You put your right foot in
You put your right foot out
You put your right foot in
And you shake it all about
You do the Hokey Pokey and you turn yourself around
That’s what it’s all about!
Ever since the United States Supreme Court decided Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), legal scholars, judges, and lawyers have struggled with the structure and validity of expert opinion on specific causation. Professor David Faigman and others have attempted to articulate the scientific basis (if any) for opinion testimony in health-effects litigation that a give person’s disease has been caused by an exposure or condition.
In 2015, as part of a tribute to the late Judge Jack Weinstein, Professor Faigman offered the remarkable suggestion that in advancing differential etiologies, expert witnesses were inventing wholesale an approach that had no foundation or acceptance in their scientific disciplines:
“Differential etiology is ostensibly a scientific methodology, but one not developed by, or even recognized by, physicians or scientists. As described, it is entirely logical, but has no scientific methods or principles underlying it. It is a legal invention and, as such, has analytical heft, but it is entirely bereft of empirical grounding. Courts and commentators have so far merely described the logic of differential etiology; they have yet to define what that methodology is.”[1]
Faigman is correct that courts often have left unarticulated exactly what the methodology is, but he does not quite make sense when he writes that the method of differential etiology is “entirely logical,” but has no “scientific methods or principles underlying it.” After all, Faigman starts off his essay with a quotation from Thomas Huxley that “science is nothing but trained and organized common sense.”[2] As I have written elsewhere, the form of reasoning involved in differential diagnosis is nothing other than iterative disjunctive syllogism.[3] Either-or reasoning occurs throughout the physical and biological sciences; it is not clear why Faigman declares it un- or extra-scientific.
The strength of Faigman’s claim about the made-up nature of differential etiology appears to be undermined and contradicted by an example that he provides from clinical allergy and immunology:
“Allergists, for example, attempt to identify the etiology of allergic reactions in order to treat them (or to advise the patient to avoid what caused them), though it might still be possible to treat the allergic reactions without knowing their etiology.”
Faigman at 437. Of course, not only allergists try to determine the cause of an individual patient’s disease. Psychiatrists, in the psychoanalytic tradition, certain do so as well. Physicians who use predictive regression models use group data, in multivariate analyses, to predict outcomes, risk, and mortality in individual patients. Faigman’s claim is similarly undermined by the existence of a few diseases (other than infectious diseases) that are defined by the causative exposure. Silicosis and manganism have played a large role in often bogus litigation, but they represent instances in which a differential diagnosis and puzzle may also be an etiological diagnosis and a puzzle. Of course, to the extent that a disease is defined in terms of causative exposures, there may be serious and even intractable problems caused by the lack of specificity and accuracy in the diagnostic criteria for the supposedly pathognomonic disease.
As I noted at the time of Faigman’s 2015 essay, his suggestion that the concept of “differential etiology” was not used in the sciences themselves, was demonstrably flawed and historically inaccurate.[4]
A year earlier, in a more sustained analysis of specific causation, Professor Faigman went astray in a different direction, this time by stating that:
“it is not customary in the ordinary practice of sociology, epidemiology, anthropology, and related fields (for example, cognitive and social psychology) for professionals to make individual diagnostic judgments derived from group-based data.”[5]
Faigman’s invocation of “ordinary practice” of epidemiology was seriously wide of the mark. Medical practitioners and scientists frequently use epidemiologic data, based upon “group-based data” to make individual diagnostic judgments. The inferences from group data to individual range abound in the diagnostic process itself, where the specificity and sensitivity of disease signs and symptoms are measured by group data. Physicians must rely upon group data to make prognoses for individual patients, and they rely upon group data to predict future disease risks for individual patients. Future disease risks, as in the Framingham risk score for hard coronary heart disease, or the Gale model for breast cancer risk, are, of course, based upon “group-based data.” Medical decisions to intervene, surgically, pharmacologically, or by some other method, all involve applying group data to the individual patient.
