for example, that in the hypothetical virus test, the incidence of false negatives increases with the length of time that the patient's blood samples are stored before the laboratory test is run. The inferences drawn from a test run by a laboratory that stores its blood samples longer that the optimum time for the test would be subject to a greater variation and uncertainty than results from a laboratory that runs its tests promptly.

A subset of questions regarding "reliability as applied," particularly where the methodology involves calculations from raw data, concerns the quality and quantity of the underlying data. In toxic torts, the data on which estimates of exposure to a toxic substance are based are often sketchy or subject to large uncertainties. Those uncertainties make the inferences of causation that depend on the exposure data similarly uncertain and unreliable.

As noted previously, the characteristics of toxic tort cases impose limitations on the ability to establish causal connections between exposure and disease. The latency periods typical of toxic tort injuries, the absence in most cases of a unique signature injury associated with a toxic substance, the fact that injury does not occur in every instance of exposure, and the absence of clinical indicators that discriminate among causes of a particular individual's disease all tend to obscure toxic injury causation.¹²⁰ In simple terms, the typical toxic tort case looks something like this: The plaintiff believes she has been exposed to a toxic chemical. She has a disease that is commonplace, or at least not unknown, in the general population. The current progress of the disease bears no relation to the continuation of exposure, and the exposure may have long since terminated. There is no diagnostic or clinical test that can determine what caused her disease.

How can such a plaintiff prove that a toxic substance caused her disease? Because of the absence of clinical indicia of cause, the plaintiff must always make out her case indirectly. First, she needs evidence that the substance can cause the condition from which she suffers and of the circumstances under which disease causation is reasonably likely to occur. The coincidence of exposure and disease in the same individual, while necessary, can never be sufficient to prove the capability of the substance to cause disease. Similar problems attend the use of anecdotal case reports or evidence of clusters of disease that have not been subjected to statistical analysis because a certain amount of coincidence and toxic chemical exposure or even clustering of a disease can occur as the result of random chance.¹²¹

Second, she needs to establish that she is within the class of persons to which inferences from the general causation evidence should be applied. This second, particularistic causation component of proof, which is discussed later in this article, usually involves two parts: proof of sufficient exposure to permit the inference that the general causation evidence is applicable to her and a demonstration that other causal explanations, including background causes, are less likely causes than the toxic substance exposure.¹²²

Epidemiologic studies are expensive to conduct and are subject to a number of limitations on the size of the effect they can detect.¹²⁷ Thus, it is sometimes argued that case reports and clusters of disease constitute sufficient evidence of the capability of a substance to cause toxic injury.¹²⁸ Case reports and disease clusters are sometimes sufficient to raise suspicions and stimulate investigation of toxic chemicals as causative agents.¹²⁹ Benzene was identified as a leukemogenic agent through clinical studies of case reports beginning in the late 1800s,¹³⁰ and vinyl chloride was more recently recognized as carcinogenic through the appearance of clusters of angiosarcoma of the liver in plant workers in the early 1970s.¹³¹ Those examples, however, are typified, in the case of benzene, by very high exposures and the accumulation of evidence over decades,¹³² or by the unexpected appearance of an otherwise very unusual disease.¹³³ Such identifications through case reports and clusters, however, depend on at least a rough sense that the incidence of the disease in the exposed group exceeds the background rates,¹³⁴ even if the reports of unusually high incidence are not initially subjected to the same rigorous statistical analysis as is typical of an epidemiologic study.¹³⁵ Moreover, those initial clusters or unusual case reports will often suggest other places to look for additional evidence, such as workplace exposures involving the same substance, or other users or consumers of the suspect chemical.¹³⁶ The absence of similarly affected individuals among other populations with similar exposures would suggest that the cluster is a statistical accident rather than a true cluster.

Case reports and apparent disease clusters are likely to be argued in toxic tort cases in circumstances where they do not have even minimal indicia of reliability. In Renaud v. Martin Marietta Corp.,¹³⁷ the plaintiffs argued that the existence of four cases of childhood cancer in Friendly Hills, an area in which only two would have been expected, was evidence that the substances allegedly in their water supply had caused their cancers.¹³⁸ Plaintiffs' experts agreed, however, that the Friendly Hills population was too small to yield meaningful results. Moreover, another expert's opinion was that four cases of childhood cancer was within the expected range for the community.¹³⁹

Animal studies are based on the theory that substances that cause harmful effects in animals are likely to cause similar harmful effects in humans.¹⁴³ That thesis is supported by observations that many substances that cause harmful effects in one species also cause harmful effects in other species.¹⁴⁴ All but one of the chemicals identified by epidemiologic studies as causing cancer in humans have also proven to be carcinogenic in one or more animal species.¹⁴⁵ Thus, there appears to be some correlation between carcinogenicity in animals and carcinogenicity in humans. Similar observations and findings have been made with respect to other kinds of toxic effects, including teratogenic effects.¹⁴⁶

As any observer of the popular media knows, however, animal testing for diseases such as cancer, which has a long latency periods, and for which even low incidence rates are of concern, are conducted under conditions that are very different from the usual human exposure scenario.¹⁴⁷ Animal studies of carcinogenicity typically utilize doses at or near the maximum level tolerated by the animal.¹⁴⁸ That practice is necessitated by the need to detect effects in relatively small groups of test subjects, in a relatively short period of time. Those same concerns also have led to protocols using animal strains bred for their susceptibility for tumor formation.¹⁴⁹ Additionally, the route of administration may differ from the likely human exposure route.¹⁵⁰

The prediction of effects in humans from animal testing involves a number of extrapolations-from animal species to humans, from one route of administration to another, and most acutely, from a high-dose exposure in which the animals are typically subjected to the maximum dose they can tolerate (the MTD),¹⁵¹ to a low-dose chronic exposure.¹⁵² Each of those extrapolations introduces uncertainty into the predictive value of animal testing in proving causation of human disease.¹⁵³ Differences in species can have a dramatic impact on the effects of a toxic substance,¹⁵⁴ as can routes of administration.¹⁵⁵

The high dose exposure scenario of typical animal testing protocols raises several concerns. One concern relates to the model used to extrapolate the results of high dose exposures to the much lower doses encountered by humans. The lack of a complete mechanistic understanding of cancer causation precludes the adoption of any particular extrapolation model with a high degree of certainty.¹⁵⁶ For example, one possible set of assumptions is that no dose of a carcinogen is completely risk free and that the disease incidence rate will be directly proportional to the dose. Those assumptions lead to a linear extrapolation model.¹⁵⁷ Another possibility, which apparently applies to some carcinogens, is that at very low levels, a toxic chemical exerts no adverse effects and that such effects appear only when a threshold level of exposure is exceeded.¹⁵⁸ The set of assumptions adopted in a particular instance can lead to vastly different predictions of the effects of low dosage exposures, sometimes as much as several factors of ten.¹⁵⁹

The accuracy of risk extrapolations from exposure of animals to the MTD has recently been called into further question by prominent researchers in the field of carcinogenesis.¹⁶⁰ Bruce Ames, the developer of the "Ames test" for mutagenicity,¹⁶¹ now argues that risk estimates obtained under such circumstances are largely due to toxic effects of the test chemical, rather than factors that might operate at lower doses in human.¹⁶² Thus, the results from animal studies may not be predictive of human carcinogenicity under the usual exposure scenario.

