McGill Law Journal Revue de droit de McGill
EPIDEMIOLOGICAL UNCERTAINTY, CAUSATION, AND
DRUG PRODUCT LIABILITY
Richard Goldberg*
Epidemiological evidence is regularly presented to
courts in determining proof of causation in medicinal
product liability litigation. Building on the foundations of
the authors previous monograph, which supported the
use of epidemiological evidence in dealing with problems
of proof of causation in alleged cases of adverse drug reac-
tions, this paper revisits this perennial problem of the role
of epidemiological evidence in assessing causation in
product liability cases in a twenty-first century context,
examining recent cases in the United Kingdom, United
States, Australia, and Canada. It seeks to determine the
extent to which the courts in the highlighted cases have
been pragmatic and fair in their interpretation and utili-
zation of epidemiological evidence, from the perspective of
both consumers and pharmaceutical manufacturers. The
paper examines the apparent tension between the levels
of proof required in law and science, including the rela-
tionship between levels of statistical significance and the
claimants burden of proof; and it assesses the wisdom of
using a doubling of the risk rule as a threshold to any re-
covery. It explores the ways in which probabilistic meth-
ods, including statistical refining with individual risk fac-
tors, can be used in conjunction with epidemiological evi-
dence to determine specific causation. The paper supports
the view that logistic regression techniques and other
forms of statistical refining mechanisms using specific
risk factors can and do help in the process of giving quan-
titative or quasi-quantitative expression to conclusions
about the cause of disease in an individual drug product
liability claim that is based on epidemiological evidence.
La preuve pidmiologique est rgulirement pr-
sente pour dmontrer la causalit dans les litiges en ma-
tire de responsabilit pour les produits thrapeutiques.
En se fondant sur lune des monographies prc-
dentes de lauteur, qui appuie le recours la preuve pi-
dmiologique pour traiter des difficults prouver la cau-
salit dans des affaires portant sur les effets indsirables
de mdicaments, cet article rexamine, dans le contexte
du XXIe sicle, le problme constant du rle de la preuve
pidmiologique dans les questions de causalit en ma-
tire de responsabilit du fabricant. Il analyse cette fin
la jurisprudence du Royaume-Uni, des tats-Unis, de
lAustralie et du Canada. Cet article cherche cerner
jusqu quel point les tribunaux ont adopt, dans les af-
faires tudies, une approche pragmatique et juste en in-
terprtant et en utilisant la preuve pidmiologique, tant
du point de vue des consommateurs que de celui des fabri-
cants de produits pharmaceutiques. Il examine la tension
entre les niveaux de preuve ncessaire dans les domaines
scientifique et juridique, y compris la relation entre les ni-
veaux de signification statistique et le fardeau de la
preuve du demandeur, et value la possibilit dutiliser la
rgle de doublement du risque comme seuil pour un re-
couvrement judiciaire. Il explore les faons dont les m-
thodes probabilistes, telles que laffinage des statistiques
avec des facteurs de risque, peuvent tre utilises en con-
jonction avec la preuve pidmiologique afin de dtermi-
ner la causalit spcifique. Cet article avance que les
techniques de rgression logistique, ainsi que dautres m-
canismes de raffinement statistique se servant de facteurs
de risque spcifiques, peuvent aider donner une expres-
sion quantitative ou quasi-quantitative aux conclusions
portant sur la causalit dans une rclamation en respon-
sabilit pour un produit thrapeutique bas sur la preuve
pidmiologique.
* Professor of Law, Durham University, UK. Earlier drafts of this paper were delivered to
the Department of Epidemiology, Biostatistics and Occupational Health, Faculty of
Medicine, McGill University, Montreal, April 2011 under a Carnegie Research Grant,
and to the first meeting of the Technological Innovations, Uncertainty, and Responsibil-
ity network, Faculty of Law, McGill University, 5 July 2012. The author wishes to
thank Professors Philip Dawid and David Goldberg for helpful discussions.
Citation: (2014) 59:4 McGill LJ 777 Rfrence : (2014) 59 : 4 RD McGill 777
Richard Goldberg 2014
778 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
Introduction
I.
Reconciling the Standards of Proof in Law and
Science in the United Kingdom
A. Evidence of Causation for Purposes of Science
and for Purposes of Law
B. Doubling of Risk Theory
C. Association Versus Causation
D. Teaching Courts Epidemiology
E. The Statistical Chance/Personal Chance Dichotomy
F. Overcoming the Statistical Chance/Personal
Chance Dichotomy: Statistical Refining Mechanisms
Using Specific Risk Factors
Conclusions
779
784
784
785
793
799
802
806
815
EPIDEMIOLOGICAL UNCERTAINTY 779
Introduction
Proof of causation in toxic tort litigation is an inherently difficult prob-
lem, which regularly requires time-consuming analysis of complex scien-
tific evidence.1 The difficulties in proving both general causation (whether
a product was capable of causing the damage alleged) and specific causa-
tion (whether the product did so in the individual case) are magnified in
the context of medicinal products.2 As Harvey Teff and Colin Munro have
highlighted:
Drugs are always potentially dangerous due to their toxicity. They
are often taken by people who are already ill and who may be unu-
sually susceptible to further ailments. Unlike many other products,
they may cause injury in unpredictable ways, depending on the in-
dividual users constitution. They may not be taken according to the
instructions. The user may be allergic to a particular drug. Alterna-
tively, what appears to be an allergy may in fact be a toxic reaction.3
With a multitude of new kinds of drugs emerging as a harvest of the
scientific and technological revolutions of both the twentieth and early
twenty-first centuries, the cases have become even more complex, de-
manding much from lawyers and scientific experts on both sides and from
judges themselves.4
1 See CJ Miller & RS Goldberg, Product Liability, 2d ed (Oxford: Oxford University
Press, 2004) at para 17.05.
2 Richard Goldberg, Causation and Risk in the Law of Torts: Scientific Evidence and Me-
dicinal Product Liability (Oxford: Hart Publishing, 1999) 5 et seq [Goldberg, Causation
and Risk in the Law of Torts].
3 Harvey Teff and Colin Munro, Thalidomide: The Legal Aftermath (Farnborough: Saxon
House, 1976) at 13536. Clearly it is often harder to prove that ones injuries are due to
an adverse drug reaction than that they have been caused by a faulty machine (Harvey
Teff, Regulation Under the Medicines Act 1968: A Continuing Prescription for Health
(1984) 47:3 Mod L Rev 303 at 322). Professor Teff also notes [t]he synergistic effects of
certain combinations (for example, barbiturates and alcohol, anti-histamines and
cheese) may prove fatal (Harvey Teff, Products Liability in the Pharmaceutical Indus-
try at Common Law (1974) 20:1 McGill LJ 102 at 115).
4 See e.g. Bonthrone v Secretary of State for Scotland, 1987 SLT 34, Jauncey LJ, cited in
Diana Brahams, Pertussis Vaccine and Brain Damage: Two Claims Before the Courts
(1985) 326 Lancet 1137 (on the existence of cryptogenic (unknown) causes to eliminate
any possible causal connection between the pertussis vaccine and brain damage); see
also Loveday v Renton (1988), [1990] 1 Med LR 117 (QB) at 185 [Loveday], Stuart-
Smith LJ, discussed in Goldberg, Causation and Risk in the Law of Torts, supra note 2
at 13743 (after examining complex scientific evidence and arguments, Lord Stuart-
Smith held that the plaintiff had failed to prove on a balance of probabilities that the
pertussis (whooping cough) vaccine could cause permanent brain damage in young
children); Kay v Ayrshire and Arran Health Board, [1987] 2 All ER 417, 1987 SC 145
(HL Eng) (penicillin overdose was held to be not capable of causing or aggravating
deafness). Consider also the US mass tort litigation concerning Bendectin (Debendox),
780 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
The practical significance of establishing causation in a medicinal
product liability case cannot be overstated. In the United Kingdom,
whether the claim is in negligence or under the strict liability provisions
of the Product Liability Directive,5 proof of causation will often lead to ei-
the anti-nausea drug used in pregnancy. In contrast to thalidomide, where many lines
of evidence have shown that the drug caused phocomelia malformations (see Henning
Sjstrm & Robert Nilsson, Thalidomide and the Power of the Drug Companies (Har-
mondsworth: Penguin, 1972) at 15659), no causal link has ever been scientifically es-
tablished between Bendectin and birth defects. The Bendectin litigation demonstrated
a persistent failure by plaintiffs lawyers to prove causation. Joseph Sanders wrote the
seminal paper on this topic: The Bendectin Litigation: A Case Study in the Life Cycle
of Mass Torts (1992) 43:2 Hastings LJ 301. For a discussion of the litigations implications
in the United Kingdom, see Goldberg, Causation and Risk in the Law of Tort, supra note 2
at 10231. The litigation spawned two formative monographs: Michael D Green, Bendectin
and Birth Defects: The Challenges of Mass Toxic Substances Litigation (Philadelphia: Uni-
versity of Pennsylvania Press, 1996); Joseph Sanders, Bendectin on Trial: A Study of Mass
Tort Litigation (Ann Arbor: University of Michigan Press, 1998). Thirty years after Ben-
dectins withdrawal from the market, the drug (now renamed Diclegis) has won Food and
Drug Administration approval as the only FDA-approved treatment for morning sickness
(FDA News Release, FDA approves Diclegis for pregnant women experiencing nausea
and vomiting, online: US Food and Drug Administration
Press, Morning Sickness Drug Returns, The New York Times (8 April 2013), online: The
New York Times
5 EC, Council Directive 85/374/EEC of 25 July 1985 on the Approximation of the Laws,
Regulations and Administrative Provisions of the Member States Concerning Liability
for Defective Products, [1985] OJ, L 210/29. By article 4 of the Directive, and section 2(1)
of the Consumer Protection Act 1987 (UK), c 43, s 2(1), the person injured by a defective
medicinal product must prove the damage, the defect, and the causal relationship be-
tween them. The damage must have been caused wholly or partly by a defect in the
medicinal product. Thus the formal distinction between negligence and strict liability is
that with negligence, it must be proven that breach of a duty caused the harm, whereas
under the 1987 Act, it must be proven that a defect caused the damage (see Pamela R
Ferguson, Drug Injuries and the Pursuit of Compensation (London: Sweet & Maxwell,
1996) at 125). It appears that each member state will rely on its own theory of causation
as established in its civil liability system, though it has been observed that some kind of
semi-autonomous European understanding of causation could be established from
common elements of the member states legal systems (see Simon Whittaker, The EEC
Directive on Product Liability (1985) 5 YB Eur L 233 at 247). The argument that cau-
sation is likely to be defined and interpreted by national law and assessed by national
courts is strengthened by the decision of the European Court of Justice in Henning
Veedfald v. rhus Amtskommune, where the Court concluded that it was for the na-
tional court to decide whether a claim was to be categorized in respect of personal inju-
ry, property damage, or non-material damage (C-203/99 [2001] ECR I-3587 at I-3599
3600). This is subject to a qualification founded on the principle of effectiveness, in that
national laws must not by their interpretation of causation render ineffective either the
protection of injured persons or the restraints on liability of producers, since both reflect
the fair apportionment of risk of the Directive. In so doing, however, courts will take
into consideration the extent to which these causal issues combine issues of fact and
their evaluation and questions of law (see Simon Whittaker, Liability for Products: Eng-
EPIDEMIOLOGICAL UNCERTAINTY 781
ther a settlement or a successful claim.6 Conversely, a failure to establish
a causal link between a medicinal product and, for example, the alleged
medical conditions of claimants, may lead to such claims being struck out
as an abuse of the process of the court on the basis that each claim has no
real prospect of success.7
Epidemiology is defined as the field of public health and medicine
that studies the incidence, distribution and etiology of disease in human
populations.8 Epidemiological evidence is regularly presented to courts in
determining proof of causation in medicinal product liability litigation.
Building on the foundations of the authors previous monograph, which
supported the use of epidemiological evidence in dealing with problems of
proof of causation in alleged cases of adverse drug reactions,9 this paper
revisits the perennial problem of the role of epidemiological evidence in
assessing causation in product liability cases in a twenty-first century
context, examining recent cases in the United Kingdom, United States,
Australia, and Canada. In essence, it seeks to determine the extent to
which the courts in the highlighted cases have been pragmatic and fair in
their interpretation and utilization of epidemiological evidence, from the
perspective of both consumers and pharmaceutical manufacturers.
In order to establish factual causation in the context of medicinal
product liability, claimants must prove both general causation (whether
a substance is capable of causing a particular injury or condition in the
general population) 10 and specific causation (whether a substance
lish Law, French Law, and European Harmonisation (Oxford: Oxford University Press,
2005) at 51213). The European Commission believes that injured parties can estab-
lish the causal link in cases where a defective product causes damage irrespective of the
differences between national procedural rules (EC, Fourth Report on the Application of
Council Directive 85/374/EEC of 25 July 1985 on the Approximation of the Laws, Reg-
ulations and Administrative Provisions of the Member States Concerning Liability for
Defective Products (Brussels: EC, 2011) at 11), though it has noted the views of consum-
ers that there is difficulty in proving the causal link between the defect and damage
when such damage is complex in nature (ibid at 7). Consumers believe that the burden
of proof should be reversed (ibid).
6 Mark Mildred, Representing the Plaintiff in Geraint G Howells, ed, Product Liability,
Insurance and the Pharmaceutical Industry: An Anglo-American Comparison (Man-
chester: Manchester University Press, 1991) 24 at 27.