Faigman’s 2014 law review article was certainly correct, however, in noting that specific causation inferences and conclusions were often left “profoundly underdefined,” with glib identifications of risk with cause.[6] There was thus plenty of room for further elucidation of specific causation decisions, and I welcome Faigman’s most recent effort to nail conceptual jello to the wall, in a law review article that was published last year.[7]
This new article, “Differential Etiology: Inferring Specific Causation in the Law from Group Data in Science,” is the collaborative product of Professor Faigman and three other academics. Joseph Sanders will be immediately recognizable to the legal community as someone who long pondered causation issues, both general and specific, and who has contributed greatly to the law review literature on causation of health outcomes. In addition to the law professors, Peter B. Imrey, a professor of medicine at the Cleveland Clinic Lerner College of Medicine of Case Western Reserve University, and Philip Dawid, an emeritus professor of statistics in Cambridge University, have joined the effort to make sense of specific causation in the law. The addition of medical and statistical expertise has added greatly to Faigman’s previous efforts, and it has corrected some of his earlier errors and added much nuance to the discussion. The resulting law review article is well worth reading for practitioners. In this post, however, I have not detailed every important insight, but rather I have tried to point out some of the continuing and new errors in the analysis.
The Sanders-Faigman-Imbrey-Dawid analysis begins with a lament that:
“there is no body of science to which experts can turn when addressing this issue. Ultimately, much of the evidence that can be brought to bear on this causal question is the same group-level data employed to prove general causation. Consequently, the expert testimony often feels jerry-rigged, an improvisation designed to get through a tough patch.”[8]
As an assessment of the judicial decisions on specific causation, there can be no dissent or appeal from the judgment of these authors. The authors use of the term “jerry-rigged” is curious. I had first I thought they were straining to avoid using the common phrase “jury rigged” or to avoid inventing a neologism such as “judge rigged.” The American Heritage and Merriam Webster dictionaries, however, describe the phrase “jerry-rigged” as a conflation of “jury-rigged,” a nautical term for a temporary but adequate repair, with “jerry-rigged,” a war-time pejorative term for makeshift devices put together by Germans. So jerry-rigged it is, and the authors are off and running to try to describe, clarify, and justify the process of drawing specific causation inferences by differential etiology. They might have called what passes for judicial decision making in this area as the “hokey pokey.”
The authors begin their analysis of specific causation with a brief acknowledgement that our legal system could abandon any effort to set standards or require rigorous thinking on the matter by simply leaving the matter to the jury.[9] After all, this laissez-faire approach had been the rule of law for centuries. Nevertheless, despite occasional retrograde, recidivist judicial opinions,[10] the authors realize that the law has evolved to a point that some judicial control over specific causation opinions is required. And if judges are going to engage in gatekeeping of specific-causation opinions, they need to explain and justify their decisions in a coherent and cogent fashion.
Having thus dispatched legal nihilism, the authors turn their attention to what they boldly describe as “the first full-scale effort to bring scientific sensibilities – and rigorous statistical thinking – to the legally imperative concept of specific causation.”[11] The claim is remarkable claim given that tort law has been dealing with the issue for decades, but probably correct given how frequently judges have swept the issue under a judicial rug of inpenetrable verbiage and shaggy thinking. The authors also walk back some of Faigman’s earlier claims that there is no science in the assessment of specific causation, although they acknowledge the obvious, that policy issues sometimes play a role in deciding both general and specific causation decisions. The authors also offer the insight, for which they claim novelty, that some of the Bradford Hill guidelines, although stated as part of assessing general causation, have some relevancy to decisions concerning specific causation.[12] Their insight is indeed important, although hardly novel.