Quite a number of toxic torts plaintiffs have offered animal studies in support of their contentions that the substances in question can cause harm in humans.¹⁶³ In some cases courts have been willing to entertain such evidence,¹⁶⁴ while other have found it inadmissible¹⁶⁵ or insufficient.¹⁶⁶ The closer examination of the assumptions and methodologies involved in animal testing, however, reveals that the extrapolation of animal test results to humans is too uncertain-the potential for error is too high-for animal testing alone to support an inference that it is more probable than not that a substance causes cancer or birth defects in humans at a specified level of exposure.¹⁶⁷ There is considerable doubt about the inference that an animal carcinogen is a human carcinogen at all. Even if that hurdle is assumed away, however, the uncertainties that attend the interspecies and high-dose to low-dose extrapolations necessary to extend animal test results to human exposure scenarios are simply too large. Policy considerations in the regulatory arena dictate or at least support the use of models that overpredict rather than underpredict risks levels.¹⁶⁸ Risk estimates based on unproven dose-response models, where the choice of model may alter results by a thousand times or more, however, are not consistent with a more likely than not standard of proof.

Another kind of evidence, typically offered as evidence of causation of birth defects or other noncancerous disease or injury, is in vitro testing, tests involving exposure of isolated groups of cells or organs¹⁷⁵ to suspect chemicals. To test for teratogenesis (birth defects), fetal cells or embryos may be used. These tests are fraught with uncertainties, however, related to whether and to what degree the chemical in question would reach or react with the sensitive cells or organs in a whole organism.¹⁷⁶ They are also fraught with the same uncertainties relating to interspecies extrapolation as is animal testing generally.

Toxicological research into the causes of human disease, when direct evidence in humans is unavailable, proceeds according to a hierarchy of reasoning, from the least costly and time-consuming, and least predictive methods, to the most costly, time-consuming methods that are believed to correspond most closely to human response. Thus, the investigation of the toxicological properties of a chemical is likely to start with the analysis of available information about chemicals with similar structures-chemical structure-activity analysis.¹⁷⁹ The most likely candidates identified by structure-activity analyses are then subjected to biological assays such as mutagenicity testing or in vitro testing on cell groups. Lastly, animal testing will likely be conducted on chemicals that exhibit toxic effects in the short-term screening procedures. Structure-activity analysis and short-term screening are not the end points of the evaluation process, even in a regulatory context, because they are recognized as significantly less valid and reliable than animal testing. Whether considered separately or in the aggregate, the methods that do not involve observations of disease in humans are too likely to lead to an erroneous conclusion to satisfy the traditional burden of proof.

To understand what kinds of evidence are probative of individual causation, we must first make reference to the kinds of evidence probative of the capability of the substance to cause harm, namely, epidemiologic evidence or possibly other human evidence of sufficient reliability. That evidence will identify one or more diseases that are believed to be causally associated with exposure to a toxic substance. Such studies will also typically be based on or identify certain levels or ranges of exposures. The plaintiff must argue that the association established through the statistical study applies to him and that background and other risk factors are less likely causal explanations.

incidence in exposed group

Relative Risk = ----------------

incidence in unexposed group

Black provides an example in which the disease rates in exposed and unexposed groups are 50 and 5 per 100,000 population respectively.¹⁸² The relative risk in that case is 50/5 or 10, indicating that the exposure increases the disease rate to ten times that of the background rate.

How can a toxic tort plaintiff use such information? At a minimum, it would seem obvious that the plaintiff must have a disease identified as associated with the toxic chemical exposure. Nonetheless, it is not uncommon for plaintiffs' experts to assert that evidence that a substance causes any cancer is evidence that it can and has caused other cancers.¹⁸³ Although substances that are discovered to cause one type of cancer may cause other types of cancer as well, that possibility does not permit a prediction of what those other cancers, if any, are likely to be.¹⁸⁴ The problem is compounded when the plaintiff's general causation evidence is not based on human evidence, but on animal studies or other less reliable methods that only generally suggest a possible carcinogenic effect;¹⁸⁵ in such cases, the evidence may not allow an identification of any particular disease associated with the substance.

If plaintiff has a disease associated with the toxic substance exposure, demonstrating that it is more likely than not that plaintiff's condition was caused by exposure usually will require the plaintiff to demonstrate that her exposure was in the range found to be associated with an increased risk of disease.¹⁸⁶ Alternatively, plaintiff might be able to show that the results of the study could be extrapolated to lower doses.¹⁸⁷

If the plaintiff can demonstrate sufficient exposure to argue that the relative risk factor in the study applies to him, he still must differentiate background causes where the disease occurs in the background population. Under the traditional more likely than not rule, the plaintiff will prevail if the relative risk identified in the epidemiologic study is greater than two. That is because relative rate greater than two corresponds to more than doubling of the disease rate, permitting the inference that more than half of the cases of the disease in the exposed population were caused by the exposure. When applied to the plaintiff, the inference can be made that it is more likely than not that the exposure caused the plaintiff's disease.¹⁸⁸ Put another way, when the relative risk is greater than two, the fraction of all disease in the exposed group attributable to the exposure is greater than 50%.¹⁸⁹ The causal connection inferred from the presence of a signature disease represents the application of this principle when the relative risk is very large and the corresponding risk attributable to the exposure may be ninety percent or more.¹⁹⁰

Plaintiffs who lack an epidemiologic basis for their proof of causation nonetheless frequently offer medical testimony from a treating physician or other expert who opines that the plaintiff's disease was caused by toxic substance exposure.²⁰³ This form of opinion is evident in Renaud v. Martin Marietta Corp.²⁰⁴ and other recent cases. When there is no clinical test that establishes a cause or distinguishes among possible causes, however, such "intuition" can only be characterized as speculation. This kind of speculation could easily be unmasked by inquiring into the reasoning behind the witness's opinion.²⁰⁵

Plaintiffs often argue and courts sometimes accept the notion that the absence of other risk factors increases the likelihood that the plaintiff's disease was caused by the toxic exposure at issue.²⁰⁸ The validity of that kind of reasoning, however, rests on two unstated, and usually untested, assumptions. First, such reasoning treats toxic exposure and the other risks as alternatives. In other words, it assumes that the disease was caused by the toxic exposure or some other cause, such as the other identified risk factors.²⁰⁹ Second, it assumes that most causes of the disease in question are known; otherwise, the elimination of other risk factors would not significantly increase the likelihood that the toxic exposure was the cause of the plaintiff's disease.