7 See Miller & Goldberg, supra note 1 at para 17.02.
8 Michael D Green, D Michal Freedman & Leon Gordis, Reference Guide on Epidemiol-
ogy in Reference Manual on Scientific Evidence, 3d ed (Washington: National Acade-
mies Press, 2011) 549 at 551 [Green, Epidemiology].
9 See generally Goldberg, Causation and Risk in the Law of Torts, supra note 2.
10 Merck & Co v Garza, 347 SW (3d) 256 at 262 (Tex 2011) [Garza], citing Merrell Dow
Pharmaceuticals v Havner, 953 SW (2d) 706 at 714 (Tex 1997) [Havner].
782 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
caused a particular individuals injury).11 Since epidemiology is based on
the study of populations and not individuals, it focuses on the question of
general causation rather than specific causation.12 Epidemiological evi-
dence may identify an association between a drug and a disease, but
whether such an association is causal requires an evaluation of the evi-
dence, with emphasis on the extent to which weaknesses of a studys de-
sign and implementation compromise its findings and inferences about
causation.13
The results of epidemiological studies cannot per se conclusively prove
specific causation. However, several cases have focused on the role that
epidemiological evidence plays in determining proof of specific causation,
which is a legal question addressed by courts.14 This paper explores the
ways in which probabilistic methods, including statistical refining with
individual risk factors, can be used in conjunction with epidemiological ev-
idence to determine specific causation.
Part IA explores the apparent tension between the levels of proof re-
quired in law and science, including the relationship between levels of
statistical significance and the claimants burden of proof. Part IB assess-
es the wisdom of using a doubling of the risk rule as a threshold to any
recovery. Notwithstanding the problems with the doubling of risk theory
in the United States, its usage appears to be gaining ground in the United
Kingdom. Moreover, in particular, the doubling of risk theory has come to
recent attention in the context of the utilization and value of epidemiolog-
ical or statistical evidence alone in determining causation on a balance of
probabilities, with discussion by the UK Supreme Court in Sienkiewicz v.
Greif.15 A cautious attitude toward the use of the doubling of risk rule in
the context of both general and specific causation is seen from the case
law explored. In examining the distinction between association and causa-
tion, Part IC discerns two main reasons for this judicial scepticism about
epidemiological evidence, namely the propriety of drawing causal infer-
ences from observed associations (a general causation issue) and the pro-
priety of drawing causal inferences in individual cases from concededly
11 See ibid; Garza, supra note 10. See also Michael D Green, The Future of Proportional
Liability: The Lessons of Toxic Substances Causation in M Stuart Madden, ed, Explor-
ing Tort Law (New York: Cambridge University Press, 2005) 352 at 366 [Green, Pro-
portional Liability].
12 See Green, Epidemiology, supra note 8 at 552; Steve Gold, Causation in Toxic Torts:
Burdens of Proof, Standards of Persuasion, and Statistical Evidence (1986) 96 Yale LJ
376 at 37980.
13 Green, Epidemiology, supra note 8 at 55253, 598.
14 Ibid at 609.
15 Sienkiewicz v Greif, [2011] UKSC 10, [2011] 2 WLR 523 [Sienkiewicz].
EPIDEMIOLOGICAL UNCERTAINTY 783
causal associations observed in samples of populations (a specific causa-
tion issue). These reasons are discussed in an analysis of the controversial
Scottish case of McTear v. Imperial Tobacco Ltd16 and the decision of the
Federal Full Court of Australia in Merck Sharp & Dohme (Australia) Pty
Ltd v. Peterson.17 In the context of McTear, Part ID discusses the necessi-
ty of requiring something more than a doubling of the risk to permit the
claimant to recover, and it stresses the role of judges in resolving this is-
sue.
Part IE discusses the implications for specific causation, in the context
of McTear, of epidemiology being based on the study of populations and
not individuals. It suggests that the limitations of epidemiological evi-
dence in determining specific causation as described by the trial judge are
somewhat inaccurate since, in establishing specific causation, epidemiolo-
gists can and do adjust for potentially confounding factors through logistic
regression techniques and other forms of statistical refining mechanisms.
Part IF therefore concludes with an examination of such statistical refin-
ing methods in determining specific causation in medicinal product liabil-
ity cases, including the use of Bayes theorem to help us understand how
statistical risks can be refined using personal risk factors. The paper does
not argue that Bayes theorem is necessarily the answer to the problem of
establishing specific causation in the context of epidemiological evidence.
Nonetheless, while recognizing the limitations of Bayes theorem, the pa-
per supports the view that logistic regression techniques and other forms
of statistical refining mechanisms using specific risk factors can and do
help in the process of giving quantitative or quasi-quantitative expression
to conclusions about the cause of disease in an individual drug product li-
ability claim that is based on epidemiological evidence. Finally, the paper
illustrates the increasing support for the refining and personalizing of ep-
idemiological evidence in cases of individual causation involving medicinal
products, as evidenced by the decision of the Ontario Superior Court of
Justice in Andersen v. St Jude.18
16 McTear v Imperial Tobacco Ltd, [2005] CSOH 69, 2 SC 1 (Ct Sess Scot) [McTear].
17 Merck Sharp & Dohme (Australia) Pty Ltd v Peterson, [2011] FCAFC 128, 284 ALR 1,
leave to appeal to HCA refused, [2012] HCA Trans 105 [Peterson].
18 See Andersen v St Jude Medical, Inc, 2012 ONSC 3660 (available on CanLII) at paras
542, 544, 555, 55859 [Andersen].
784 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
I. Reconciling the Standards of Proof in Law and Science in the United
Kingdom
A. Evidence of Causation for Purposes of Science and for Purposes of Law
There is an apparent tension between the levels of proof required in
law and in science. For the law of negligence, it is sufficient to show that
the balance of probabilitiesmeaning more than fifty per cent, or on a
preponderance of the evidenceindicates a causal connection. It is some-
times erroneously assumed by lawyers that scientists regard an associa-
tion as causal if it is ninety-five per cent certain.19 However, this is a mis-
interpretation of the so-called p value, which is merely the level of statis-
tical significance used to exclude the possibility that when something
transpires in a cohort of cases, it does so by chance (i.e., the null hypothe-
sis). When the p value falls below the threshold of 0.05, the investigator is
able to reject the null hypothesis since there is a less than one in twenty
chance that the link between exposure and disease is random.20 While
there is no generally accepted standard of scientific proof for causation,21
and neither the claimant nor the defendant is required to apply scientific
standards of proof when determining causation on a balance of probabili-
ties,22 such a standard must be much more than marginal.23 In light of
this apparent tension between the balance of probabilities standard and
19 See Green, Epidemiology, supra note 8 at 577 n 81. Equating statistical significance
with the legal burden of proof has been described as being like trying to find the short-
est path from Oxford to Cambridge by scrutinizing a map of London (DH Kaye, Ap-
ples and Oranges: Confidence Coefficients and the Burden of Persuasion (1987) 73:1
Cornell L Rev 54 at 66). Kaye demonstrates the distinction between statistical signifi-
cance and the civil burden of persuasion by using a hypothetical case (ibid at 6673).
There is often judicial reference to a statement that the level of 0.05 for statistical sig-
nificance is a much higher burden of proof than the civil burden of a preponderance of
the evidence or balance of probabilities (that is, greater than fifty per cent): see Green,
Epidemiology, supra note 8 at 577, citing In re Ephedra Products Liability Litigation,
393 F Supp (2d) 181 at 193 (SD NY 2005); Marmo v IBP, Inc, 360 F Supp (2d) 1019 at
1021 (D Neb 2005); Peter Feldschreiber, Leigh-Ann Mulcahy & Simon Day, Biostatis-
tics and Causation in Medicinal Product Liability Suits in Richard Goldberg, ed, Per-
spectives on Causation (Oxford: Hart, 2011) 179 at 190. Recent case law has referred to
Wyeths citation of the Reference Manual on Scientific Evidence to point to the errone-
ous nature of this approach (see Giles v Wyeth, Inc, 500 F Supp (2d) 1048 at 105657
(SD Ill 2007)).
20 Feldschreiber, Mulcahy & Day, supra note 19 at 184, 190.
21 See Loveday, supra note 4 at 124, Stuart-Smith LJ.
22 See Carter v Basildon and Thurrock University Hospitals NHS Foundation Trust,
[2007] EWHC 1882 at para 92 (available on BAILII) (QB) [Carter].
23 See Dingley v The Chief Constable, Strathclyde Police, 1998 SC 548 at 603 (Ct of Sess),
[1998] GWD 677, Lord Prosser [Dingley cited to SC], affd 2000 SC (HL) 77, cited in
Sienkiewicz, supra note 15, Lord Phillips.
EPIDEMIOLOGICAL UNCERTAINTY 785
the standard of statistical significance, courts must be alert to the prob-
lem that may be faced by an expert in readjusting his focus from the
ninety-five per cent confidence limit approach to the balance of probabili-
ties test.24 However, in Vadera v. Shaw25 the English Court of Appeal
reconciled the legal standard of proof on a balance of probabilities with
the scientific standard of statistical significance, in holding that a failure
to establish a statistically significant connection between the oral contra-
ceptive Logynon and the occurrence of strokes was fatal to the establish-
ment of proof of causation on a balance of probabilities. Lord Justice Hen-
ry stated:
The judge concluded, and in our respectful view was right on the ev-
idence to conclude, that the studies carried out and referred to by Dr
Lidegaard [for the plaintiff] did not establish a statistically signifi-
cant connection between Logynon and strokes. Such evidence cannot
be ignored by a judge. It is as common sense a conclusion as one
could wish to say that if the connection between A and B cannot be
shown with confidence to be other than a coincidence, then it cannot
be held on a balance of probabilities that A caused B. This is not to
allow scientists or statisticians to usurp the judges function, but ra-
ther to permit him to use their skills to discern a connection, or a
lack of connection, between two phenomena.26
B. Doubling of Risk Theory
Epidemiologists investigating disease causation measure the associa-
tion between exposure to an agent and the incidence of disease by using
the concept of relative risk. Relative risk is defined as the ratio of the in-
cidence of a disease in a population exposed to the agent to the incidence
of disease in a population that has not been exposed.27 For example, if ten
per cent of all people exposed to a drug develop a disease, compared with
five per cent of people who are unexposed, the disease occurs twice as fre-
quently among the exposed people. The relative risk is ten per cent/five
24 See Carter, supra note 22 at para 97. A confidence interval or confidence limit is a range
of values within which the true value is likely to fall (see Green, Epidemiology supra
note 8 at 621; Goldberg, Causation and Risk in the Law of Torts, supra note 2 at 137;
American Law Institute, Reporters Study: Enterprise Responsibility for Personal Injury,
vol 2: Approaches to Legal and Institutional Change (Philadelphia: American Law Insti-
tute, 1991) at 32428).
25 Vadera v Shaw (1998), 45 BMLR 162 (CA), (2000) 8 Med LR 316.
26 Ibid at 174. However, it is suggested that there was a failure by the trial judge and the
Court of Appeal to scrutinize adequately the scientific evidence in respect of causation
in this case (see Richard Goldberg, The Contraceptive Pill, Negligence and Causation:
Views on Vadera v. Shaw (2000) 8 Med L Rev 316 at 33135).
27 Green, Epidemiology, supra note 8 at 566, 627; Green, Proportional Liability, supra
note 11 at 366.
786 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
per cent (i.e., two). A relative risk of one shows no association between ex-
posure and disease.28
A significant attempt to reconcile the apparent tension between the
balance of probabilities standard and the standard for epidemiology has
emerged with the theory that causation can be proven on the balance of
probabilities by reference to the doubling of risk of injury theory. That
theory has long been recognized in the United States,29 where it has been
said that [t]he use of scientifically reliable epidemiological studies and
the requirement of more than a doubling of the risk strikes a balance be-
tween the needs of our legal system and the limits of science.30 However,
the theory has also been subject to trenchant criticism.31 In particular,
academics have argued that judges have adopted substantive changes in
causation law through the rubric of evidentiary admissibility decisions32
and have frequently conflated admissibility decisions and sufficiency of
28 See ibid; Green, Epidemiology, supra note 8.
29 See especially Daubert v Merrell Dow Pharmaceuticals, 43 F (3d) 1311 (9th Cir 1995),
63 USLW 2420, cert denied, 516 US 869, 116 S Ct 189 (1995) [Daubert II]. In that
case, the Court of Appeals, on remand from the Supreme Court of the United States,
held that the plaintiffs had to show not merely that Bendectin increased the likelihood
of injury, but that it more likely than not caused their injuries. In terms of statistical
proof, it had to be shown that plaintiffs mothers ingestion of Bendectin more than
doubled the likelihood of birth defects (ibid at 1320). This was reaffirmed by the Su-
preme Court of Texas in Havner, supra note 10 at 71618. The Supreme Court of Tex-
as has now expanded on its holding in Havner and adopted the position that a dou-
bling of risk is a necessary but not sufficient condition to prove causation (see Garza,
supra note 10 at 265). Vermont has also adopted the doubling of risk theory in a
slightly diluted form in the context of specific causation (see Blanchard v Goodyear
Tire and Rubber, 30 A (3d) 1271, 2011 VT 85 (Vt Sup Ct) at 127577). For an excellent
discussion of the implications of both cases, see Steve C Gold, Revisiting Relative Risk
Rules: Garza, Blanchard, and the Ever Evolving Role of Epidemiologic Proof in Toxic
Tort Cases (2012) 40 Prod Safety & Liab Rep (BNA) 50 [Gold, Revisiting Relative Risk
Rules].