Drawing upon some of the clearer judicial decisions, the authors identify three necessary steps to reach a conclusion of specific causation:
“(a) making a proper diagnosis;
(b) supporting (“ruling in”) the plausibility of the alleged cause of the injury on the basis of general evidence and logic; and
(c) particularization, i.e., excluding (‘ruling out’) competing causes in the specific instance under consideration.”[13]
Although this article is ostensibly about specific causation, the authors do not reach a serious discussion of the matter until roughly the 42nd page of a 72 page article. Having described a three-step approach, the authors feel compelled to discuss step one (describing or defining the “diagnosis,” or the outcome of interest), and step two, the “ruling in” process that requires an assessment of general causation.
Although ascertaining general causation is not the focus of this article, the authors give an extensive discourse on it. Indeed, the authors have some useful things to say about steps one and two, and I commend the article to readers for some of its learning. As much as the lawsuit industry might wish to do away with the general causation step, it is not going anywhere soon.[14] The authors also manage to say some things that range from wrong to not even wrong. One example of professoriate wish casting is the following assertion:
“Other things being equal, when the evidence for general causation is strong, and especially when the strength of the exposure–disease relationship as demonstrated in a body of research is substantial, the plaintiff faces a lower threshold in establishing the substance as the cause in a particular case than when the relationship is weaker.”[15]
This assertion appears, sans citation or analysis. The generalization fails in the face of counterexamples. The causal role for estrogen in many breast cancers is extremely strong. The International Agency for Cancer Research classifies estrogen as a Category I, known human carcinogen for breast cancer, even though estrogen is made naturally in the human female, and male, body. In the Women’s Health Initiative clinical trial, researchers reported a hazard ratio of 1.2,[16] but plaintiffs struggled to prevail on specific causation in litigation involving claims of breast cancer caused by post-menopausal hormone therapy. Perhaps the authors meant, by strength of exposure relationship, a high relative risk as well, but that point is taken up when the authors address the “ruling in” step of the three-step approach. In any event, the strength of the case for general causation is quite independent of the specific causation inference, especially in the face of small effect sizes.
On general causation itself, the authors begin their discussion with “threats to validity,” a topic that they characterize as mostly implicit in the Bradford Hill guidelines. But their suggestion that validity is merely implicit in the guidelines is belied by their citation to Dr. Woodside’s helpful article on the “forgotten predicate” to the nine Bradford Hill guidelines.[17] Bradford Hill explicitly noted that the starting point for considering an association to be causal occurred when “[o]ur observations reveal an association between two variables, perfectly clear-cut and beyond what we would care to attribute to the play of chance.”[18] Sir Austin told us in no uncertain terms that there is no need to consider the nine guidelines until random and systematic error have been rejected.[19]
In this article’s discussion of general causation, Professor’s Dawid’s influence can be seen in the unusual care to describe and define the p-value.[20] But the discussion devolves into more wish casting, when the authors state that p-values are not the only way to assess random error in research results.
They double down by stating that “[m]any prominent statisticians and other scientists have questioned it, and the need for change is increasingly accepted.”[21] The source for their statement, the American Statistical Association (ASA) 2016 p-value Statement, did not questioned the utility of the p-value for assessing random error, and this law review provides no other support for other unidentified methods to assess random error. For the most part, the ASA Statement identified misuses and misstatements of p-values, with the caveat that “[s]cientific conclusions and business or policy decisions should not be based only on whether a p-value passes a specific threshold.” This is hardly questioning the importance or utility of p-values in assessing random error.
When one of the cited authors, Ronald Wasserstein, published an editorial in 2019, proclaiming that it was time to move past the p-value, the then president of the ASA, Professor Karen Kafadar, commissioned a task force on the matter. That task force, consisting of many of the world’s leading statisticians, issued a short, but pointed rejection of Wasserstein’s advocacy, and by implication, the position asserted in this law review.[22] Several of the leading biomedical journals that were lobbied by Wasserstein to abandon statistical significance testing reassessed their statistical guidelines and reaffirmed the use of p-values and tests.[23]
Similarly, this law review’s statements that alternatives to frequentist tests (p-values) such as Bayesian inference are “ascendant” have no supporting citations, and generally are an inaccurate assessment of what most biomedical journals are currently publishing.