The assumption that risk factors, including the toxic exposure, represent alternative causes is true only if the various risks are additive. Additivity is only one of several ways in which risk factors for the same disease may relate. The combined effects may be the same, greater, or less than the sum of the effects as measured separately.²¹⁰ Additive effects represent the absence of interaction between risk factors.²¹¹ Thus, each factor adds an incremental level of risk to the background risk that is independent of the presence or absence of other risk factors. Additive risks are properly treated as alternative risks in a causation analysis.²¹² Often, however, the information necessary to support that assumption is not available.

Risk factors whose combined effects are greater than additive are considered interactive or synergistic.²¹³ In this situation, each risk factor enhances the risk contributed by the other factor so that the total incidence of the disease is greater than the sum of the incidence attributable to each factor separately, sometimes approaching a multiplicative effect. Perhaps surprisingly, the presence or absence of other risk factors that are multiplicative does not increase or decrease the fraction of disease attributable to the toxic exposure. Thus, when the causes are synergistic, as with smoking and asbestos and lung cancer,²¹⁴ it is incorrect to pose the question as one of whether the disease was caused by one factor or another.²¹⁵

There are several scenarios under which this issue could arise, but the actual cases tend to be grouped into two extremes. Where epidemiologic data are available that address the contributions of both the toxic substance and other causal factors, the plaintiff's attributable risk and probability of causation by the toxic substance can be determined by calculating attributable fractions, whether the risks are additive, multiplicative or antagonistic. From those calculations, it can be determined whether the plaintiff can satisfy the more likely than not standard of proof on causation.²¹⁶

The other extreme is represented by toxic tort cases where multiple risk factors are treated in a vague or qualitative fashion and data are not available to support a quantitative analysis. The plaintiff's expert may opine that because other known risk factors are absent in plaintiff's case, it is the expert's opinion that plaintiff's disease was caused by the toxic substance exposure.²¹⁷ This argument has superficial appeal. It can only be valid, however, when risk factors that account for most cases of the plaintiff's injury and their interactions are understood. Although some risk factors for cancers and birth defects have been identified, the causes of background incidence of most birth defects and cancers remain unknown. Even if the identified risk factors are alternative, independent causes, as this line of analysis assumes, the expert's opinion distinguishes among factors that make up only a small part of the total picture, while ignoring the probability that the plaintiff's injury may stem from the same unidentified factors that are responsible for most cases of the injury. Whether the differential diagnosis argument is made on behalf of plaintiff or defendant, it adds little to the resolution of the case when it is based on vague, qualitative assumptions about alternative causes.²¹⁸

Some cases, particularly those involving workplace exposures, involve situations that are known to involve exposure to a toxic substance.²²² The severity of the plaintiff's exposure can be inferred from the length of time he was present in the environment. The situation would be similar where there is an ongoing exposure, such as a drinking water exposure, that can be measured. In such cases, inferences about past conditions can be drawn from the present ones.²²³

Sometimes there is clinical evidence of the toxic substance that will suffice to prove the plaintiff's exposure.²²⁴ Substances such as asbestos²²⁵ and PCBs²²⁶ remain in tissues indefinitely, and can be detected by appropriate analytical tests. Other substances may result in subclinical changes that can be detected through appropriate testing.²²⁷

Models, at least conceptually, sometimes begin with theories or hypotheses about how different kinds of data might be related. A scientist would be unlikely to propose a model that was not at least plausible, based on her understanding of how the phenomena in question are connected. Plausibility alone, however, like the mechanistic theories of cancer causation, is often insufficient to eliminate other plausible but untested models or theories. Thus, a model must be validated before an expert or a court should assume that it has any probative value.

Validation of a model involves testing the model's predictive capability in circumstances where the expected results can be independently measured.²²⁹ In the case of the blood/adipose tissue partition model or the much more complex groundwater contaminant migration modeling, actual concentrations can be measured and compared with values predicted by the model. From such data, it can be determined whether the model has any predictive value and how accurate or reliable those predictions are. If the model proves valid and reasonably accurate, then it is reasonable to apply it to other situations similar to the one for which the model has been tested.

A number of cases, however, have involved modeling of exposures where the models have not been subjected to the most cursory validation, sometimes in the face of data that contradict the model. In In re Paoli Railroad Yard PCB Litigation,²³⁰ one of the ways the plaintiffs attempted to prove exposure to PCBs was by showing that their PCB levels were elevated above background levels. Because PCBs accumulate in fatty tissue and are not quickly eliminated from the body, it is possible to measure PCB levels in blood or tissue samples from individuals and compare them to norms for the general population.²³¹ The Agency for Toxic Substances and Disease Registry (ATSDR) had done a study on blood PCB levels in Paoli residents and concluded that the residents' serum PCB levels did not differ significantly from those of the general population.²³² Plaintiffs contended that the ATSDR study's conclusions regarding background PCB levels in the general population were erroneous. Their experts sought to show that the plaintiffs' adipose or fat PCB levels exceeded norms found in the Environmental Protection Agency's National Human Adipose Tissue Study.²³³ Because most of the plaintiffs' PCB levels were determined by blood tests alone, plaintiffs' expert used his own formula to calculate their adipose tissue PCB levels, which he then compared with results reported in the national study.²³⁴

As the trial court recognized, however, neither of the plaintiffs' experts on this issue²³⁵ cited any basis for the claimed relationship between PCB concentrations in blood and adipose tissue.²³⁶ Moreover, where blood and adipose tissue PCB levels were measured in the same plaintiffs, the results did not bear out the ratios asserted by plaintiffs' experts.²³⁷ Thus, although a relationship between blood and adipose tissue levels of PCBs is plausible, the direct ratio posited by plaintiffs' witnesses was not validated and was demonstrably inaccurate as indicated by comparison of predicted levels with actual measurements in plaintiffs who had both tests. Conclusions based on models that have not been validated by actual measurement,²³⁸ or worse, which are contradicted by actual measurements, are based on invalid reasoning and should be rejected by the courts.