30 Havner, supra note 10 at 718 (echoing the views of the court of appeals in Daubert II,
supra note 29).
31 See e.g. Lucinda M Finley, Guarding the Gate to the Courthouse: How Trial Judges
Are Using Their Evidentiary Screening Role to Remake Tort Causation Rules (1999)
49:2 DePaul L Rev 335 (criticizing the doublinginrisk evidentiary requirement for
epidemiological proof, describing the trend as seriously scientifically and legally mis-
guided at 348); Margaret A Berger, Upsetting the Balance Between Adverse Interests:
The Impact of the Supreme Courts Trilogy on Expert Testimony in Toxic Tort Litiga-
tion (2001) 64:23 Law & Contemp Probs 289 (criticizing the doubling of the risk rule
as a legal invention that creates a hard and fast rule that disposes of cases efficiently
but rests on assumptions that cannot be scientifically validated at this time at 30406);
Sander Greenland & James M Robins, Epidemiology, Justice, and the Probability of
Causation (2000) 40:3 Jurimetrics J 321 at 32526; Mark Geistfeld, Scientific Uncer-
tainty and Causation in Tort Law (2001) 54:3 Vand L Rev 1011 at 1015, 1018, 1020.
32 Finley, supra note 31 at 336.
EPIDEMIOLOGICAL UNCERTAINTY 787
evidence decisions.33 Those courts which require plaintiffs to produce ep-
idemiological studies with a relative risk of two are making a legal policy
determination to equate epidemiology, relative risk, general causation,
and the burden of proof on individual causation.34 Moreover, while the to-
tal number of judicial opinions that at least mention the concept of dou-
bling of risk has increased, US courts disagree as to the proper role of the
doubling of risk theory in deciding questions of both sufficiency and ad-
missibility of scientific evidence of causation in toxic tort cases. They do
not agree on whether to adopt the doubling of risk as a threshold, nor do
they agree on the meaning of such a threshold.35 As the reporters for the
American Law Institutes Restatement Third of Torts have noted:
Many courts accept the doubling of the incidence of disease in group
studies; some courts insist on doubling of risk as a minimum thresh-
old for establishing specific causation. Others have recognised that if
other known causes can be identified and eliminated, something less
than a doubling would still be sufficient to find specific causation. 36
Accordingly, the requirement of a relative risk of two for the admissibility
or sufficiency of epidemiological evidence has been subject to much scepti-
cism.37 The reporters for the Restatement Third of Torts, in discussing the
considerations that affect the appropriateness of determining the proba-
bility of specific causation based on the outcome of group studies, have
concluded that a judicial requirement that plaintiffs show a threshold in-
crease in risk (or a doubling of incidence in a group study) to satisfy the
burden of proof of specific causation is usually inappropriate.38
Notwithstanding the problems with the doubling of risk theory in the
United States, its existence appears to be gaining ground in the United
33 Ibid; see also Jean Macchiaroli Eggen, Clinical Medical Evidence of Causation in Toxic
Tort Cases: Into the Crucible of Daubert (2001) 38:2 Hous L Rev 369 at 37879; Mi-
chael D Green, The Future of Proportional Liability: The Lessons of Toxic Substances
Causation in M Stuart Madden, ed, Exploring Tort Law (Cambridge: Cambridge Uni-
versity Press, 2005) 352 at 36869. A recent instance of the conflation of both admissi-
bility and sufficiency of evidence requirements is Garza, supra note 10, discussed in
Gold, Revisiting Relative Risk Rules, supra note 29 at 53 (where he argues that by
framing a totality of evidence test as a matter of reliability, Garza explicitly conflated
rules of admissibility and substantive sufficiency in the weighing of evidence).
34 Finley, supra note 31 at 362.
35 Russellyn S Carruth & Bernard D Goldstein, Relative Risk Greater Than Two in Proof
of Causation in Toxic Tort Litigation (2001) 41:2 Jurimetrics Journal 195 at 199, 202
03.
36 Restatement (Third) of the Law of Torts 28(a) (2010) [Restatement].
37 See Carl F Cranor, Toxic Torts: Science, Law and the Possibility of Justice (Cambridge:
Cambridge University Press, 2007) at 23438, 281.
38 Restatement, supra note 36.
788 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
Kingdom. Of particular significance was the case XYZ v. Schering Health
Care Ltd,39 a trial of seven lead cases in group litigation against three
pharmaceutical companies in respect of cardiovascular injuries coming
under the collective description of venous thromboembolism (VTE). The
claimants alleged that their injuries were caused by taking the defend-
ants different brands of third-generation combined oral contraceptives.
The claimants alleged that the products they took were defective under
the Consumer Protection Act 1987 and the Product Liability Directive.
While the cause of action was based on strict liability, the requirement
common to both negligence and strict liability of proving a causal link be-
tween the product and the damage (i.e., the issue of general causation)
emerged as the first central issue requiring determination.
Justice Mackay stated that the claimant could prove that an exposure
to risk caused injury if that exposure had more than doubled the risk of
the injury occurring.40 This method of proving causation had previously
been applied in a case of bladder cancer, where the claimant had been tor-
tiously exposed to carcinogens and non-tortiously exposed to cigarette
smoke, both of which are potent causes of the condition.41 However, it has
been argued that since the doubling of risk approach is only valid where
the risk estimate represents mutually exclusive ways in which the injury
may have been caused and is sought to estimate the likelihood it was
one way which had operated in a particular case rather than one of the
39 70 BMLR 88, [2002] EWHC 1420 (QB) [XYZ].
40 Ibid at para 21, Mackay J. See also Miller & Goldberg, supra note 1 at paras 17.06
17.08.
41 See Cookson v Novartis Grimsby Ltd, [2007] EWCA Civ 1261, [2007] All ER (D) 465
(Nov) at para 74, Smith LJ. See also Ministry of Defence v AB, [2010] EWCA Civ 1317,
117 BMLR 101 at 149 [AB]. In the appeal for AB (AB v Ministry of Defence, [2012]
UKSC 9, [2013] 1 AC 78 [AB UKSC]), the UK Supreme Court found the doubling of risk
theory relevant in the context of examining the strength of claimants cases on causa-
tion and in determining whether the trial courts exercise of discretion under section 33
of the Limitation Act 1980 was appropriate. The trial judge was found to have wrongly
exercised his discretion. In dismissing the claimants appeals, the Supreme Court ob-
served that it was undesirable that a court which conducts an inquiry into whether a
claim is time-barred should, even when it considers its power under section 33 of the
1980 Act, have detailed regard to the evidence with which the claimant aspires to prove
its case. Nonetheless, because of the complexity of the claims placed before the trial
judge and the nature of the submissions about knowledge in section 14(1) of the 1980
Act, the trial judge was able to make a microscopic survey of the written evidence, es-
pecially in respect of causation. The Court of Appeal had been unusually well-placed in
exercising its discretion under section 33 to assess the claimants prospects of establish-
ing causation. Since the Court of Appeal had concluded that the claimants faced very
great difficulties in establishing causation, and the claimants had no real prospects of
success, it had been correct not to exercise its discretion to allow the claims to proceed.
To have done so would have been absurd (ibid, Lord Wilson at 100).
EPIDEMIOLOGICAL UNCERTAINTY 789
other possible ways.42 As such, the doubling of risk approach is not valid-
ly applicable as a method to cases of bladder cancer, where the mecha-
nism by which an agent (e.g., amines) present in two sources (e.g., occupa-
tional amine exposure and amines contained in cigarette smoke) causes
bladder cancer is unknown.43 By contrast, such comparisons of risk esti-
mates in the doubling of risk approach would be statistically valid where
the estimates relate to mechanisms which, even if their details are not
understood, are known to involve different agents, such as a birth defect
that may be attributable either to a medicinal product or to a background
risk.44
The utilization and value of epidemiological or statistical evidence
alone in determining causation on a balance of probabilities was subject to
some interesting debate in the [UK] Supreme Court in Sienkiewicz v.
Greif.45 The reason for this discussion, as pointed out by Baroness Hale,46
was the presence of an obiter observation by Lady Justice Smith in her
judgment in Sienkiewicz that in a case of multiple potential causes, a
claimant can demonstrate causation by showing that the tortious expo-
sure has at least doubled the risk arising from the non-tortious cause or
causes.47 Their Lordships were postulating the scenario where, having
established general causation between the toxic agent and the disease, ep-
idemiological evidence might be used to establish specific causation. While
their Lordships held unanimously that there was no room for introducing
the doubling of risk approach to single exposure48 mesothelioma cases or
multiple defendant mesothelioma cases,49 differences in view emerged
42 Jane Stapleton, Factual Causation, Mesothelioma and Statistical Validity (2012)
128:2 Law Q Rev 221 at 22325 [emphasis added].
43 Ibid at 226.
44 Ibid at 223.
45 Richard Goldberg, Using Scientific Evidence to Resolve Causation Problems in Product
Liability: UK, US and French Experiences in Richard Goldberg, ed, Perspectives on
Causation (Oxford: Hart Publishing, 2011) 149 at 153.
46 See Sienkiewicz, supra note 15 at para 169.
47 Ibid at para 63 (quoting Smith LJ in the Court of Appeal, [2010] QB 370).
48 Ibid at para 67. As Lord Kerr noted, [t]he use of the expression single exposure may
be misleading in this context (ibid at para 199). It is probably better expressed as sin-
gle tortious exposure cases (ibid at para 173, Hale B). These are cases where only one
defendant exposed the victim to asbestos and there was only one possible tortious
source for the exposure, and the only other exposure creating a risk of developing meso-
thelioma was environmental exposure to low level asbestos dust in the general atmos-
phere (ibid at paras 113, Lord Rodger; 199, Lord Kerr; 207, Lord Dyson).
49 Ibid at paras 106, Lord Phillips; 160, Lord Rodger; 169, Hale B; 188, Lord Mance; 203,
Lord Kerr; 220, Lord Dyson. In such cases, the Fairchild v Glenhaven Funeral Services
Ltd ([2002] UKHL 22, [2003] 1 AC 32) and Barker v Corus (UK) plc ([2006] UKHL 20,
[2006] 2 AC 572) exception for mesothelioma applied to provide the claimant with an
790 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
with the obiter discussion of the general applicability of the doubling of
risk theory using epidemiological evidence to determine proof of causation
in personal injury cases.
Lord Phillips discussed the XYZ decision and took the view that, while
the case contained a detailed and illuminating discussion of epidemiolo-
gy, it did not afford any direct assistance to the question whether the
doubles the risk testas he called itwas appropriate for determining
causation in a case of multiple potential causes.50 His reasoning was
somewhat obscured by his misclassification of the contraceptives in this
case. He stated that the issue was whether a second generation of oral
contraceptives more than doubled the risk of causing deep vein throm-
bosis (DVT) that was created by the first generation of contraceptives … It
was not whether the DVT suffered by the claimants had been caused by
the second generation of oral contraceptives.51
In fact, the issue was whether the claimants had proved that third
generation combined oral contraceptives caused a true excess risk of VTE,
which was more than twice the risk caused by second generation com-
bined oral contraceptives. Both sides in XYZ agreed that if the claimants
failed to prove this, the action could not succeed. However, both parties
had also agreed that if the claimants could prove a true excess risk of
VTE, they would also succeed on the second issue, which was whether the
relevant products were defective within the meaning of section 3 of the
Consumer Protection Act 1987 (i.e., that their safety would not be such as
persons generally were entitled to expect)52. The test of defectiveness un-
der section 3(2)(a) of the Act includes consideration of instructions or
warnings associated with the product. Thus, the reasoning behind the
doubling of risk theorys relevance to establishing that the third genera-
tion contraceptives were defective was that if the UK Supreme Court
ruled that the true risk of VTE was more than doubled with third genera-
tion combined oral contraceptives, women and their prescribers were enti-
tled to be told this before making their decisions or giving their advice, re-
spectivelyand they had not been.53
action if it proved that the defendant materially increased the risk that the claimant
might develop mesothelioma (see Sienkiewicz, supra note 15 at paras 103, 107, Lord
Phillips). But now see the Supreme Court decision in Durham v BAI (Run Off) Ltd
([2012] UK SC 14, [2012] 1 WLR 867), where the Supreme Court held that the actual
development of mesothelioma was an essential element of the cause of action (ibid at
paras 52, 6566, Lord Mance; see also para 124, Lord Phillips, dissenting).
50 Sienkiewicz, supra note 15 at para 74.
51 Ibid.
52 See XYZ, supra note 39 at para 20.
53 Ibid at paras 2021. See also Miller & Goldberg, supra note 1 at paras 17.0617.08.
Lord Phillips reasoning seems to ignore the fact that causation was
inherently behind the courts approach. As Justice Mackay explained in
XYZ:
EPIDEMIOLOGICAL UNCERTAINTY 791
The reason why the Claimants accept, through Lord Brennan QC,
that this first issue is capable of disposing of the claims should be set
out. It is not because an increase of less than two would fail to ren-
der the product defective within the meaning of the Act, though the
Defendants would so argue if they had to. It is for reasons of causa-
tion that he accepts this burden, correctly in my view. If factor X in-
creases the risk of condition Y by more than two when compared
with factor Z, it can then be said, of a group of say 100 with both ex-
posure to factor X and the condition, that as a matter of probability
more than 50 would not have suffered Y without being exposed to X.