Despite the care with which this law review article has defined p-values, the authors run off the road when defining a confidence interval:
“A 95% confidence interval … is a one-sided or two-sided interval from a data sample with 95% probability of bounding a fixed, unknown parameter, for which no nondegenerate probability distribution is conceived, under specified assumptions about the data distribution.”[24]
The emphasis added is to point out that the authors assigned a single confidence interval with the property of bounding the true parameter with 95% probability. That property, however, belongs to the infinite set of confidence intervals based upon repeated sampling of the same size from the same population, and constant variance. There is no probability statement to be made for the true parameter, as either in or not in a given confidence interval.
In an issue that is relevant to general and specific causation, the authors offer some ipse dixit on the issue of “thresholds”:
“with respect to some substance/injury relationships, it is thought that there is no safe threshold. Cancer is the injury for which it is most frequently thought that there is no safe threshold, but even here the mechanism of injury may lead to a different conclusion.”[25]
Here as elsewhere, the authors are repeating dogma, not science, and they ignore the substantial body of scientific evidence that undermines the so-called linear no threshold dose-response curve. The only citation offered is a judicial citation to a case that rejected the no threshold position![26]
So much for “ruling in.” In the next post, I will turn my attention to this law review’s handling of the “ruling out” step of differential etiology.
[1] David L. Faigman & Claire Lesikar, “Organized Common Sense: Some Lessons from Judge Jack Weinstein’s Uncommonly Sensible Approach to Expert Evidence,” 64 DePaul L. Rev. 421, 444 (2015).
[2] Thomas H. Huxley, “On the Education Value of the Natural History Sciences” (1854), in Lay Sermons, Addresses and Reviews 77 (1915).
[3] See, e.g., “Differential Etiology and Other Courtroom Magic” (June 23, 2014) (collecting cases); “Differential Diagnosis in Milward v. Acuity Specialty Products Group” (Sept. 26, 2013).
[4] See David Faigman’s Critique of G2i Inferences at Weinstein Symposium (Sept. 11, 2015); Kløve & D. Doehring, “MMPI in epileptic groups with differential etiology,” 18 J. Clin. Psychol. 149 (1962); Kløve & C. Matthews, “Psychometric and adaptive abilities in epilepsy with differential etiology,” 7 Epilepsia 330 (1966); Teuber & K. Usadel, “Immunosuppression in juvenile diabetes mellitus? Critical viewpoint on the treatment with cyclosporin A with consideration of the differential etiology,” 103 Fortschr. Med. 707 (1985); G.May & W. May, “Detection of serum IgA antibodies to varicella zoster virus (VZV)–differential etiology of peripheral facial paralysis. A case report,” 74 Laryngorhinootologie 553 (1995); Alan Roberts, “Psychiatric Comorbidity in White and African-American Illicity Substance Abusers” Evidence for Differential Etiology,” 20 Clinical Psych. Rev. 667 (2000); Mark E. Mullinsa, Michael H. Leva, Dawid Schellingerhout, Gilberto Gonzalez, and Pamela W. Schaefera, “Intracranial Hemorrhage Complicating Acute Stroke: How Common Is Hemorrhagic Stroke on Initial Head CT Scan and How Often Is Initial Clinical Diagnosis of Acute Stroke Eventually Confirmed?” 26 Am. J. Neuroradiology 2207 (2005);Qiang Fua, et al., “Differential Etiology of Posttraumatic Stress Disorder with Conduct Disorder and Major Depression in Male Veterans,” 62 Biological Psychiatry 1088 (2007); Jesse L. Hawke, et al., “Etiology of reading difficulties as a function of gender and severity,” 20 Reading and Writing 13 (2007); Mastrangelo, “A rare occupation causing mesothelioma: mechanisms and differential etiology,” 105 Med. Lav. 337 (2014).
[5] David L. Faigman, John Monahan & Christopher Slobogin, “Group to Individual (G2i) Inference in Scientific Expert Testimony,” 81 Univ. Chi. L. Rev. 417, 465 (2014).