Use of unvalidated models is not the only concern about models. Models inherently involve approximations-generalizations about physical phenomena and estimations that must be made because the actual situation cannot be studied directly.²³⁹ Those approximations inevitably introduce inaccuracies into the model's predictions. At some point, the uncertainty or inaccuracy may become so large that the model's results are too unreliable to prove the fact on which they are offered.

Modeling of groundwater and surface water contamination was at issue in Renaud v. Martin Marietta Corp.,²⁴⁰ in which the plaintiffs claimed that contamination of the public water supply had caused childhood cancers and other diseases.²⁴¹ Plaintiffs' case on exposure involved the issue of whether contaminants released at the Martin Marietta facility had reached their taps through the Denver culinary water distribution system. Because the circumstances that created the discharges at issue had changed in the years preceding the suit, the plaintiffs relied on hydrological modeling of ground and surface water movement as proof that contaminants had reached the water distribution system.²⁴²

Defendants argued that there were serious flaws in the modeling, most notably that plaintiffs had erred by failing to consider all relevant factors when deriving the decay coefficient for the contaminants.²⁴³ Further, the experts had not taken into account the possibility that chlorination at the water intake plant had destroyed or greatly reduced the concentration of contaminants.²⁴⁴ Although the court stated that "[t]he issues of which factors should have been considered and what impact each should have been given" were questions for the jury, it seems clear that the factors that plaintiffs' experts ignored would have had a large impact on the concentrations predicted by the fate and transport modeling.²⁴⁵ Models are always subject to dispute over the factors that are included or excluded, and thus cannot be judged by too rigorous a standard. Nonetheless, where experts have excluded significant factors that would tend to produce results at odds with their conclusions, the court should exclude the modeling results unless there is some more direct way to demonstrate the model's validity and accuracy.²⁴⁶

Another concern with modeling is that uncertain input data affect the reliability of the results of all kinds of exposure modeling. If the input data are very limited, there will be uncertainty about how representative those data are, and those uncertainties will produce corresponding uncertainties about the modeling results, no matter how good the model is. The greatest uncertainty about the exposure modeling in Renaud resulted from such a scarcity of data.²⁴⁷ In that case, the plaintiffs' exposure estimate, obtained through the transport modeling described above, was based on a single loading concentration for the contaminants. The court recognized that the single data point on which the modeling was based could not be said to be representative of the 11-year period over which releases occurred.²⁴⁸ It found the single data point to be a fatal flaw in the plaintiffs' exposure case.²⁴⁹

The Third Circuit's decision in In re Paoli Railroad Yard PCB Litigation²⁵⁰ represents one of the most troubling decisions on the admissibility and sufficiency of challenged scientific evidence. Paoli involved an action by 38 neighbors and employees of an electric railcar maintenance facility contaminated by PCBs.²⁵¹ The action, which was brought against owners and operators of the site, and suppliers of PCBs and transformers, made claims for various injuries and for medical monitoring costs necessary to protect against latent disease.²⁵²

Defendants' motions to exclude plaintiffs' evidence of exposure to PCBs and other causation evidence, and for summary judgment were granted by the trial court.²⁵³ The Third Circuit, however, reversed, finding that the trial court had improperly excluded sufficient evidence to survive summary judgment.²⁵⁴ The case represents a virtually complete catalog of the unprobative and insufficient kinds of proof identified in this article.

Two factual issues on which scientific evidence was crucial were: (1) whether plaintiffs received any exposure above background levels of PCBs that could be attributed to the Paoli railyard; and (2) whether PCBs are capable of causing the ailments of which plaintiffs complained or of which they believed they were at risk.²⁵⁵ The primary problem with plaintiffs' evidence on exposure involves unvalidated methodology through which plaintiffs attempted to show that their exposure to PCBs exceeded background levels.

Plaintiffs offered several forms of evidence in addition to the blood/adipose tissue calculations discussed in the preceding Part,²⁵⁶ in support of their contentions of higher than background levels of PCB exposure. First, they offered the testimony of Dr. Herbert Allen, an environmental chemist who used a formula of his own devising to calculate airborne exposure levels from PCB levels measured in neighborhood soils.²⁵⁷ Nothing in either the district court's or the appellate court's opinions, however, suggests that Dr. Allen's formula had been tested, that is, that its validity had been demonstrated by comparing the airborne concentrations predicted by the formula and actual, measured air concentrations.²⁵⁸ In fact, the district court opinion states that the "actual measurements that were taken showed an amount much lower than [Allen] calculated."²⁵⁹

Plaintiffs also offered the testimony of Dr. Deborah Barsotti, a toxicologist employed by the Agency for Toxic Substances and Disease Registry, who claimed to have correlated gas chromatography tracings of PCBs in the plaintiff's blood with tracings from soil samples from the Paoli railyard.²⁶⁰ Dr. Barsotti, however, was apparently unable to support her general statements with reference to any specific plaintiffs' blood samples or soil samples.²⁶¹ Later, she apparently conceded that the equipment she had used was not capable of yielding the results she claimed.²⁶²

Plaintiffs' evidence on general causation and on distinguishing background causes was hardly more probative. On the question of whether PCBs are capable of causing the kinds of illnesses complained of, the plaintiffs were faced with various studies that had failed to find a correlation between PCB exposure and significant human disease. One such study was the ATSDR's study, Toxicological Profile for Selected PCBs.²⁶³ As summarized in the Foreword, the ATSDR study found that only skin lesions and liver effects that were not associated with "clinically detectable disease" had been observed in PCB-exposed workers. The study also concluded that adverse effects had not been observed in persons with non-occupational exposures.²⁶⁴

The Third Circuit, however, found various plaintiffs' witnesses' testimony sufficient to create a jury question.²⁶⁵ Some of this testimony can only be described as conclusory.²⁶⁶ Several of plaintiffs' experts relied on animal studies and on studies involving incidents of accidental ingestion of a mixture of PCBs and PCDFs,²⁶⁷ the "Yusho" and "Yu Cheng" incidents that occurred respectively in Japan in 1968, and Taiwan in 1979.²⁶⁸ Lastly, the plaintiffs offered the testimony of Dr. William J. Nicholson, a professor of community medicine who had performed a "meta-analysis" of existing (negative) epidemiologic studies²⁶⁹ and concluded that his analysis showed that PCBs can cause liver, biliary tract and gall bladder disorders.²⁷⁰

The evidence on general and individual causation was clearly insufficient to permit the inference that would satisfy the more likely than not standard of proof. Animal studies, as discussed earlier, are at best subject to large uncertainties when extrapolated to humans, particularly for effects of chronic, low-level exposures.²⁷¹