If medical science cannot identify the members of the group who
would and who would not have suffered Y, it can nevertheless be
said of each member that she was more likely than not to have
avoided Y had she not been exposed to X [emphasis added].54
While Lord Phillips concluded55 that there was no scope for the dou-
bling the risk test in cases where two agents operated cumulatively and
simultaneously in causing the onset of a disease, since in such cases the
material contribution rule in Bonnington Castings v. Wardlaw56 would
apply, he submitted57 that there was no reason in principle why the dou-
bles the risk test should not be applied where the initiation of a disease
was dose-related and there had been consecutive exposures to an agent or
agents that cause the disease (e.g., McGhee v. National Coal Board).58
Lord Phillips regarded Hotson v. East Berks Area Health Authority59 as an
example of the latter situation.60
However, neither Lord Rodger nor Baroness Hale took such a view,
both holding that a doubling of risk approach was not an appropriate test
of causation.61 Lord Rodger stressed that where statistical evidence estab-
lished that exposure to a substance more than doubled the risk of a dis-
ease, this would not amount to proof, on the balance of probabilities, that
the exposure actually caused the disease.62 Meanwhile, Lord Dyson did
54 XYZ, supra note 39 at para 21.
55 See Sienkiewicz, supra note 15 at para 90.
56 Bonnington Castings Ltd v Wardlaw, [1956] AC 613 (HL) at 620, Lord Reid.
57 See Sienkiewicz, supra note 15 at para 93.
58 McGhee v National Coal Board, [1972] UKHL 7, [1973] 1 WLR 1.
59 Hotson v East Berkshire Health Authority, [1988] UKHL 1, [1987] 1 AC 750 [Hotson].
60 See Sienkiewicz, supra note 15 at para 93.
61 Ibid at paras 156, 158, 161, Lord Rodger; 170, 173, Hale B.
62 Ibid at para 156.
792 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
not find it necessary to decide whether there are any circumstances in
which, as a matter of English law, causation can be proved on the basis of
epidemiological evidence alone.63 He expressed the view that there [was]
no a priori reason why, if the epidemiological evidence [was] cogent
enough, it should not be sufficient to enable a claimant to prove his case
without more.64 By contrast, Lord Kerr stressed the need to treat the use
of epidemiological evidence to seek to establish any specific proposition in
an individual case with great caution.65 He felt that there was a real dan-
ger that so-called epidemiological evidence [would] carry a false air of
authority.66
Finally, Lord Mance felt that whether and when epidemiological evi-
dence could prove a case was a question best considered not in the ab-
stract but in a particular case, when and if that question [arises].67 If it
could arise, he would hope and expect that this would only occur in the
rarest of cases.68
This cautious attitude toward the use of the doubling of risk rule in
the context of specific causation has been reflected in medicinal product
liability litigation concerning the anti-inflammatory drug Vioxx. In Merck
Sharp & Dohme (Australia) Pty Ltd v. Peterson,69 plaintiffs alleged in rep-
resentative proceedings that consumption of Vioxx increased the risk of a
myocardial infarction (heart attack) and that Vioxx had caused or con-
tributed to the myocardial infarction of the class representative, Mr. Pe-
terson.70 The trial judge, Justice Jessup, had held that the epidemiological
evidence had demonstrated that Vioxx had doubled the risk of heart at-
tack across the population as a whole,71 and that consumption of Vioxx
materially contributed to Petersons heart attack. 72 Yet in upholding
Merck Australias appeal on the issue of causation, the Full Court criti-
cized the doubling of risk approach as being apt to mandate an award of
63 Ibid at para 221.
64 Ibid at para 222.
65 Ibid at para 205.
66 Ibid at para 206.
67 Ibid at para 192.
68 Ibid.
69 Peterson, supra note 17.
70 Ibid at 2.
71 Merck Sharp & Dohme (Australia) Pty Ltd v Peterson (2010), 184 FCR 1 at para 570,
[2011] FCAFC 128; see generally Claudia Newman-Martin, Manufacturers Liability
for Undiscoverable Design Flaws in Prescription Drugs: A Merck-Y Area of the Law
(2011) 19:1 Torts Law Journal 26.
72 Peterson, supra note 17 at para 772.
EPIDEMIOLOGICAL UNCERTAINTY 793
compensation to applicants who have not, in truth, been injured by the
respondent.73 It also noted that, while a relative risk of two might imply a
fifty per cent probability that the risk had been realized in a typical case,
a relative risk of less than two would imply a probability of less than fifty
per cent. The trial judges finding of relative risk had been about two.74
C. Association Versus Causation
It is arguable that it would be an oversimplification to think that the
views of Lord Phillips in Sienkiewicz will help to signal a green light to
the establishment of proof of causation on a balance of probabilities by a
mere doubling of relative risk. The matter was addressed in Merrell Dow
Pharmaceuticals, Inc. v. Havner75 where, having stated that a balance be-
tween the needs of the legal system and the limits of science could be
achieved by the use of scientifically reliable epidemiological studies and
the requirement of more than doubling the risk, the Supreme Court of
Texas added the caveat:
We do not hold, however that a relative risk of more than 2.0 is a
litmus test or that a single epidemiological test is legally sufficient
evidence of causation. Other factors must be considered. As already
noted, epidemiological studies only show an association. There may
in fact be no causal relationship even if the relative risk is high.76
The latter sentence is of particular importance, and while Lord Phil-
lips in Sienkiewicz referred to the caveat expressed in Havner,77 he omit-
ted that last sentence and ignored its import in his final analysis. Unlike
Lord Phillips, Lord Rodger stressed the importance of the distinction be-
tween association and causation. In this context, Lord Rodgers reason for
73 Ibid at para 110. This would be the case since those applicants who were actually in-
jured by causes other than the respondents actionable conduct will be able to recover
compensation because, for them too, a relative risk of greater than 2 can be said to im-
ply probability of greater than 50% that the respondents actionable conduct was the
cause of their loss (ibid). However, this criticism is misconceived, since the problem of
compensation to those not injured by a defendant is generic in any system that uses a
preponderance of the evidence rule and has no relevance to the type of evidence em-
ployed to determine whether the plaintiff has met the preponderance threshold.
74 Ibid at para 111. See also Seltsam Pty Ltd v McGuiness, [2000] NSWCA 29, 49 NSWLR
262, Spigelman CJ (while in Australian law the test of actual persuasion did not re-
quire epidemiological studies to reach the level of risk of 2.0, the closer the ratio ap-
proaches 2.0, the greater the significance that can be attached to the studies for the
purposes of drawing an inference of causation in an individual case. The strands in the
cable must be capable of bearing the weight of the ultimate inference at para 137).
75 Havner, supra note 10.
76 Ibid at 718.
77 Supra note 15 at para 88.
794 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
scepticism about epidemiological evidence concerns the propriety of draw-
ing causal inferences from observed associations (a general causation is-
sue); yet there is seemingly a further reason for scepticism in his speech,
regarding the propriety of drawing causal inferences in individual cases
from concededly causal associations observed in samples of populations (a
specific causation issue).78 Lord Rodgers speech is more compelling in
that it shows a greater understanding of both the significance and the
limitations of epidemiological evidence, and it demonstrates a reluctance
to support the general application of the doubling of risk theory to deter-
mining proof of both general and specific causation in personal injury cas-
es.79 Lord Rodger accepted that epidemiological and statistical evidence
may form an important element in proof of causation, and he supported
the utilization and value of epidemiological evidence where a claimant
was required to prove his case on a balance of probabilities.80 However, he
emphasized that, since by its very nature statistical evidence does not
deal with the individual case, the court should not proceed to find a causal
relationship in that particular case without further non-statistical evi-
dence (e.g., evidence of temporality of the appearance of results of the ex-
posure).81 In so doing, he cited Phipson on Evidence, which states that
[w]here there is epidemiological evidence of association, the court should
not proceed to find a causal relationship without further, non-statistical
78 This can be contrasted with the other reason for judicial scepticism about epidemiologi-
cal evidence, namely the propriety of drawing causal inferences from observed associa-
tions. It is often difficult to tease out from the decisions which form of judicial treatment
is taking place.
79 See Sienkiewicz, supra note 15 at paras 163, Lord Rodger; 173, Hale B.
80 Ibid at para 163, Lord Rodger.
81 Ibid. Baroness Hale also opined that the existence of a statistically significant associa-
tion between factor X and disease Y does not prove that in the individual case it is more
likely than not that factor X caused disease Y (ibid at para 170). Lord Mance accepted
that epidemiological evidence, used with proper caution, could be admissible and rele-
vant in conjunction with specific evidence related to the individual circumstances and
parties. The significance a court might attach to it depended on the nature of the epi-
demiological evidence, and of the particular factual issues before the court (ibid at para
191). Lord Kerr considered that [i]t is an essential and minimum requirement … that
there be evidence connecting avowedly relevant statistical information produced by the
epidemiological studies to the facts of the case (ibid at para 205). Lord Dyson also
stressed the association/causation dichotomy, stating that epidemiology … seeks to es-
tablish associations between alleged causes and effects … However, in an individual
case, epidemiology alone cannot conclusively prove causation (ibid at para 218). See al-
so the recent discussion by the High Court of Australia in Amaca Pty Ltd v Booth,
[2011] HCA 53, 283 ALR 461 at para 49 [Amaca] (where French CJ distinguished be-
tween mere statistical correlation between conduct and injury and the need to estab-
lish causal connection between the conduct and injury).
EPIDEMIOLOGICAL UNCERTAINTY 795
evidence.82 Lord Rodger illustrated his example of evidence of temporali-
ty in the context of a medicinal product and an adverse effect, where there
was a strong epidemiological association between a drug and some condi-
tion that could have been caused in some other way.83 He submitted that
epidemiological evidence, along with evidence that the claimant devel-
oped the condition immediately after taking the drug, could be sufficient
to allow the judge to conclude that the drug caused the condition on the
balance of probability.84
The Federal Full Court of Australias decision in Peterson is another
example of courts reluctance to draw inferences from a population to an
individual in the context of medicinal products.85 There the Full Court
upheld the but for test of causation and found that the trial judges find-
ings of fact were insufficient to sustain the position that, on the balance of
probabilities, but for the consumption of Vioxx, Petersons myocardial in-
farction would not have occurred.86 The court concluded that while the ep-
idemiological evidence meant that it was possible Vioxx had caused Peter-
sons myocardial infarction, there were other strong potential causes, such
as age, gender, hypertension, hyperlipidaemia, obesity, left ventricular
hypertrophy and [a] history of smoking. 87 Peterson was therefore a
member of a group within the community, 25% of whom were expected by
… cardiologists to suffer a heart attack within 5 years.88 These personal
circumstances seriously diminished the strength of the epidemiological
evidence as a strand in the cable of circumstantial proof.89 Accordingly,
the Full Court held that it was not more probable than not that Vioxx,
whether alone or in combination with Petersons personal risk factors,
was a necessary condition of the occurrence of his heart attack.90 While a
relative risk of two could be converted into a fifty per cent statistical like-
lihood that Vioxx was causally implicated in the occurrence of a myocar-
dial infarction, there were other candidates as causes of the injury. The
strength of the epidemiological strand did not rise above the possibility
that it was in the mix of factors which may have caused Petersons heart
82 Hodge M Malek, ed, Phipson on Evidence, 17th ed (London, UK: Thomson Reuters Le-
gal, 2010) at paras 3427.
83 Sienkiewicz, supra note 15 at para 163.
84 Ibid.
85 Peterson, supra note 17.
86 Ibid at paras 10305.
87 Ibid at para 120.
88 Ibid.
89 Ibid at para 113.
90 Ibid at para 120.
796 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
attack.91 While the fact that the plaintiff in Peterson suffered from several
personal risk factors prima facie cuts against recovery, this mere fact
alone does not resolve the import of the epidemiological evidence. The dif-
ficulty lies with the fact that epidemiological evidence conflates people
with different underlying conditions, and it may not be known what the
relative risk is for those individuals with no history of heart disease com-
pared to individuals, such as Peterson, with a long history of heart dis-
ease. Indeed, there is a strong argument, based on Mercks own VIGOR
study of Vioxx, that the relative risk of taking Vioxx is equally strong in
both subgroups and that Vioxx could have caused a heart attack even in
someone with a history of heart disease.92 There is thus no data to support
the Full Courts conclusion that personal circumstances seriously dimin-
ish the strength of the epidemiological evidence. Accordingly, the courts
approach was arguably a guess by a sceptical court that Vioxx is incapable
of being identically implicated both in cases of individuals with pre-
existing heart problems and in cases of those without.
In light of Lord Rodgers observations in Sienkiewicz, and from the
perspective of the propriety of drawing causal inferences from observed
associations, the mere existence of a statistically significant association is
insufficient to establish a causal relationship without the presence of fur-
ther non-statistical evidence. To establish a causal relationship, factors
such as those enumerated by Sir Austin Bradford Hill would need to be
utilized to determine whether a reported association is causal.93 This point
91 Ibid at para 123. Special leave to appeal to the High Court of Australia was refused
since the applications were deemed not suitable vehicles for the consideration of the
relevant questions of principle that would warrant the grant of leave, having regard to
the findings of fact of the primary judge and the Full Courts treatment of them (Peter-
son, supra note 17, leave to appeal to HCA refused, [2012] HCATrans 105).
