Dayton Law Review


The Scientific Basis of Causality in Toxic Tort Cases

Appendix: The Logical Structure of Scientific Studies Relevant to Toxic Tort Cases

Consider an analysis of the effect of breathing asbestos on the development of cancer in rats. A population of identical rats is randomly divided into two groups which are housed and fed in exactly the same manner, except one of the groups receives a predetermined amount of asbestos. The dose of asbestos constitutes the “x” in the putative causal statement “x caused y,&lrdquo; and the percentage of animals that develop cancer is the “y.” Since asbestos is not necessary for cancer to occur, some members of both groups are expected to develop the disease. The point of the experiment is to determine whether the cancer rate in the exposed group is higher than in the control group.

Suppose that after an observation period of two years, twenty percent of the asbestos-breathing animals and ten percent of the control rats developed cancer. It might seem proper to conclude that asbestos caused an increase in the incidence of cancer, but such a conclusion could not be justified (based on the hypothetical facts thus far given) over the conclusion that, despite the difference in percentages, reliable evidence of a causal relationship was not adduced. Why? Because it is possible that the dose of asbestos and the increased cancer rate were simultaneous but wholly independent events-like the relation between the length of women’s skirts and the stock-market average. If the study were repeated, it is unlikely that the precise percentages observed in the first study would be found in the second study because it is a biological fact that groups of animals are inherently different despite all efforts to make them identical. Consequently, successive studies using different but nominally identical groups of animals are expected to yield a range of values. It is possible the group of rats used to form the control group naturally fell in the low part of the range, whereas the animals assigned to the asbestos group fell at the high end of the range. If so, natural biological variation accounted for the observed differences in percentages.

Statistical methods have been developed that permit an objective answer to the question whether the asbestos caused the difference in cancer rates. Some methods are preferable to others, depending on considerations such as what was measured, how it was measured, and how much the data varied from animal to animal. The feature common to all the statistical methods is that they permit an objective answer to the question whether the apparent difference between an experimental and control group is real, and that answer can be expressed mathematically in the form of a probability.

An impossible event has a probability of zero, whereas an event that is certain is assigned a probability of one; all other possible outcomes have probabilities between these two extremes. The general convention in science (the standard used by most practicing experimental scientists and broadly enforced by journal editors and peer review panels of public and private granting and contracting agencies) is that if the calculated probability that the results are different is greater than .95, or ninety-five percent, they are then accepted as actually being different, absent a supervening consideration. Thus, if the calculated likelihood that the observed cancer rates in the asbestos study were different was greater than ninety-five percent, it would follow that asbestos caused the cancer as that concept is used in science. As a scientific statement, the phrase caused the cancer means that, to a certainty of greater than ninety-five, the observed difference in cancer rates would not have occurred but for the presence of the asbestos.133

The possibility of supervening factors in a biological experiment is termed bias, and may manifest itself as follows. Assume that the results of the statistical analysis of the asbestos study justified a statement of the form “x caused y,” where “x” is the concentration of asbestos and “y” is the observed difference in cancer rates. The experimental and control groups were formed at the beginning of the experiment by randomly assigning animals to the respective groups. The purpose of the randomization was to insure that all factors actually or potentially pertinent to the development of cancer would be, on average, identical between the two groups and hence not a possible explanation for a subsequently observed intergroup difference. If this assumption is violated, the logical structure of the experiment is complicated because the investigator must now justify “x caused y” when “z caused y” (where “z” is the biasing factor) may be true, based on the experimental procedures followed. In general, if the groups differed with regard to any factor whatsoever, that difference could, in principle, serve as a supervening factor that destroys the logical structure of the inference “x caused y.”134

In performing scientific studies, care is taken to avoid bias and thereby foster the situation drawing the inference of causation, the possible existence of which is actually the point of the study. But some bias exists in every experiment, and consequently it is always a matter of judgment whether that bias was causally related to the result observed. The biasing factors may be regarded as minor; for example, the asbestos rats were located on a different shelf than the control rats, and exposed to slightly different light levels. Alternatively, potential bias may be of greater concern; if, the asbestos rats were located closer to the door of the animal care facility where the average temperature was lower than the location of the control rats. If a strong biasing factor were present—for example, eighty percent of the asbestos rats but only ten percent of the control rats were males—the groups would be essentially non-comparable and the results of statistical testing would therefore be useless.

A criticism based on bias can be asserted against any controlled observation,135 and evaluation of the potential role of the bias involves the exercise of judgment within the orthodox framework of the particular branch of science. The judgment regarding possible supervening bias is made first by the investigator conducting the study; thereafter, the question is further evaluated by other scientists, journal editors, peer review panels at government granting agencies such as the National Institutes of Health, and by other mechanisms that may be created by governmental or industrial entities that paid for the research.

Successful scientists develop powers of judgment regarding whether the costs involved in controlling specific factors are warranted in view of the aims of the particular study. For this reason, and because an investigator can exert great control over both environmental factors and experimental subjects, the question of bias in a laboratory study involving animals or human subjects is relatively rare. However, there are biological studies, called epidemiological studies, in which bias is common because the experimental and control groups almost certainly differ with regard to characteristics other than those chosen for study.

In an epidemiological study,136 the investigator does not exert control over the experimental subjects. Instead, the subjects either intentionally or inadvertently apply the potentially toxic agent to themselves. The amount of exposure the subjects receive is usually assessed indirectly based on job categories or place of residence because it is impractical or impossible to actually measure exposure levels for each subject.

Epidemiological studies have the advantage of providing scientific data about the effects of toxic agents on human beings, as opposed to laboratory animals. Consequently, the information obtained from epidemiological studies is directly relevant to toxic tort cases because there is no need for extrapolation. On the other hand, the omnipresence of bias in epidemiological studies complicates the interpretation of epidemiological studies, and renders it improper to use data from a single study to justify a causal assertion.137 For these reasons, the statistical link between “x” and “y” in an epidemiological study is described in terms of the euphemism “associated,” as in “x is associated with y.” Only if there exist multiple independent epidemiological studies involving the same or similar “x’s” in which similar or consistent “y’s” were observed would it be reasonable to infer the existence of a causal relation.138

 

HOME | Contact Marino |