The use of the Yusho and Yu Cheng studies were not challenged as invalid,²⁷² but their use for the purposes of proving that PCBs cause significant adverse effects in humans involves invalid reasoning.²⁷³ The studies identified harmful effects from two incidents involving the ingestion of oil mixtures containing PCBs and PCDFs. Assuming that the conclusions regarding the Yusho and Yu Cheng incidents were accurate, the results can at most be said to prove the following proposition: The harmful effects observed in the Yusho and Yu Cheng incidents were caused by either: (1) PCBs; (2) PCDFs; or (3) PCBs and PCDFs in combination.²⁷⁴ The studies do not allow the conclusion that PCBs alone can cause the effects observed in the study. Additionally, because PCDFs are regarded as more toxic than PCBs,²⁷⁵ the proposition that plaintiffs argued from the study is not supported by the study.²⁷⁶

The meta-analysis of epidemiologic studies presents a somewhat different problem. Meta-analyses are not outside the scope of recognized scientific methodology.²⁷⁷ The defendants did raise questions, however, about the way in which Dr. Nicholson analyzed the existing data, contending that he omitted data that were inconsistent with his conclusions.²⁷⁸ Thus, the trial court could have examined the bases on which Dr. Nicholson included and excluded data to determine whether there were logical criteria, systematically applied, in combining and evaluating the data from previous studies.

Poali also involved questions related to expert testimony on individual causation.²⁷⁹ Several witnesses appear to have asserted that based on test results indicating the presence of PCBs in the railyard and surrounding area, the presence of PCBs in plaintiffs' blood, plaintiffs' medical records, and the literature on effects of PCBs, they could state "to a reasonable medical certainty"²⁸⁰ that plaintiffs' various ailments were caused by exposure to PCBs.²⁸¹

Reference to the evidence on which these conclusions were purportedly based reveals that those conclusions amounted to nothing more than speculation. Animal studies and studies of incidents involving several chemicals provide only uncertain evidence that the substance will cause human disease at all. Animal studies simply do not produce results that permit reliable conclusions about the likelihood of human disease at particular exposure levels. The Yusho and Yu Cheng studies involved ingestion of much larger quantities of PCBs than the Paoli plaintiffs were exposed to, and that exposure involved another, probably more toxic chemical. In fact, Paoli, like other recent cases,²⁸² involves a scenario in which plaintiffs complaining of a variety of common ailments²⁸³ attempt to attribute those ailments to exposure to a toxic chemical for which human effects have not been demonstrated or are different from those complained of by the plaintiffs. Even if one assumes that some level of exposure above background has occurred, a proposition that appears doubtful in Paoli and other cases,²⁸⁴ the question remains as to whether the individual plaintiffs' conditions were caused by the exposure or by the more commonplace causes of such diseases in the general population, whether known or unknown, a question that cannot be answered without evidence demonstrating that a substance can cause the plaintiff's disease and indicating increased disease incidence at the levels to which plaintiffs were exposed.²⁸⁵ Animal testing and studies involving high level exposure to several toxic substances simply cannot provide that information.

The logical extension of the plaintiffs' position is that virtually anyone with one or more of a whole host of commonplace ailments who may have come into contact with toxic substances should be able to recover from the entity responsible for the toxic substance. Such a proposition clearly goes too far; yet it is difficult to draw any principled distinctions about who should or should not recover if the reasoning of the Paoli plaintiffs and their experts is accepted.²⁸⁶

Concerns about the evidentiary basis of causation in toxic torts are not limited to cases involving large numbers of plaintiffs and an array of alleged injuries. Cases involving one or a few plaintiffs may raise similar concerns. Further, a court may tend to view such cases in isolation, even though extension of the case's underlying logic may lead to results similar to those in cases such as Paoli, namely, that there is no principled way to distinguish persons whose disease was caused by a toxic substance exposure from those whose diseases were not so caused.

The New Jersey Supreme Court's decision in Rubanick v. Witco Chemical Co.,²⁸⁷ another case involving injuries claimed to have resulted from PCB exposure, is an example of that scenario. Plaintiffs, the relatives of two Witco employees who had died of colon cancer, claimed that their decedents' cancers were caused by PCB contamination at the site.²⁸⁸ Their expert on causation was Dr. Balis, a Ph.D. biochemist with extensive research experience on colon cancer.²⁸⁹ Dr. Balis cited studies indicating that PCBs cause cancer in animals, reports on the effects of PCBs on animals and humans, a high rate of cancer at Witco "during the relevant period," and the personal history of one of the deceased, indicating the absence of other risk factors.²⁹⁰ The difficulties of drawing inferences from animal studies and studies of the effects of PCBs in humans paralleled the difficulties in Paoli.²⁹¹ Further, the absence of other risk factors is of little probative value where the causes of most instances of a disease are unknown.²⁹²

The New Jersey court focused disapprovingly on the trial court's finding that Dr. Balis' theory of causation was not generally accepted in the scientific community.²⁹³ Citing the need for a more liberal standard for determining the reliability of scientific theories of causation in toxic torts cases,²⁹⁴ the court held that "a scientific theory of causation that has not yet reached general acceptance may be found to be sufficiently reliable if it is based on sound, adequately-founded scientific methodology involving data and information of the type reasonably relied on by experts in the scientific field."²⁹⁵ The court went on to state that the theory must be offered by an expert with "a demonstrated professional capability to assess the scientific significance of the underlying data and methodology, and to explain the bases for the opinion reached."²⁹⁶

Remanding the case for further proceedings on the admissibility of Dr. Balis' testimony, the court admonished the trial court not to scrutinize the expert's methodology itself to determine its soundness, however, but to refer to the opinions of comparable experts in the field.²⁹⁷ The problem with that recommendation is that it was unclear that the expert used any methodology at all. The trial court's opinion makes it clear that Dr. Balis testified in terms of possibilities, not probabilities.²⁹⁸ Moreover, the trial court had the benefit of the testimony of the defendant's witnesses, who opined that the scientific literature did not support the plaintiffs' expert's opinion that PCBs can cause cancer in humans.²⁹⁹ The court itself had read the articles cited by the plaintiff's expert and concluded that they "do not say what plaintiff's expert concludes."³⁰⁰

Rubanick, like Paoli, is a case in which an appellate court appears to approve of toxic tort causation testimony based on uncertain exposure levels, animal testing and other indicators of possible carcinogenicity, despite the absence of any evidence suggesting a connection between PCB exposure and the plaintiff's specific disease. The court cites the appropriate criteria, including reliability, but neither conducts nor allows the trial court to conduct a reliability analysis. Rather than accepting the trial court's findings, which were based on testimony of other experts, the appellate court interjects its own assessment. In reality, the New Jersey Supreme Court's rationale is based almost entirely on the qualifications of the expert, since it harks back to the witness's qualifications as a point of reference for determining the reliability of the data on which the expert relies as well as his methodology.³⁰¹

The problems of deferential review are not restricted to environmental exposure cases; they also occur in products liability cases where the costs of erroneous findings of liability in terms of withdrawal of useful products and disincentives to new product development are perhaps more apparent. Ferebee v. Chevron Chemical Co.,³⁰² perhaps the paradigm decision involving deferential review, involved injuries allegedly caused by a pesticide.³⁰³ Plaintiff's causation case was essentially based on the testimony of treating or examining physicians who claimed to have seen a few similar cases. It therefore illustrates the pitfalls of medical testimony based on the coexistence of exposure and disease.