92 See McDarby v Merck & Co, 949 A2d 223 at 234 (NJ Super App Div 2008) [McDarby]
(noting that the results of the VIGOR study in March 2000 revealed a higher incidence
of adverse cardiovascular events with those who received rofecoxib (Vioxx) than with
those patients who received naproxen, in patients with and without a history of ather-
osclerotic cardiovascular disease, and in patients with or without classic risk factors for
cardiovascular disease at 234). See also Gold, Revisiting Relative Risk Rules, supra
note 29. Consider the following hypothetical. For those with no pre-existing heart prob-
lems, taking Vioxx raises the risk of heart attacks 101%, more than doubling the risk
from 1% to 2.01%. For those with pre-existing heart conditions like Peterson, taking Vi-
oxx raises the risk of heart attacks 101%, from 10% to 20.01%. Vioxx is identically im-
plicated in both scenarios.
93 Sir Austin Bradford Hill, The Environment and Disease: Association or Causation?
(1965) 58:5 Proceedings of the Royal Society of Medicine 295 at 295 [Hill, Association
or Causation]. These aspects of association (that is, strength of association, consisten-
cy, specificity, temporality, biological gradient, plausibility, coherence, experiment,
and analogy) are utilized to determine whether a reported association is causal or non-
genuine. For recent support for the Bradford Hill factors as providing a guide to the
EPIDEMIOLOGICAL UNCERTAINTY 797
was emphasized by the Scottish Court of Session in McTear v. Imperial
Tobacco Ltd,94 a decision which takes a cautious approach to the use of
epidemiological evidence and stresses the impossibility of applying epi-
demiological studies to determine causation in individual cases. The case
illustrates the courts scepticism about the epidemiological evidence, as
the court questioned the propriety of drawing causal inferences from ob-
served associations when determining general causation. The court was
also sceptical about the propriety of drawing causal inferences in individ-
ual cases from causal associations observed in samples of populations.
While this case concerns tobacco products, its implications are particular-
ly pertinent to problems involving medicinal products, where the role of
epidemiological evidence in proving both general and specific causation is
prominent.
UK developments in this area have often focused on the difficulty in
proving general and specific causation using epidemiological evidence95
derived from trends in general populations. This was graphically illus-
trated by McTear. In that case, the pursuer, the widow of a smoker,
sought to recover damages from the defenders, who had manufactured the
John Player brand cigarettes that the pursuers late husband had
smoked. The pursuers husband had contracted squamous cell carcinoma
of the lung, and the pursuer averred both that cigarette smoking could
cause lung cancer (an issue of general causation) and that her husbands
lung cancer was caused by his smoking (an issue of individual or specific
causation).
The problem of establishing a general causal link between cigarette
smoking and cancer was exacerbated by the fact that, unlike all the ciga-
rette companies in the United States and all the other cigarette compa-
nies in the United Kingdom, Imperial Tobacco had not accepted that there
kind of considerations that lead to an inference of causal association, see Amaca, su-
pra note 81 at para 49.
94 McTear, supra note 16 at para 6.158. However, the presentation of the list of factors in
textbooks as criteria for inferring causality or associations in a way as to imply that
all the conditions are necessary has been described as unfortunate (Sander Green-
land, ed, The Evolution of Epidemiologic Ideas: Annotated Readings on Concepts and
Methods (Los Angeles: Epidemiology Resources Inc, 1987) at 14). As Greenland cor-
rectly observes, Sir Austin Bradford Hill expressly stated that he did not intend to lay
down hard and fast rules of evidence that must be obeyed before we accept cause and
effect (ibid, citing Hill, Association and Causation, supra note 93 at 299). Hill added
that [n]one of [his] nine viewpoints can bring indisputable evidence for or against the
cause-and-effect hypothesis and none can be required as a sine qua non (ibid at 299
[emphasis in original]). See also Cranor, supra note 37 at 10205.
95 Epidemiology has been defined as the study of patterns of disease occurring in human
populations and the factors that influence these patterns (McTear, supra note 16 at
para 6.157).
798 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
was a causal link between smoking and disease, especially lung cancer.96
In respect of establishing general causation, Lord Nimmo Smith conclud-
ed that, in the absence of such an admission, and indeed of any evidence
that this was an inference that should be drawn, the burden of proof lay
on the pursuer to show that cigarette smoking could cause lung cancer.97
In the absence of support from animal experiments, proof of causation be-
tween cigarette smoking and lung cancer depended on what was proven
before the court about epidemiological studies.
Lord Nimmo Smith held that, in accordance with the Scots law of ex-
pert evidence, it was necessary to consider whether the evidence of any
expert witness had imparted to the court special knowledge of the subject
matter of epidemiology so as to enable the court to draw its own conclu-
sions from epidemiological evidence. Accordingly, it was not open to the
court to form its judgment on the evidence without being taught how to
analyze the epidemiological evidence to a sufficient extent, and without
being provided with sufficient factual material to enable proof on the bal-
ance of probabilities not only that there was an association between ciga-
rette smoking and lung cancer, but also that the proper conclusion to be
drawn from this was that there was a causal connection between them.98
This distinction between association and causation in the context of the
general causation issue lay at the heart of Lord Nimmo Smiths conclu-
sions. In his view, when an association between an exposure and a condi-
tion was judged to be statistically significant, that in itself did not consti-
tute a judgment that there was a causal connection between an exposure
and a condition.99 He explained:
The finding of an association between an exposure and a condition or
disease, even if judged to be statistically significant, does not of itself
connote that a causal connection between the two is established.
This is a matter for further exercise of judgment, taking account of
96 See ibid at paras 2.58, 2.76, 6.30. This was notwithstanding the generally accepted view
for over 50 years that cigarette smoking could cause lung cancer (see Richard Doll & A
Bradford Hill, Smoking and Carcinoma of the Lung: Preliminary Report [1950] 4682
Brit Med J 739; Richard Doll & A Bradford Hill, The Mortality of Doctors in relation to
their Smoking Habits: A Preliminary Report [1954] 4877 Brit Med J 1451; McTear, su-
pra note 16 at para 5.208 (evidence of Sir Richard Doll)). The defence in McTear admit-
ted that the World Health Organization, along with United Kingdom and United States
governments, had accepted for years that cigarette smoking can cause lung cancer.
However, they averred that [c]igarette smoking has not been scientifically established
as a cause of lung cancer and, although various theories have been advanced, the cause
or causes of lung cancer are unknown and the mechanism or mechanisms whereby lung
cancer develops are unknown (ibid at para 2.7).
97 See ibid at paras 2.78, 2.80.
98 See ibid at para 6.155.
99 See ibid at para 6.158.
EPIDEMIOLOGICAL UNCERTAINTY 799
such criteria as the consistency, the strength, the specificity, the
temporal relationship and the coherence of the association … This
must, I think, especially be so when, in the view of Sir Richard Doll
… cigarette smoking is not a necessary cause nor a sufficient cause of
lung cancer[.]100
Lord Nimmo Smith then addressed the concept of relative risk,
concluding that even a relative risk derived from comparison of the inci-
dence of lung cancer in smokers and non-smokers, of a magnitude such
that a positive association may be judged to be strong enough to establish
causation between the two, did not connote the establishment of a causal
link.101
As we shall now see, this scepticism about epidemiological evidence
and questions about the propriety of drawing causal inferences from ob-
served associations was not the only problem that the pursuer had in es-
tablishing general causation in McTear. The court also had to be taught
the relevant epidemiology.
D. Teaching Courts Epidemiology
In respect of general causation, Lord Nimmo Smith held that the pur-
suer had failed to prove, in accordance with the requirements of the Scots
law of evidence relating to expert witnesses, that cigarette smoking could
cause lung cancer.102 This was because the pursuer had failed to lead suf-
ficient evidence, in the form of primary epidemiological literature that
drew a causal connection between cigarette smoking and lung cancer, to
impart to the court special knowledge of the subject matter so as to enable
the court to form its own judgment about it and the conclusions to be
drawn from it.103 Lord Nimmo Smith stated that a fundamental defect in
the presentation of the pursuers case was the failure to present in court
any of the primary literature that had concluded that there was a causal
connection between cigarette smoking and lung cancer.104 In his view, this
was a missed opportunity:
This could have been done: it is clear that the survey of British doc-
tors, on which Sir Richard Doll and colleagues have worked for
many years, is regarded as a classic of its kind, both because of the
pioneering nature of the research, a preliminary report of which was
published as Doll and Hill (1950), and because this has been followed
100 Ibid.
101 See ibid at para 6.159.
102 Ibid at paras 6.170.
103 Ibid at paras 6.155, 6.1626.163.
104 Ibid at para 6.163.
800 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
up with subsequent papers over several decades. I could at least
have been shown these papers, which I assume disclosed the data,
the statistical techniques and all the other considerations which led
to the authors conclusions, so that I could see for myself whether
these conclusions were soundly based. The opportunity was there,
with Sir Richard Doll in the witness box, and indeed Prof[essor]
Friend for one thought that evidence would be given about this sur-
vey. Warning had been given on behalf of [Imperial Tobacco Ltd] …
that Sir Richard Dolls data were of potential interest to the court.
But in the event no attempt was made to show me the data.105
A recent Scottish case, Smith v. McNair,106 reaffirms this cautious ap-
proach to the interpretation of epidemiological evidence.107 It stresses the
need for experts to teach a court how to analyze epidemiological evidence
before it can come to a judgment by interpreting that evidence. While ac-
knowledging that medical witnesses are entitled to refer to medical litera-
turein particular to published papers by epidemiologistseven if they
themselves are not epidemiologists,108 Lord McEwan in McNair stressed
the need to look at such evidence critically because its writers could not be
cross-examined themselves. Such scientific evidence only becomes a factor
for consideration if it is intelligible, convincing and tested.109 According-
ly, in Scotland, the cases are at one in emphasizing that where a pursuer
seeks to rely on epidemiological evidence of disease to prove causation, the
pursuer must impart to the court special knowledge of the subject matter
of epidemiology, so that the court can form its reasoned judgment on the
epidemiological evidence.110
Such a cautious approach to epidemiological evidence was central to
the decision in McNair. While sympathetic to the experts who were out-
105 Ibid at para 6.162.
106 [2008] CSOH 154 (available on WL UK) (OH Scot) [McNair]; see also Richard Goldberg,
Causation, Idiopathic Conditions and the Limits of Epidemiology (2009) 13:2 Ed L
Rev 282.
107 See Dingley, supra note 23 at 555, Lord President Rodger; 604, Lord Prosser, concur-
ring; McTear, supra note 16 at 5.11, Lord Nimmo Smith.
108 Main v McAndrew Wormald Ltd (1988), [1988] SLT 141 at 142.
109 McNair, supra note 1086 at para 18, citing Davie v Magistrates of Edinburgh (1952),
1953 SC 34 at 40, 1953 SLT 54 Ct Sess (Scot).
110 See also United States, the Advisory Committee Note (2000 Amendment) to Fed R Evi-
dence 702. The Amendment not only stresses that the expert conducts the application
of principles and methods to the facts of cases reliably, but also reiterates the venera-
ble practice of using expert testimony to educate the factfinder on general principles
(ibid). It notes that it might be important in some cases for an expert to educate the
finder of fact about general principles, without ever attempting to apply these principles
to the specific facts of the case (ibid).
EPIDEMIOLOGICAL UNCERTAINTY 801
with their chosen discipline and abroad in the field of epidemiology,111
Lord McEwan concluded nonetheless that the experts were unable to ex-
plain the studies, which seemed to him to raise more questions than an-
swers.112 Unlike McTear, however, McNair shows less of an impression
of what Chris Miller has described as a dogmatic aversion113 to statisti-
cal evidence. Lord McEwan felt that many of the concerns about the evi-
dence might have been assuaged if the authors of the reports had been
called to testify and if there had been some statistical evidence presented.
Without such assistance, the judge was at once disabled from being able
properly to evaluate the worth of the study or to draw the proper conclu-
sions.114 In his view, therefore, this was an appropriate case for epidemi-
ologists to give evidence and for experts to explain their studies. He did
not, however, believe that this was always the case, and he suggested that
reliance on doctors and epidemiologists can almost lead the court unwit-
tingly into a kind of satellite litigation on issues away from the pursuers
case. 115 He seemed to regard McTear and another Scottish decision,
Dingley,116 as two recent examples of this.117 However, the use of statistics
in determining causation is hardly satellite litigation. In both McTear and
Dingley, it was a primary issue which required resolution in the face of
scientific uncertainty. The concern with Scots law taking such a cautious
approach to epidemiological evidence is therefore that such an approach
may make it harder to even discern that there is any possible reconcilia-
tion of the legal standard of proof on a balance of probabilities with the
scientific standard of statistical significance.
Even more importantly, there is also concern that the placing an obli-
gation on a plaintiff to teach epidemiology to a court suggests that the
court can remain passive in this process. This is surely an unhelpful ap-
proach in cases such as McTear and in cases involving adverse reactions
allegedly caused by medicinal products. In such cases, there is a clear so-
cial expectation that judges will resolve these matters to the satisfaction
of both parties. As a leading American judge has observed about cases
where judges preside over non-jury trials:
111 McNair, supra note 106 at para 80.
112 Ibid at para 81.
113 Chris Miller, Causation in Personal Injury: Legal or Epidemiological Common Sense
(2006) 26:4 LS 544 at 566 [Miller, Causation in Personal Injury].