[6] Id. at 448.
[7] Joseph Sanders, David L. Faigman, Peter B. Imrey, and Philip Dawid, “Differential Etiology: Inferring Specific Causation in the Law from Group Data in Science,” 63 Ariz. L. Rev. 851 (2021) [Differential Etiology]. I am indebted to Kirk Hartley for calling this new publication to my attention.
[8] Id. at 851, 855.
[9] Id. at 855 & n. 8 (citing A. Philip Dawid, David L. Faigman & Stephen E. Fienberg, “Fitting Science into Legal Contexts: Assessing Effects of Causes or Causes of Effects?,” 43 Sociological Methods & Research 359, 363–64 (2014). See also Barbara Pfeffer Billauer, “The Causal Conundrum: Examining the Medical-Legal Disconnect in Toxic Tort Cases from a Cultural Perspective or How the Law Swallowed the Epidemiologist and Grew Long Legs and a Tail,” 51 Creighton L. Rev. 319 (2018) (arguing for a standard-less approach that allows clinicians to offer their ipse dixit opinions on specific causation).
[10] Differential Etiology at 915 & n.231, 919 & n.244 (citing In re Round-Up Prods. Liab. Litig., 358 F. Supp. 3d 956, 960 (N.D. Cal. 2019).
[11] Differential Etiology at 856 (emphasis added).
[12] Differential Etiology at 857.
[13] Differential Etiology at 857 & n.14 (citing Best v. Lowe’s Home Ctrs., Inc., 563 F.3d 171, 180 (6th Cir. 2009)).
[14] See Margaret Berger, “Eliminating General Causation: Notes Toward a New Theory of Justice and Toxic Torts,” 97 Colum L. Rev. 2117 (1997).
[15] Differential Etiology at 864.
[16] Jacques E. Rossouw, et al., “Risks and benefits of estrogen plus progestin in healthy postmenopausal women: Principal results from the Women’s Health Initiative randomized controlled trial,” 288 J. Am. Med. Ass’n 321 (2002).
[17] Differential Etiology at 884 & n.104, citing Frank Woodside & Allison Davis, “The Bradford Hill Criteria: The Forgotten Predicate,” 35 Thomas Jefferson L. Rev. 103 (2013).
[18] Austin Bradford Hill, “The Environment and Disease: Association or Causation?” 58 Proc. Royal Soc’y Med. 295, 295 (1965).
[19] Differential Etiology at 865.
[20] Differential Etiology at 869.
[21] Differential Etiology at 872, citing Ronald L. Wasserstein and Nicole A. Lazar, “The ASA Statement on p-Values: Context, Process, and Purpose,” 72 Am. Statistician 129 (2016).
[22] Yoav Benjamini, Richard D. De Veaux, Bradley Efron, Scott Evans, Mark Glickman, Barry I. Graubard, Xuming He, Xiao-Li Meng, Nancy M. Reid, Stephen M. Stigler, Stephen B. Vardeman, Christopher K. Wikle, Tommy Wright, Linda J. Young, and Karen Kafadar, “ASA President’s Task Force Statement on Statistical Significance and Replicability,” 15 Ann. Applied Statistics 1084 (2021), 34 Chance 10 (2021).
[23] See “Statistical Significance at the New England Journal of Medicine” (July 19, 2019); See also Deborah G. Mayo, “The NEJM Issues New Guidelines on Statistical Reporting: Is the ASA P-Value Project Backfiring?” Error Statistics Philosophy (July 19, 2019).
[24] Differential Etiology at 898 n.173 (emphasis added).
[25] Differential Etiology at 890.
[26] Differential Etiology at n.134, citing Chlorine Chemistry Council v. Envt’l Protection Agency, 206 F.3d 1286 (D.C. Cir. 2000), which rejected the agency’s assumption that the carcinogenic effects of chloroform in drinking water lacked a threshold.