The Bendectin cases are based on a more complex assemblage of evidence and testimony, consisting of chemical structure-activity analysis, in vitro testing, animal studies and purported reanalyses of existing epidemiologic studies.³⁰⁴ A review of one of the early cases decided in favor of plaintiffs, Oxendine v. Merrell Dow Pharmaceuticals, Inc.³⁰⁵ reveals much of the same evidence that plaintiffs have argued in other cases alleging birth defects caused by Bendectin. The chemical structure-activity analysis consisted of the observation that one of Bendectin's ingredients is an antihistamine and that some antihistamines are teratogenic.³⁰⁶ The in vitro and in vivo animal test results cited by plaintiffs' witness Dr. Done are subject to the same concerns for high rates of false positives that were discussed previously. The remaining evidence consisted primarily of Dr. Done's unpublished reanalysis of a previous epidemiologic study, which involved selective elimination of data.³⁰⁷

As will be discussed in more detail in Section VI.B.2 of this Article, reanalyses of epidemiologic studies are particularly susceptible to manipulation to achieve a preconceived result. The reanalyses offered by plaintiffs studies stand in contrast to a large body of epidemiologic evidence that has failed to confirm a statistically significant association between Bendectin and birth defects. Further, the fact that the studies offered by plaintiffs have been unpublished and therefore not subjected to peer review,³⁰⁸ lends further support to other courts' decisions to exclude them.³⁰⁹

The Bendectin litigation has also produced opinions that discerningly review scientific evidence; two such cases are Brock v. Merrell Dow Pharmaceuticals, Inc.³¹⁵ and Lynch v. Merrell-National Laboratories.³¹⁶ Bendectin has been the subject of over 2000 suits for birth defects allegedly caused by in utero exposure to the anti-nausea drug. Plaintiffs have sought recovery for a variety of malformations, but a number of the cases have involved limb reduction defects.³¹⁷ In Brock, the court based its reversal of a jury verdict for plaintiffs on the absence of any statistically significant epidemiologic evidence of an association between Bendectin and birth defects.³¹⁸ Both the Brock and the Lynch courts concluded that the plaintiffs' in vitro testing and animal studies evidence was insufficient, the Lynch court noting particularly the inability of in vivo and in vitro animal studies to prove causation in humans "in the absence of confirmatory epidemiologic data," which it contrasted with a number of studies that failed to find an association between Bendectin and birth defects.³¹⁹

In what has become one of the more controversial aspects of the Bendectin litigation, both courts rejected reanalyses of existing epidemiological studies that purported to show an association between Bendectin exposure and birth defects. The Lynch court noted the plaintiffs' failure to file any description of the expected testimony of Dr. Shanna Swan, whose reanalysis of epidemiologic data was offered by plaintiffs. The court went on to examine the basis of Swan's opinion from testimony in other litigation, observing that Swan's control group consisted of children with genetic birth defects, a group that had a lower than background risk for certain types of birth defects, raising the question whether genetic defects might make that control group less susceptible to non-genetic defects such as limb reduction. A lower susceptibility in the control group would skew the relative risk observed for the Bendectin-exposed group.³²⁰

In contrast, the Brock court's rationale focused on the fact that the elevated risk found in the reanalysis conducted by Dr. Jay Glasser lacked statistical significance.³²¹ As will be discussed in more detail in the following Section, reanalyses of epidemiological data are susceptible to advertent and inadvertent introduction of bias. Although the statistical significance point is arguable, the reanalyses were unpublished and therefore lacked the safeguards against biased or result-oriented data selection that peer-reviewed publication would have provided.

The Third Circuit and the New Jersey Supreme Court, who authored the Paoli and Rubanick decisions respectively, have demonstrated their understanding of complex scientific evidence. In DeLuca v. Merrell Dow Pharmaceuticals, Inc.,³²² the Third Circuit discussed the epidemiologic evidence on Bendectin and birth defects. At issue was the admissibility of a meta-analysis of existing epidemiologic studies. The meta-analysis in question³²³ did not meet the level of statistical significance³²⁴ typically required for epidemiologic studies.³²⁵ The court's view of the meta-analysis was overly generous,³²⁶ but the opinion demonstrates the court's understanding of the concepts of statistical significance and the effects of bias in epidemiologic studies.³²⁷ Similarly, when confronted with a causation issue on which epidemiologic evidence was offered, the New Jersey Supreme Court also evidenced a sophisticated understanding of such evidence. In Landrigan v. Celotex Corp.,³²⁸ the court discussed the proffered epidemiologic evidence and the causal inferences to be drawn from it, as well as the concept of attributable risk.³²⁹

No doubt there are times when expert witness testimony and scientific evidence are obscure. The details of statistical significance calculations could undoubtedly lose all but the most mathematically inclined and dedicated lay observer. But judges need not examine expert testimony and scientific evidence at that level of detail. Courts can and should, however, require the proponent of such evidence to demonstrate to the court that the evidence is valid and reliable, that is, that it makes sense and is sufficiently likely to produce an accurate result.

The fraction of cancers and other diseases that could be prevented by reducing or eliminating exposure to man-made toxic chemicals is still in dispute. Studies that have attempted to estimate the fraction of cancers caused by environmental pollution have placed the figure at about six percent, up to as much as fifteen percent.³³⁴ Other exposures and industrial products are thought to add an additional four to five percent, perhaps as much as ten percent.³³⁵

The debate about the role of synthetic chemicals in cancer causation has occurred against a backdrop of increasing cancer rates.³³⁶ The meaning of the data is unclear, however, because most if not all of the increase can be attributed to increases in smoking-related cancers and aging of the population.³³⁷ Even if the more pessimistic experts are correct in there conclusion that age adjusted rates are increasing for some cancers,³³⁸ the data do not support the proposition that most cancers are caused by toxic pollution or toxic products (other than cigarettes). Thus, there is no factual basis for a presumption that environmental pollutants cause most cancers.