114 McNair, supra note 106 at para 80.
115 Ibid at para 16.
116 Dingley, supra note 23.
117 McNair, supra note 106 at paras 27, 29.
802 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
Passivity of the court is no virtue when serious scientific questions of
more than passing importance are involved. The court owes an obli-
gation to the parties, to society, and to itself to assist in obtaining the
best possible answers to the scientific questions before it. That will
mean forcing the parties to gather and present evidence effectively,
calling upon other experts as necessary, and studying to obtain the
understanding needed to maintain effective control.118
Had the pursuer in McTear explained the epidemiological evidence
properly, and had Lord Nimmo Smith been more receptive to evidence of
relative risk as well as taken a more active role in forcing the pursuer to
present her evidence effectively, it would seem that general causation
could have been established. Moreover, Lord Nimmo Smith should have
given more weight119 to the surely important fact that the defenders ad-
mitted that the World Health Organization, along with the governments
of the United Kingdom and the United States, had accepted for many
years that cigarette smoking can cause lung cancer.120
Of course, irrespective of the conclusions on general causation, there
remained the problem of establishing individual causation in the context
of naked statistical evidence.121 It is to this that we now turn.
E. The Statistical Chance/Personal Chance Dichotomy
It has been argued that there is a dichotomy between two kinds of
chancesone statistical and the other personal. A statistical chance is
a figure collected from previous unconnected outcomes, giving a probabil-
118 Jack B Weinstein, Improving Expert Testimony (1986) 20:3 U Rich L Rev 473 at
49596. Judge Weinstein also encourages judges presiding over non-jury trials to be-
come familiar with the scientific background by reading about the issues and discussing
them with the experts (ibid at 494). Weinstein is cited in Snyder v United States (De-
partment of Health & Human Services) 2009 WL 332044 at 2 (Ct Fed Cl 2009).
119 Counsel for the pursuer had submitted (unsuccessfully) that considerable weight should
be placed on the fact that this proposition had come to be generally accepted (see
McTear, supra note 16 at para 6.41).
120 Ibid at paras 2.7, 6.30.
121 As Gary Wells has observed, the term naked statistical evidence is ill-defined in the
legal literature (see e.g. David Kayes use of the term in in David Kaye, Naked Statisti-
cal Evidence, Book Review of Quantitative Methods in Law: Studies in the Application
of Mathematical Probability and Statistics to Legal Problems by Michael Finkelstein,
(1980) 89:3 Yale LJ 601 at 603; David Kaye, The Limits of the Preponderance of the
Evidence Standard: Justifiably Naked Statistical Evidence and Multiple Causation
(1982) 7:2 Law & Soc Inquiry 487 at 488), though it typically refers to probabilities that
are not case specific to the events in issue but rather existed prior to or independently
of the particular case being tried (Gary L Wells, Naked Statistical Evidence of Liabil-
ity: Is Subjective Probability Enough? (1992) 62:5 Journal of Personality and Social
Psychology 739 at 739).
EPIDEMIOLOGICAL UNCERTAINTY 803
ity of that outcome in any non-individual case, whereas a personal
chance is peculiar to a particular individual.122 A statistical chance has
no compensatory value, until the data is personalised.123
The impossibility of applying statistics derived from epidemiological
studies to determine causation in individual cases was cited as the princi-
pal reason for the pursuers failure to prove individual, or specific, causa-
tion in McTear. Epidemiological evidence could not prove that it was more
likely than not that but for his smoking of cigarettes, the deceased would
not have contacted lung cancer.124 As Lord Nimmo Smith put it:
The information provided in an observational epidemiology is gener-
ally such that it can neither confirm nor refute a causal relationship,
particularly when the exposure in question is not specifically associ-
ated with a certain condition (ie the exposure is always associated
with the condition, and vice versa). Epidemiology cannot provide in-
formation on the likelihood that an exposure produced an individu-
als condition. The population attributable risk is a measure for pop-
ulations only and does not imply a likelihood of disease occurrence
within an individual, contingent upon that individuals exposure.
The fact that cases and noncases can emerge both from the unex-
posed and the exposed groups show that the likelihood of the indi-
vidual occurrence cannot be reliably predicted from his or her expo-
sure group membership alone. The group estimates obscure the un-
derlying heterogeneity of the population, so that it is entirely possi-
ble that other group memberships besides exposure, like genetic pro-
file, socio-economic status, workplace, diet and other exposures
make a major contribution to disease occurrence. The question of us-
ing epidemiological data for individual causation raises the problem
of identifying a particular individual who was harmed by the expo-
sure. While models such as the assigned share concept, derived from
attributable fractions, have attempted to deal with this, they suffer
from the limitations mentioned by Dr Lewis. The attempt to identify
exposure as the sole cause of disease in an individual produces a
statement counter to fact in that it implies that the individual would
have remained healthy if the exposure had not occurred. This, as Dr
Lewis said, is not provable and cannot be derived from epidemiologi-
cal data.125
122 Timothy Hill, A Lost Chance for Compensation in the Tort of Negligence by the House
of Lords (1991) 54:4 Mod L Rev 511 at 512 [Hill, Lost Chance].
123 Ibid at 518. See also Hotson v East Berkshire AHA, [1987] 2 WLR 287, 303769, Croom-
Johnson LJ.
124 See McTear, supra note 16 at paras 6.180, 6.1846.185. See also the above discussion in
Sienkiewicz, supra note 15 at paras 163, 170, 191, 205, 218. As in the UK, US courts of-
ten bifurcate both specific and general causation elements (see Joseph Sanders, The
Controversial Comment C: Factual Causation in Toxic-Substance and Disease Cases
(2009) 44:4 Wake Forest L Rev 1029 at 1032).
125 McTear, supra note 16 at para 6.180.
804 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
Lord Nimmo Smith concluded that, given there were other possible
causes of lung cancer other than cigarette smoking, and given that lung
cancer could occur in a non-smoker, it was not possible to determine in
any individual case whether but for an individuals cigarette smoking he
probably would not have contracted lung cancer.126 In doing so, Lord
Nimmo Smith referred to [t]he fallacy of applying statistical probability
to individual causation.127
However, his dicta require closer scrutiny. While Lord Nimmo Smith
was correct to observe that there are limitations to epidemiological evi-
dence, his description of these limitations is somewhat inaccurate. In stat-
ing that group estimates obscure the underlying heterogeneity of the
population, so that it is entirely possible that other group memberships
besides exposure, like genetic profile, socioeconomic status, workplace,
diet and other exposures make a major contribution to disease occur-
rence,128 he fails to appreciate that epidemiologists can and do adjust for
these potentially confounding factors through logistic regression statisti-
cal techniques.129 Notwithstanding Lord Nimmo Smiths doubts about
causal proof based on population estimates of relative risk, these esti-
mates are relevant to individual cases, even though they do not directly
measure the probability of causation in an individual case.130 Moreover,
Miller has suggested that, while Lord Nimmo Smiths dogmatic aversion
to statistical evidence means that epidemiology alone will never secure
recovery in respect of specific causation in such cases,131 use of epidemio-
logical evidence that satisfies the criteria developed by Sir Austin Brad-
ford Hill would seem to be hard to gainsay.132 Thus, Miller has argued
that if an individual had been one of the cases in a case control study that
yields strength of association (relative risk), then in light of such strength
of association and other Bradford Hill criteria, it seems perverse to hold
126 Ibid at paras 6.1846.185.
127 Ibid at para 6.184. For the need to exercise caution in the use of general statistics in es-
tablishing causation, and the importance of looking at the claimants individual circum-
stances, see the observations of Brooke LJ in Wardlaw v Farrar, [2003] EWCA Civ
1719, [2003] 4 All ER 1358. See also Amaca Pty Ltd v Ellis, [2010] HCA 5 at para 62,
263 ALR 576; Sienkiewicz, supra note 15 at paras 152, 163, Lord Rodger; 170, 172, Hale
B; 19092, Lord Mance; 20406, Lord Kerr.
128 McTear, supra note 16 at para 6.180.
129 See also Kenneth J Rothman & Sander Greenland, Modern Epidemiology, 2d ed (Phil-
adelphia, Lippincott-Raven Publishers, 1998) at 39495.
130 See Steve C Gold, When Certainty Dissolves Into Probability: A Legal Vision of Toxic
Causation for the Post-Genomic Era (2013) 70:1 Wash & Lee L Rev 237 at 281, 303
[Gold, Certainty Dissolves]; Greenland & M Robins, supra note 31 at 32122.
131 See Miller, Causation in Personal Injury, supra note 113 at 566.
132 Ibid.
EPIDEMIOLOGICAL UNCERTAINTY 805
that it is less probable than not that the exposure caused that individuals
condition.133 I contend that Miller is correct in concluding that a causal
relationship would exist in such circumstances. Indeed, Sir Austin Brad-
ford Hill emphasized that [n]one of my nine viewpoints can bring indis-
putable evidence for or against the cause-and-effect hypothesis and none
can be required as a sine qua non,134 and this has been judicially ap-
proved in the United States.135 Sir Austin Bradford Hill specifically cau-
tioned against overly emphasizing the importance of specificity at the ex-
pense of strength of association, referring specifically to smoking and lung
cancer. 136 In doing so, he provided a particularly apt example:
Coming to modern times the prospective investigations of smoking
and cancer of the lung have been criticized for not showing specifici-
tyin other words the death rate of smokers is higher than the
death rate of non-smokers from many causes of death. … But here
surely one must return to my first characteristic, the strength of as-
sociation. If other causes of death are raised 10, 20 or even 50% in
smokers whereas cancer of the lung is raised 9001,000% we have
specificitya specificity in the magnitude of the association.
…
We must also keep in mind that diseases may have more than one
cause.
…
In short, if specificity exists we may be able to draw conclusions
without hesitation; if it is not apparent, we are not thereby neces-
sarily left sitting irresolutely on the fence.137
I suggest that Lord Nimmo Smith in McTear undervalued the signifi-
cance of the widely accepted magnitude of strength of association between
cigarette smoking and cancer, and that he was wrong to treat the Brad-
ford Hill factors as criteria that all needed to be satisfied before such an
association could amount to a causal connection between smoking and
lung cancer. In his discussion of the impossibility of applying statistics de-
rived from epidemiological studies to determine causation in individual
cases, Lord Nimmo Smith failed to appreciate that, in determining specif-
ic causation, epidemiologists can and do adjust for potentially confounding
133 Ibid [emphasis in original].
134 Hill, Association or Causation, supra note 93 at 299 [emphasis in original].
135 See Cook v Rockwell Intern Corp, 580 F Supp 2d 1071 at 1098 (D Colo 2006).
136 Hill, Association or Causation, supra note 93 at 297. The High Court of Australia has
recently stressed that reference to relative risk ratio may act as an indicator of strength
of association (see Amaca, supra note 81 at para 49).
137 Hill, Association or Causation, supra note 93 at 297.
806 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
factors through logistic regression techniques and other forms of statisti-
cal refining mechanisms. It is to these techniques that we now turn.
F. Overcoming the Statistical Chance/Personal Chance Dichotomy:
Statistical Refining Mechanisms Using Specific Risk Factors
The problem of using statistics deriving from trends in general popu-
lations to prove causation in an individual case has been recognized judi-
cially by the House of Lords in Hotson v. East Berkshire Area Health Au-
thority138 and in Gregg v. Scott,139 and by the UK Supreme Court in Sien-
kiewicz v. Greif.140
Yet it is arguable that while epidemiological evidence reaches conclu-
sions on the incidence of a disease in a population in the form of relative
risk, this relative risk can be refined to draw conclusions about the cause
of disease in an individual using specific risk factors,141 such as those pre-
sent in Mr. McTears case and in Mr. Petersons case. This has been ac-
cepted by American courts in the context of pharmaceutical product liabil-
ity litigation.142 In McDarby v. Merck & Co, Inc.,143 a case involving the
drug Vioxx, epidemiological evidence was combined by experts with the
presence of the plaintiffs personal heart attack risk factors, namely his
age, low levels of good cholesterol, weight, and diabetes. The New Jersey
court regarded this as ample evidence to support an increased risk result-
ing from the combined effects of diabetes and Vioxx, and concluded that
Vioxx had been a substantial contributing factor to the plaintiffs heart at-
138 Hotson, supra note 59 at 789, Lord Mackay.
139 See Gregg v Scott, [2005] UKHL 2, [2004] 2 AC 176 at paras 2633, Lord Nicholls; 153,
Lord Phillips.
140 Sienkiewicz, supra note 15 at paras 152, 163, Lord Rodger; 170, 172, Hale B; 19092,
Lord Mance; 205, Lord Kerr.
141 See Berger, supra note 31 at 306.
142 Green, Epidemiology, supra note 8 at 616, citing Havner, supra note 10 at 720; see al-
so Smith v Wyeth Ayerst Laboratories Co, 278 F Supp (2d) 684 at 70809 (WDNC 2003)
(discussing an experts attempt to apply principles of relative risk from an epidemiologi-
cal study on the relationship between diet drugs and primary pulmonary hypertension
(PPH) to the risk faced by the individual plaintiff, who developed PPH after taking pre-
scription appetite suppressants based on specific risk characteristics (duration of use
and timing of use). The experts opinion was deemed unreliable).
143 McDarby, supra note 92 at 270.
EPIDEMIOLOGICAL UNCERTAINTY 807
tack.144 In so concluding, the court applied a substantial factor standard in
the context of concurrent causation, in preference to the but for test.145
In this context, I have suggested that statistics regarding evidence of
general causal links between a drug and an injury (a statistical chance)
could be refined into statistics establishing a specific causal link between
the drug and the adverse reaction in the case at issue (a personal
chance)146 using logistic regression techniques and other forms of statisti-
cal refining mechanisms.147
Logistic regression techniques identify determinants of a particular
outcome and assess the extent of the contribution of these determinants,
adjusting for confounding factors148 that may influence the contribution.149
Logistic regression is also closely linked to other forms of statistical refin-
ing, such as Bayes theorem. Bayes theorem can modify evaluations of
probability based on initial assumptions in the light of more data that lat-
er becomes available. It expresses the relationship between the probabil-
ity of a proposition (A) evaluated before the utilization of new data (B)
(prior probability), and the probability of the same proposition evaluated
after the utilization of the new data (posterior probability).
Thus:
Posterior
Probability
of A given B
=
Prior Probability of A
Probability of B given A
1
x
Unconditional Probability
of B
i.e. P(A/B) = P(A) x [P(B/A) / P(B)]
Prior probabilities can therefore be updated in the light of new data
from epidemiological studies as they accumulate, providing both fact find-
144 Ibid at 26970.
145 Ibid (applying the substantial factor standard, causation was appropriately demon-
strated by long term use of Vioxx and medical and/or scientific proof of a nexus be-
tween [that use] and … plaintiffs condition at 271).
146 Hill, Lost Chance, supra note 122 at 518.
147 This builds on the authors discussion in Goldberg, Causation and Risk in the Law of
Torts, supra note 2 at 3940.
148 A confounding factor is a factor that is both a risk factor for the disease and one associ-
ated with the exposure in issue. Confounding refers to the situation where an associa-
tion between an exposure and an outcome is all or partly due to a factor that affects the
outcome but which is unaffected by the exposure (see Green, Epidemiology, supra
note 8 at 621).
149 Rothman & Greenland, supra note 129; O Caster et al, Logistic Regression in Signal
Detection: Another Piece Added to the Puzzle, Letter to the Editor, (2013) 94:3 Clinical
Pharmacology & Therapeutics 312.
808 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
ers in individual product liability cases, and policy-makers such as the
Food and Drug Administration and the European Medicines Agency, with
an update of the estimated risk.150 The main difficulty with such posterior
probabilities is that frequentist statisticians151 who rely on epidemiological
evidence regard them as necessarily subjective, since they reflect not only
data but also subjective prior probabilities.152 However, objective Bayesi-
ans153 use Bayes theorem without eliciting prior probabilities from sub-
jective beliefs, avoiding the charge of subjectivism.154 This has been sup-
ported in the pharmaceutical product liability context by Professor Joseph
Gastwirth, who has adopted a data-based approach to ensure that the
choice of prior distribution is objective and unbiased. He uses the first
case control study or an analysis of adverse event and case reports to de-
termine two prior distributions, one the most favourable to the defendant,
and the other centred on or near the estimated relative risk from the first
study. This method of determining two prior distributions restricts the
degree of subjectivity that an analyst can insert into a Bayesian approach.
This is very important in the legal context, where lawyers would likely
choose the expert who obtains the more favourable result for them. The
data-based approach helps to avoid bias in the choice of prior distribu-
tion.155 Others have also tried to apply Bayes theorem in the evaluation of
the reliability of medical and scientific evidence in toxic tort cases.156
150 See Joseph L Gastwirth, Should Law and Public Policy Adopt Practical Causality as
the Appropriate Criteria for Deciding Product Liability Cases and Public Policy? (2013)
12:3 Law, Probability and Risk 169 [Gastwirth, Practical Causality]; see also Patrick
Ryan et al, Learning from Epidemiology: Interpreting Observational Database Studies
for the Effects of Medical Products (2013) 5:3 Statistics in Biopharmaceutical Research
170 at 178 (supporting a Bayesian framework to interpret observational database stud-
ies for the effects of medical products and suggesting that future work can extend the
Bayesian framework to include such elements as the Bradford Hill factors).
151 Frequentist statisticians are those who define probability as the frequency of a certain
measurement or observation. The frequentist approach focuses on the probability of the
data, given the hypothesis. See Maarten HP Ambaum, Frequentist vs Bayesian Statis-
ticsA Non-Statisticians [sic] View (July 2012), online: Department of Meteorology, Uni-
versity of Reading, UK
152 See David H Kaye & David A Freedman, Reference Guide on Statistics in Reference
Manual on Scientific Evidence, 3d ed (Washington, DC: National Academies Press,
2011) 211 at 258, 273.
153 Bayesian statisticians define probability as the plausibility of a hypothesis given in-
complete knowledge or data (see Ambaum, supra note 150).
154 Kaye & Freedman, supra note 152 at 259 n 123.
155 Gastwirth, Practical Causality, supra note 150.
156 See e.g. Neal C Stout & Peter A Valberg, Bayes Law, Sequential Uncertainties, and
Evidence of Causation in Toxic Tort Cases (2005) 38:4 Mich JL Reform 781 at 787
(submitting that judges should apply Bayesian probabilistic approaches in toxic tort
EPIDEMIOLOGICAL UNCERTAINTY 809
However, the strongest criticism of Bayes theorem is the difficulty of ar-
riving at a sufficiently accurate evaluation of a pre-existing probability to
which experimental data can be applied.157
Bayes theorem tells us that the value of a piece of evidence in testing
a particular assertion is determined by its likelihood ratio. The likelihood
ratio (LR) is the probability of the evidence supposing our assertion is
true, divided by the probability of the evidence if the assertion is not
true.158 The Centre for Evidence Based Medicine at the University of Ox-
ford provides a helpful example of the LR in the following:
[Y]ou have a patient with anaemia and a serum ferritin of 60mmol/l
and you find in an article that 90 per cent of patients with iron defi-
ciency anaemia have serum ferritins in the same range as your pa-
tient (= sensitivity) and that 15 per cent of patients with other caus-
es for anaemia have serum ferritins in the same range as your pa-
tient (1 specificity). This means that your patients result would be
six times as likely (90/15) to be seen in someone with, as opposed to
someone without, iron deficiency anaemia, and this is called the LR
for a positive test result.159
An alternative statement of Bayes theorem explains it in terms of
odds.160 Bayes theorem expresses the relationship between the odds in fa-
vour of a hypothesis before the utilization of new data (prior odds) and the
odds in favour of the hypothesis after taking into account the new data
(posterior odds). The prior odds must be multiplied by the likelihood ratio
of the new piece of data to generate the posterior odds.
Applying this to the Peterson case, 161 Vioxx-induced MI could be
compared with a catch-all alternative, no Vioxx-induced MI. Alternative-
Posterior Odds = Prior Odds x Likelihood Ratio
Thus:
cases when evaluating the reliability of medical and scientific evidence, and in so doing
permitting the fact finder to decide only those toxic tort claims for which there is relia-
ble and relevant scientific support for each link in the causal chain).
157 See Sir Richard Eggleston, Evidence, Proof and Probability, 2d ed, (London, UK: Wei-
denfeld and Nicolson, 1983) at 171.
158 See Bernard Robertson & GA Vignaux, Interpreting Evidence: Evaluating Forensic Sci-
ence in the Courtroom (Chichester, UK; John Wiley & Sons 1995) at 17.
159 Centre for Evidence Based Medicine, Likelihood Ratios, online: CEBM
160 The relationship between odds and probability is:
Odds = Probability / (1 Probability)
Thus the probability of 0.9 = odds of 9:1.
161 Peterson, supra note 17.
810 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
ly, one could compare Vioxx-induced MI with some specific alternative,
such as diet-induced MI, totally uncaused MI, or no MI. The likeli-
hood ratio would then be the ratio of the probabilities of developing MI
under these two hypotheses.162
A statistical chance could be refined and personalized into a personal
chance using specific factors which are embodied in the likelihood ratio.
The probabilities in the likelihood ratio can be decomposed into factors in
the light of specific case information in respect of patient history. Such
factors could include the risk factors in Peterson,163 namely Petersons age
(LR (Ag)), gender hypertension (LR (Gh)), hyperlipidaemia (LR (Hypl),
obesity (LR (Ob)), left ventricular hypertrophy (LR (LVH)), and a history
of smoking (LR (Hs)).164 The likelihood ratio is then found by obtaining
the product of all the individual likelihood ratio factors.
Diagrammatically this can be expressed by:
LR = LR (Ag) x LR (Gh) x LR(Hypl) x LR(Ob) x LR(LVH) x LR(Hs)
(Caveat: components, i.e. Ag etc., must be statistically independent)
The use of all these factors is dependent on the specific case infor-
mation available. If all specific case information in respect of the factors is
available, the posterior odds are calculated as follows:
Posterior Odds = Prior Odds x LR = LR (Ag) x LR (Gh) x LR(Hypl) x
LR(Ob) x LR(LVH) x LR(Hs)
Thus the posterior odds can be further refined by combining the prior
odds, based on background information, with the likelihood ratios, based
on case-specific information, to produce as accurate a posterior probability
as possible.165 The nature of each risk factor likelihood ratio can represent
a particularistic property of the individual claimant, provided they can be
determined in the case in issue.166 There is therefore a need to obtain sta-
162 I am grateful to Professor Philip David for his explanation of this point.
163 Peterson, supra note 17.
164 Ibid at para 120.
165 See B Donatini, I Le Baye & P Krupp, Causality Assessment of Spontaneous Report-
ing: Correlation Between Bayesian and Other Approaches (1993) 7:4 Pharmaceutical
Medicine 255 at 256.
166 Personal Communication, Professor Philip David, Statistical Laboratory, Centre for
Mathematical Sciences, Cambridge University, 12 July 2013; Robertson and Vignaux
advocate that scientific evidence concerning an issue should be combined with other ev-
EPIDEMIOLOGICAL UNCERTAINTY 811
tistics with an evidentiary foundation before such likelihood ratios can be
calculated.167
This would seem to be a possible tool that can improve probabilistic
precision in the Petersontype case and in other cases involving medicinal
products. In so doing, this tool can overcome the difficulties associated
with the statistical chance/personal chance dichotomy.
It is clear that while Bayes theorem could provide a normative ap-
proach to legal decision making in the context of causation and medicinal
products, implementing the theorem, in practice, is likely to be difficult.168
We have seen that Bayes theorem assumes the presence of conditionally
independent new evidence to update the previous evidence, but this new
evidence is absent in many cases involving alleged adverse drug reactions.
This complicates the application of the theorem. The use of individual risk
factor likelihood ratios in respect of individual items of evidence is poten-
tially valuable, but these may be difficult to calculate in practice. It should
also be conceded that if sample sizes are so small that one cannot dis-
aggregate data to provide information on individual risk factors, then the
statistical refining process will fail. Moreover, while more detailed indi-
vidual ratios might improve the accuracy of the posterior odds, the intro-
duction of too many additional quantities with imperfect estimation could
degrade it.169 However, the basic point here is not to suggest that Bayes
theorem is necessarily the answer to the problem of establishing specific
causation in the context of epidemiological evidence. It is rather that lo-
gistic regression techniques and other forms of statistical refining mecha-
nisms using specific risk factors can and do help in the process of giving
quantitative or quasi-quantitative expression to conclusions about the
idence relating to the same issue, and that the most effective way of doing so is to ex-
press the evidence in likelihood ratio form for it to be subsequently combined with other
evidence (Robertson & Vignaux, supra note 158 at 220). In addition, they have observed
that the likelihood ratios importance is that it determines relevance and probative val-
ue, the key determinants of admissibility of expert evidence (ibid at 22). Robertson and
Vignaux have submitted that it is not essential to have precise numbers for each of the
probabilities to assess the likelihood ratio (ibid). However, this would seem arguable in
complex cases involving the establishment of causation with medicinal products (see
Goldberg, Causation and Risk in the Law of Torts, supra note 2 at 43 n 242).
167 See Katherine Grevling, Book Review of Interpreting Scientific Evidence: Evaluating
Forensic Science in the Courtroom by Bernard Robertson & GA Vignaux, (1996) 112
LQR 509 at 510.
168 Stephen E Fienberg & Mark J Schervish, The Relevance of Bayesian Inference for the
Presentation of Statistical Evidence and for Legal Decisionmaking (1986) 66:4 BUL
Rev 771 at 772, 782, 794.
169 See also A Philip Dawid, The Role of Scientific and Statistical Evidence in Assessing
Causality in Richard Goldberg, ed, Perspectives on Causation (Oxford: Hart, 2011) 133
at 14045.
812 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
cause of disease in an individual claim that is based on epidemiological
evidence.