The inability of epidemiologic studies to detect small increases in risks has been a major concern in the debate over toxic tort causation evidence.³⁴² The power of an epidemiologic study to identify a small increase in risk is a function of the size of the study groups and the background rate of disease, with larger study groups corresponding to greater statistical power.³⁴³ Meta-analysis, in which the data from a number of smaller studies are combined and reanalyzed, can enhance the likelihood of detecting an effect, if one exists.³⁴⁴ Meta-analysis can also provide the opportunity to refine the selection of data included in the analysis to address potential bias in sample selection, as can reanalysis of a single study.³⁴⁵

Systematic error, of which bias is one form, can be introduced into epidemiologic studies in a number of ways, including the failure to control for causal factors other than the factor under study and the failure to accurately delineate exposed and unexposed populations.³⁴⁶ One of the potential sources of bias in the Bendectin studies is recall bias, the possibility that mothers of children born with defects will be more likely to recall drug use during pregnancy than mothers of normal infants. Such recall bias will tend to result, in some kinds of studies, in an overestimation of the effect of the drug.³⁴⁷ Another concern with inaccurate recall is that the "unexposed" group will, in fact, have some individuals who were exposed and who exhibit effects caused by the exposure.³⁴⁸ If there is an effect, part of that effect will be attributed to the unexposed group, tending to diminish the magnitude of the observed effect.³⁴⁹

To counter the negative epidemiologic evidence that predominates in the published literature concerning Bendectin, several plaintiffs have variously offered meta-analyses or reanalyses by one or both of two expert witnesses, Dr. Alan Done, a professor of pediatrics and pharmacology, and Dr. Shanna Helen Swan, an epidemiologist and chief of the a unit of the California Department of Health Services.³⁵⁰ Meta-analyses are subject to questions about the propriety of combining data from studies in which the original criteria for selection of subjects and controls differed.³⁵¹ Both meta-analyses and reanalyses involve selection of data for inclusion and exclusion, which create the opportunity for "data dredging" that may turn up statistically significant correlations that are actually due to chance.³⁵² The methodology by which data were selected for inclusion and exclusion in meta-analyses and reanalyses should therefore be carefully scrutinized.

A number of objections can be made to the reanalyses and meta-analyses offered by various Bendectin plaintiffs. In the case of Dr. Done's reanalysis at issue in Lynch, the basis of the data selection seems less than clear,³⁵³ although the Oxendine opinion indicates that in the reanalysis offered by Done in that case, some pairs of exposed and unexposed children were eliminated because Done considered the risk of recall bias to be especially high among Canadian subjects who could have purchased Bendectin without a prescription.³⁵⁴ Dr. Shanna Swan's methodology is explained more completely in Lynch; it involved the reanalysis of data previously analyzed by four members of the Center for Disease Control.³⁵⁵ All of the subjects in the original group had involved abnormal children. Swan reanalyzed the data, using only children with genetic abnormalities as the control group so that the control group's abnormalities could not have resulted from Bendectin.³⁵⁶ Her reanalysis concluded that Bendectin is associated with an increased risk of birth defects. Swan's reanalysis raises questions because the control group for her reanalysis was acknowledged to have only a 0.57 relative rate (i.e., a 40% lower rate) for certain categories of birth defects. As the First Circuit noted, "Swan made no allowance for the possibility that the very fact of having such a severe genetic deficiency as Down's Syndrome might operate to make other rare deficiencies such as limb reduction less likely," thus skewing the apparent differences between the exposed and control groups.³⁵⁷ The possibility that both Done's and Swan's reanalyses were based on result oriented "data dredging" or other inadvertent introduction of bias cannot be ignored; further, none of the reanalyses or meta-analyses has been published in peer-reviewed scientific journals, although ample time has elapsed for review and publication.³⁵⁸ The failure of either to publish their results leaves courts without any reassurance that concerns about bias are unwarranted.

In any event, the insensitivity of epidemiologic studies in the case of Bendectin is probably an overrated concern. Although the studies cannot be said to eliminate all possibility that Bendectin is teratogenic, they at least indicate that if Bendectin is a teratogen, it is a weak one.³⁵⁹ Moreover, the insensitivity of epidemiologic studies does not improve the probative value of other evidence. Animal studies, mutagenicity testing and structure-activity relationships do not become more persuasive because of the absence of other kinds of proof.

The appropriate level of certainty is particularly an issue when epidemiologic evidence that does not meet epidemiologists' criteria for certainty is offered. Epidemiologists typically are unwilling to conclude that increased disease incidence in an exposed population is associated with a toxic substance exposure unless a statistical analysis of the data shows that the probability of a false positive is 5% or less.³⁶⁴ That requirement represents a 95% confidence level,³⁶⁵ a level that is considerably higher than the more probable than not standard would seem to suggest.³⁶⁶ Moreover, the 5% cutoff for statistical significance is arbitrary, and has been used to meet epidemiologists' perceived needs for certainty.³⁶⁷

Some commentators have questioned whether statistical significance is relevant to the more probable than not standard of proof.³⁶⁸ Green suggests that focus on the relative risk found in a study is more appropriate.³⁶⁹ That approach seems untenable, however, because it fails to distinguish the issue of whether exposure to the toxic substance causes any effect at all, which is the function of statistical significance testing, from the issue of the likelihood that a particular plaintiff's case resulted from the exposure rather than background or other causes, a conclusion that is inferred from the magnitude of the relative risk.

Relative risk greater than 1.0 in an exposed population is sufficient evidence of an association of disease with exposure only if we can be reasonably certain that the unequal distribution of disease in exposed and unexposed populations is not due to chance. Ignoring the possibility that an increased incidence of disease is due to chance leads to the obviously absurd result that a disease cluster, no matter how small, could be argued as sufficient evidence of an association between an exposure and disease, a result that is indefensible. The evaluation of the role of chance in an epidemiologic study is thus an essential part of determining the probative value of the evidence.

The appropriate confidence level is a more difficult question, however. At a minimum, the more probable than not standard of proof would seem to tolerate epidemiologic data on the issue of general causation if there is less than a 50% probability that the result is due to chance, a confidence level far lower than the 95% level typically employed by epidemiologists. Additionally, Green and others have noted that typical statistical significance testing is concerned only with the risk of false positives, that is, the risk that an effect will be inferred when there is actually no effect.³⁷⁰ The legal system is also concerned, however, with the risk of false negatives, namely, in toxic torts the risk that no effect will be detected when there actually is an effect.³⁷¹ Decreasing the risk of false positives tends to increase the risk of false negatives, though not in a straightforward way.³⁷² Thus, there is an argument that in some instances, epidemiological studies should be admitted with less stringent significance criteria than are typically applied. Before such a rule, which significantly lowers the standard of acceptability of epidemiologic evidence of increased risk, is adopted, however, it would be well to consider other sources of error. Epidemiologic studies are plagued to a greater or lesser degree with other, nonrandom sources of error. Exposure data can be highly uncertain. There is always the possibility that there are unknown confounding causes that are not randomly distributed between the exposed and control populations. Although statistical testing usually does not address nonrandom error, the possibility of other confounding factors may have a great deal to do with the high confidence levels that epidemiology has typically required to minimize the risk of error due to chance.