Support for the refining and personalizing of epidemiological evidence
in cases of individual causation involving medicinal products is now gain-
ing traction in courts. One relevant recent case is Andersen v. St Jude, a
Canadian trial on the merits of a class claim concerning the safety of the
mechanical prosthetic heart valves and annuloplasty rings with Silzone
that were designed and manufactured by the defendants and approved for
sale in Canada in the late 1990s.170 In Andersen, the Ontario Superior
Court recognized that the doubling of risk standard is merely a presump-
tive threshold, so that a negative finding on causation could be rebutted
using probative individualized evidence in a subsequent individual tri-
al.171
Silzone was a proprietary term for a coating comprising layers of tita-
nium, pallodium, and an outer layer of metallic silver, which was applied
to a polyester sewing cuff that surgeons used to attach a prosthetic heart
valve to heart tissue. Silver is known as an antimicrobial, and the Silzone
coating was designed to inhibit the growth of bacteria that could cause
endocarditis, an infection of the lining of the heart that is a potential seri-
ous complication of heart valve surgery. Other than the application of the
coating to the sewing cuff, the Silzone valves were of the same design as
conventional mechanical valves that the defendants had manufactured for
many years. 172 Following a randomized clinical trial called AVERT,
which had revealed a small, but statistically significant increase in ex-
plants due to a medical complication known as paravalvular leak (PVL) in
patients who had received a Silzone implant, the defendants in Andersen
issued a worldwide recall of all Silzone-coated products in early 2000.173 A
class action against St. Jude Medical was commenced in 2001. At its core
was a claim in negligence, which focused on the breach of St. Judes duty
of care to patient class members and questions of general causation.174
The plaintiffs advanced the theory that Silzone [was] a toxic sub-
stance that interfere[d] with the cells involved in tissue healing and
impair[ed] the bodys ability to properly incorporate the Silzone de-
vice into the heart, thereby causing or contributing to a variety of se-
rious medical complications for Silzone patients. As medical compli-
cations can occur with all prosthetic heart valves, a key inquiry in
170 Andersen, supra note 18 at para 1.
171 Ibid at para 1.
172 Ibid at para 2.
173 Ibid at para 1.
174 Ibid at para 4.
EPIDEMIOLOGICAL UNCERTAINTY 813
this trial was whether Silzone … materially increased [the] risk of
[patients] experiencing one or more of these complications.175
While the couching of this inquiry in terms of material increase in risk
may seem peculiar,176 the issue being addressed was whether the plain-
tiffs could prove that the Silzone valve caused a true excess risk of the
medical complicationsabove the risk caused by the conventional valves.
In essence, the Ontario Superior Court was adopting the same approach
to the issue as the English High Court in XYZ v. Schering Health Care
Ltd.177
Notwithstanding that Justice Lax found that the defendants did not
breach any duty of care in the pre-market design, manufacture, and test-
ing, or in the post-market surveillance, warning, and recall of Silzone-
coated products,178 she proceeded to determine the common issues of cau-
sation had the court found differently on the breach of duty issue. She ex-
plained that
statistical epidemiological evidence ha[d] been presented to aid [her]
in determining whether or not Silzone valve patients experience a
higher risk of medical complications than conventional valve pa-
tients. In other words, the purpose of this evidence [was] to deter-
mine the risk of medical complications posed by the Silzone valve
relative to the risk posed by the conventional valve.179
This introduced the concept of relative risk, which was a numerical ex-
pression of the risk of medical complications for one class of patients rela-
tive to another.180 While recognizing the limitations of epidemiological
evidence, in that it ought not to be considered determinative of individual
causation,181 Justice Lax used simple arithmetic, the application of the
but for test, and the balance of probabilities standard to conclude that
for the purposes of issues of general causation in a class action trial, a
doubling of risk standard should be adopted. A product (here the Silzone
valve) thereby creates a material risk of an adverse event where the risk
175 Ibid at para 5.
176 The reason for the use of the word material, which was formulated by Justice Cullity
in his certification decision (ibid at paras 5, 520), was to ensure that findings with re-
spect to whether Silzone increases the risk of complications would be sufficiently mean-
ingful that they would be indicative of something more than a remote possibility of cau-
sation (ibid at para 528).
177 XYZ, supra note 40 at paras 2021.
178 See Andersen, supra note 18 at paras 6, 18283, 214.
179 See ibid at para 384.
180 Ibid.
181 See ibid at para 395.
814 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
is at least twice the risk of the adverse effect occurring in the absence of
the products use (namely, when using the conventional valve).182
However, in an important development which may help to constrain
the emergence of overly optimistic emphasis on doubling of risk as some
magic formula with which to prevent cases from going forward to trial in
the future, Justice Lax explained that the establishment of material risk
and the application of the doubling of risk standard were not determina-
tive of individual causation. Instead, for the purpose of individual class
member claims, the application of the doubling of risk standard is merely
a presumptive as opposed to a prescriptive threshold, so that a negative
finding on causation (where the relative risk is below two) could be rebut-
ted using probative individualized evidence in a subsequent individual
trial.183 Justice Lax added that if she had found the defendants to be neg-
ligent, she would have presumptively applied the doubling of risk stand-
ard for materiality.184 Accordingly, patients who suffered complications for
which the increase in risk was not material (i.e., where the relative risk
was below two) or even not statistically significant would still be able to
recover at the individual stage of those proceedings, provided they pre-
sented sufficient individualized evidence to rebut the presumption of a
lack of causation flowing from a relative risk below two, and that they
were able to persuade their trier of fact that Silzone was the but for
cause of their complications.185 The benefit of adopting this approach is
that it does not shut the door on individual class members solely on the
basis of evidence regarding group risk.186 As Justice Lax explained, the
adoption of a presumptive approach to materiality, permitting negative
findings on causation to be rebutted by individualized evidence, allowed
her to advance the litigation and to outline how a trier of fact at the indi-
vidual stage of similar proceedings could properly utilize relative risk as
ascertained by epidemiological data.187
182 See ibid at paras 53238. The arithmetical explanation for adopting the doubling of
risk rule (ibid at paras 53234) is almost identical to that provided in XYZ, supra note
40 at para 21.
183 Andersen, supra note 18 at paras 542, 544, 555, 55859.
184 Materiality means determining whether the Silzone valve materially increased the
risk of a particular medical complication (ibid at para 427).
185 Ibid at para 559.
186 Ibid at para 560.
187 Ibid at para 562.
EPIDEMIOLOGICAL UNCERTAINTY 815
Conclusion
We can make the following observations about recent cases from the
United Kingdom that examine the role of epidemiological evidence in as-
sessing causation in medicinal product liability claims.
There remain considerable difficulties in reconciling standards of proof
in law and in science. Despite the trenchant criticisms of the doubling of
risk theory in the United States, the theory appears to be gaining ground
in the United Kingdom. However, the majority of the UK Supreme Court
in Sienkiewicz appears to be sceptical of introducing a threshold for the
use of epidemiological evidence and remain of the view that such evidence
can be useful but must be viewed with caution. Without further non-
statistical evidence, there is reluctance for courts to proceed to find the ex-
istence of a causal relationship. The danger otherwise is that counsel, in
assessing the chances of success of no win, no fee multi-party product li-
ability litigation, especially that which involves medicinal products, may
regard this doubling of risk theory as the sole basis on which to allow or
prevent cases from going forward to trial, even where epidemiological evi-
dence is lacking. This could potentially prejudice access to justice in fu-
ture cases. If the doubling of risk approach is to be embraced by UK
courts, it should be treated as it was in the Canadian decision of Ander-
sen, where the standard operated as merely a presumptive as opposed to a
prescriptive threshold, so that a negative finding on causation (where the
relative risk is below two) could be rebutted using probative individual-
ized evidence in a subsequent individual trial. In such cases where there
is a dearth of epidemiological evidence, courts and, for that matter, fund-
ing bodies should learn from the US experience and should avoid insisting
on epidemiological studies which have a relative risk of greater than two,
allowing all evidence which falls within a zone of reasonable [scientific]
disagreement188 to be considered.
While it seems the United Kingdom is becoming more receptive to the
need for epidemiologists to come to court to speak to their evidence and
for it to be taught to the fact finder, courts have nonetheless recently de-
veloped an overly cautious approach to the use of epidemiological evi-
dence, particularly in Scots law. We have seen two main reasons for judi-
cial scepticism about epidemiological evidence emerging from the case
law, namely the propriety of drawing causal inferences from observed as-
sociations (a general causation issue) and the propriety of drawing causal
188 Cranor, supra note 37 at 366; see also ibid at 28990, 335. Courts should not exclude
causal opinions based on non-epidemiological evidence where a body of epidemiological
data does not exist (David L Faigman et al, How Good is Good Enough?: Expert Evi-
dence Under Daubert and Kumho (2000) 50:3 Case W Res L Rev 645 at 663).
816 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
inferences in individual cases from concededly causal associations ob-
served in samples of populations (a specific causation issue). The concern
with taking such a cautious approach to epidemiological evidence is that
it may make it harder to discern that there is any reconciliation of the le-
gal standard of proof on a balance of probabilities with the scientific
standard of statistical significance. Moreover, there is also concern that
placing an obligation on a plaintiff to teach epidemiological analysis to a
court suggests that the court can remain passive in this process. This is
surely an unhelpful approach in cases such as McTear, where there is a
clear societal expectation that a judge will resolve these matters to the
satisfaction of both parties. Had the pursuer explained the epidemiologi-
cal evidence properly, and had Lord Nimmo Smith been more receptive to
evidence of relative risk, taken a more active role in forcing the pursuer to
present her evidence effectively, and given adequate weight to the gener-
ally accepted scientific evidence that cigarette smoking can cause lung
cancer, general causation could have been established in this case.
There also remains a lack of clarity on the extent to which generalized
epidemiological evidence can be useful in determining individual, or spe-
cific, causation. Accordingly, this paper supports the use of logistic regres-
sion techniques and other forms of statistical refining mechanisms using
specific risk factors to give quantitative or quasi-quantitative expression
to conclusions about the cause of disease in an individual drug product li-
ability claim that is based on epidemiological evidence. Logistic regression
is also closely linked to other forms of statistical refining such as Bayes
theorem. We have seen that while Bayes theorem can modify evaluations
of probability based on initial assumptions in light of more data using
specific factors embodied in the likelihood ratio, implementation of the
theorem, in practice, is likely to be difficult. It is important to stress that
Bayes theorem is not necessarily the answer to the problem of establish-
ing specific causation in the context of epidemiological evidence. However,
the crucial point is that statistical refining mechanisms using specific risk
factors can assist courts in determining specific causation in drug product
liability cases when the dominating evidence is epidemiological in nature.
This is likely to be increasingly true, as the quality of scientific evidence
increases with time.189
It has been suggested that this approach could have been adopted
with the specific case information available in Peterson, instead of the
plaintiffs personal circumstances being blindly treated as diminishing the
189 For support for a probabilistic model of specific causation in toxic torts, when the domi-
nating evidence comprises population-based data of the toxic effect, see especially Gold,
Certainty Dissolves, supra note 130 at 281, 30304, 33839.
EPIDEMIOLOGICAL UNCERTAINTY 817
strength of the epidemiological evidence.190 Indeed, the interdisciplinary
Vaccine Safety Committee of the Institute of Medicine adopted such an
informal Bayesian approach to assessing case reports in its review of
scientific and medical literature on specific risks to children associated
with vaccines.191 Courts could use this information to refine generalized
statistics to produce as accurate a posterior probability as possible, espe-
cially in the pharmaceutical field. This, however, would require epidemi-
ologists and physicians to assist courts in such an exercise, and clearly,
without courts having access to existing prior probabilities and the ability
to quantify likelihood ratios, the utility of the process would be limited.192
Notwithstanding the scepticism of the majority of the UK Supreme
Court in Sienkiewicz, there is little doubt that the use of epidemiological
evidence in medicinal product liability cases, especially where non-
numerical solutions are elusive, has now come of age. Albeit with caution,
courts are recognizing the importance of such evidence. The challenge is
now for lawyers and epidemiologists to come to some consensus as to what
amounts to a suitable use of epidemiological evidence in such cases when
establishing proof on a balance of probabilities. It is arguable that the so-
called doubling of risk approach mooted in Sienkiewicz is overly simplis-
tic. In particular, doubling of risk does not consider absolute risk (that is,
the risk of something occurring without any context)193 and the severity of
190 Peterson, supra note 17.
191 See Kathleen R Stratton, Cynthia J Howe & Richard B Johnston, Jr, eds, Adverse
Events Associated with Childhood Vaccines: Evidence bearing on Causality (Washing-
ton, DC: Division of Health Promotion and Disease Prevention, Institute of Medicine,
National Academy Press, 1994) at 25.
192 See Cranor, supra note 37 at 25659. For further discussion of the National Childhood
Vaccine Injury Compensation Program, see Goldberg, Causation and Risk in the Law of
Torts, supra note 2 at 16370.
193 Feldschreiber, Mulcahy, and Day provide a good illustration of the failure to take ac-
count of absolute risk: If there is an incidence of disease in an unexposed population of
one in a million cases and in an exposed population of two in a million cases, the RR is
two but the absolute risk is very low (Feldschreiber, Mulcahy & Day, supra note 19 at
188). The Federal Full Court of Australia observe in Peterson that [d]oubling a very
low absolute risk of an adverse result may produce an absolute risk which itself re-
mains so low that a positive finding of causation on the balance of probabilities would
itself be an affront to common sense (supra note 17 at para 119). However, I respectful-
ly submit that as a matter of statistics, this observation is incorrect. If one accepts the
premise of this paper that population-based estimates are relevant to causal conclu-
sions in individual cases, then doubling of risk is doubling of risk, irrespective of abso-
lute risk. One can concede the intuitive appeal of the courts statement. Thus, if in a
population of 100 million unexposed individuals, only one case of disease were expected,
who could submit that finding two cases represented anything other than a fluke?
However, that intuition is merely an illustration of the difficulty in obtaining statistical-
ly significant results in the epidemiological investigation of rare conditions. If there
were a way of designing an epidemiological study of sufficient quality, capable of identi-
818 (2014) 59:4 MCGILL LAW JOURNAL REVUE DE DROIT DE MCGILL
the outcome. Any attempt to reach a consensus in the future must ad-
dress these, and related, difficult issues.
fying an association that truly exists (i.e., one with sufficient power), it could be said
with great confidence that the exposure (generally) causes the disease. I am grateful to
an anonymous reviewer for this point.