It may be instructive to consider Bendectin because it has been the subject of over thirty epidemiologic studies and at least one published meta-analysis of those studies.³⁷³ If statistical significance criteria are indeed too stringent, causing scientists to miss a real effect, one would expect to see relative risks from the various studies falling above 1.0 more often than below that number. In other words, the results should vary around the "true" relative risk even if no single study qualifies as statistically significant.³⁷⁴ In his comprehensive study of the Bendectin litigation, Sanders notes that of twenty-six studies from which he was able to extract a value indicative of relative risk, thirteen reported a value greater than one, twelve reported values less than one, and one study reported a value of exactly one.³⁷⁵ That result is roughly consistent with a published meta-analysis of seventeen prior studies that concluded that Bendectin is not associated with human birth defects.³⁷⁶ If statistical significance criteria were lowered to a 50% confidence level, one is left to wonder whether both plaintiffs and defendants would be offering "statistically significant" evidence, respectively, that Bendectin causes and prevents birth defects. Thus, it is not clear without further evaluation that scientific confidence level criteria are too stringent where epidemiologic evidence is concerned.

A more basic concern with courts' perceptions that scientists require too much certainty is that such views seem to form the basis for rejection of scientific reasoning altogether. The problem with Ferebee and its progeny is that they fail to recognize that in most cases,³⁷⁷ there are no alternative proofs available that amount to anything more than speculation or estimation with a great deal of uncertainty.³⁷⁸ Courts' unwillingness to scrutinize testimony on disease causation leaves the door open to the self-validating experts who can be found to testify to virtually any proposition.³⁷⁹ Even the courts that have deemed such evidence admissible have recognized the hazards of their approach.³⁸⁰ Nonetheless, they are willing to risk that kind of error because scientific evidence is unavailable to satisfy traditional standards of proof.³⁸¹ The irony of that rationale is that it rests on courts' and commentators' acceptance and even distortion of scientific speculation that widespread dissemination of new chemicals might result in increases in cancer, birth defects and other disease. Having accepted scientific speculation, they then reject the cautionary statements of scientists who want greater certainty before they reach conclusions.

Brennan's primary suggestion is to propose that questions involving significant scientific uncertainty be resolved by referring those questions to court-appointed experts or science panels.³⁸⁵ There are obviously cases, however, that are not significant enough to warrant science panels, or perhaps even court-appointed experts. Moreover, Brennan does not really come to grips with how evidence with such uncertain probative value can satisfy the more probable than not standard of proof, whether reviewed by a science panel or a lay jury. He recognizes that the acceptance of evidence associated with a high degree of uncertainty is a policy question, but does not provide a rationale for such a radical change in policy.³⁸⁶

Green, on the other hand, recognizes that difficulty. His solution is equally troubling: He states that "plaintiffs should be required to prove causation by a preponderance of the available evidence."³⁸⁷ This proposal is at least directly addresses the problem with animal studies and other, even more uncertain kinds of proof. The problem that Green's and Brennan's proposals present, however, is that they create potentially unlimited and ultimately arbitrary liability for cancer, birth defects, and other diseases that lack definitive causal explanations. Rare will be the cancer victim who cannot find some arguably toxic exposure, whether it be the pesticide application on the neighbor's lawn, pumping her own gas at the gas station or other such cause. Rarer still will be the plaintiff who cannot find a treating physician or other expert who is willing to state that based on past experience and review of the literature, that a particular toxic substance exposure is consistent with the plaintiff's disease and that the plaintiff lacked other predisposing factors. Reliance on the available evidence when such evidence suggests only the possibility, not the probability, of causation suggests that plaintiffs would do well to proceed to court when the evidence on whether a substance can cause disease is in an unformed stage. Such plaintiffs apparently will not have to contend with the messy questions of distinguishing background risk or other known risks that become issues when epidemiologic evidence is available. Indeed, they would have no basis for making such distinctions.

If there were a way to ease plaintiffs' evidentiary burdens without opening the door to arbitrary and potentially devastating liability for defendants, it would undoubtedly garner considerable support. The zone of uncertainty about the role of toxic chemicals in the causation of many diseases is simply too wide however, to suggest a reasonable way to split the difference.

It must be noted that courts' concerns are not all scientific. Other policy concerns, sometimes unspoken but often implied, seem to underlie courts' willingness to entertain unfounded and poorly reasoned evidence. Those concerns are the indignation and outrage felt by the public in general and plaintiffs in particular over exposure to contaminants or products involving substances suspected of causing harm or whose properties are simply unknown.³⁸⁸ In many of the environmental exposure cases, the exposures or the contamination that could have led to exposure occurred without the plaintiffs' knowledge or consent.³⁸⁹ In the case of potentially toxic products such as breast implants, the exposures have occurred with implicit or explicit assurances that the products were safe.

Traditional tort doctrines, however, do not provide for compensation for egregious conduct without causally related physical injury unless it rises to the level of intentional infliction of emotional distress.³⁹⁰ Commentators have suggested creation of causes of action based on creation of risk,³⁹¹ and a limited number of courts have adopted such theories.³⁹² Those theories are implicitly and sometimes explicitly premised on assumptions that some significant level of risk can be proved,³⁹³ assumptions that in many cases would be erroneous.³⁹⁴

In any event, the tort system is probably not the best forum for addressing public concerns over uncertain risk. The inability of toxic tort claimants to prove causation has been one of the more important rationales for environmental regulation.³⁹⁵ Indeed regulation is an area in which risk is explicitly recognized as a basis for restricting the dissemination of a substance in products or in the environment. Regulation does not compensate those who are injured despite regulation or by unregulated risks, but it has an important role to play in minimizing risks.

However desirable it might be to have the tort system fill all the gaps where toxic injury occurs, the current state of knowledge simply does not permit the necessary causal connections to be made. Given that state of affairs, what is at stake is whether the "more probable than not" standard of proof will continue to apply to toxic torts. Whether that burden should be lessened or even shifted to defendants are policy issues of the greatest importance. They should be addressed directly and changes, if any, should be based on their fullest consideration of the implications. To effect a reallocation of burdens of proof under the pretext of admitting reliable evidence which is in fact not probative, is not the appropriate way to bring about a change in such a fundamental principle of tort law.