Liability for negligently increased risk: the repercussions of Barker v. Corus UK (plc)

Transcription

1 Law, Probability and Risk (2009) 8, Advance Access publication on April 1, 2009 doi: /lpr/mgp009 Liability for negligently increased risk: the repercussions of Barker v. Corus UK (plc) CHRIS MILLER Salford Centre of Legal Research, Salford University, Salford M5 4WT, UK [Received on 16 July 2008; revised on 4 February 2009; accepted on 12 February 2009] The philosophical problem of references classes is discussed within the context of probabilistic causation and, more particularly, liability for the negligent increase of the risk of personal injury. The implications of a recent addition to the U.K. case law on asbestos-related disease are then examined. Keywords: causation; reference classes; risk; probability; uncertainty; mesothelioma. 1. Introduction In this article, discussion of the role of probability in the courtroom tends to be dominated by analyses of identification techniques such as DNA profiling and fingerprint evidence in criminal trials. But civil cases can also oblige lawyers to confront probabilistic reasoning. Reference will be made below to a few of the articles which have contributed to the enormous literature resulting from the (predominantly U.S.) litigation concerned with asbestos, tobacco, defective drugs, medical negligence, etc. The associated accumulation of case law and academic commentary has, not surprisingly, influenced jurisdictions outside the USA, especially those with a common law tradition. Given this far wider context, the recent ruling in Barker 1 hardly demands attention since, at first sight, it appears to be little more than business left unfinished by the House of Lords when it decided a very similar asbestos case, Fairchild v. Glenhaven Funeral Services Ltd. 2 (Instead of the joint and several liability imposed on the defendants in Fairchild, each defendant in the conjoined appeals comprising Barker was required to shoulder the less onerous burden of several liability factored in proportion to his contribution to the overall risk, which eventuated in his particular claimant s mesothelioma.) Barker might therefore be interpreted as simply a matter of quantum and a short-lived one at that conscious of the numbers of mesothelioma victims whose damages could be reduced as a consequence of this ruling, the U.K. government quickly brought in legislation 3 to restore joint and several liability in mesothelioma cases. But a closer examination reveals that the Lords concession towards proportionate damages is not without wider repercussions for the vexed issue of causation in personal injury. Probabilistic causation is a subject which generates deep divisions among legal theorists. In turn these divisions have underpinned rival approaches to deciding civil cases where evidence on causation is beset with uncertainty. The trenchant c.e.miller@salford.ac.uk 1 Barker v. Corus (UK) plc (formerly Saint Gobain Pipelines plc); Murray v. British Shipbuilders (Hydrodynamics) Ltd; Patterson v. Smiths Dock Ltd and others [2006] UKHL Fairchild v. Glenhaven Funeral Services Ltd, Fox v. Spousal (Midlands) Ltd, Matthews v. Associated Portland Cement Manufacturers (1978) Ltd and others [2002] UKHL Compensation Act 2006, Chapter 29, s.3 Mesothelioma: damages. c The Author [2009]. Published by Oxford University Press. All rights reserved.

2 40 C. MILLER endorsement of liability 4 for increased risk by the (4:1) majority in Barker needs to be viewed within this context. This article begins with a brief account of the problem of references classes and its relevance to risk liability. A preoccupation with the reference classes problem has sometimes impeded a pragmatic solution to inherently uncertain causation. Barker can be seen as a recent addition to a pragmatic tradition in the English common law of personal injury. But it is suggested that there remains a confusion which the (English) Supreme Court may well be obliged to revisit in the future. It will further contend that this confusion stems largely from the Lords continuing policy aim of restricting liability for increased risk of injury to those instances in which a narrowly defined set of empirical circumstances (which they now term the Fairchild exception ) 5 obtains. The confusion is illustrated by examining a number of cases lying outside of asbestos litigation and involving only single defendants. The article further contends that the confusion could become more apparent once claims for increased risk involve a numerical estimate of that increase. (It will be argued below that it is not incidental that Fairchild and McGhee v. National Coal Board, 6 on which it relied, were not complicated by such estimates.) 2. The particularists objection In view of the clear statements made by Lords Hoffmann, 7 Scott 8 and Walker 9 (but emphatically not Lord Rodger) 10 in Barker that liability should attach to negligently increased risk, it is necessary to consider the arguments of the many academic analysts who are equally convinced that probabilistic causation (and risk is a probabilistic concept) is an error. Their concern is with verdicts which are derived not from information which is specific to the circumstances of the case being tried, but from statistics produced elsewhere, e.g. an epidemiological study which concludes that the toxic substance S is implicated in N% of the type of cancer suffered by the claimant. Richard Wright 11 is perhaps the most implacable opponent of using such naked statistics to establish causation: [n]aked statistics are not useful for causal explanation or causal prediction, since they are not related to any possibly applicable causal generalisation. 12 For him, what is required is particularistic 4 See below at n Above n. 1 at para 3. 6 [1972] 3 All ER the basis of liability is the wrongful creation of a risk or chance of causing the disease, the damage which the defendant should be regarded as having caused is the creation of such a risk or chance... n. 1 at para where each defendant, in breach of duty, has exposed the claimant to a significant risk of the eventual damage, [liability] should be... commensurate with the degree of risk for which that defendant was responsible. Ibid at para [Fairchild] was decided by the majority... not on the fictional basis that the defendants should be treated as having caused the claimant s... damage, but on the factual basis that they had wrongfully exposed him to the risk of damage. The damage was indivisible, but the risk was divisible a matter of statistics. Ibid at para In his minority opinion, Lord Rodger was no less insistent that... this adventure of redefining the nature of the damage suffered by the victims should be avoided, ibid at para 86. Interestingly, Lady Hale expressed her agreement with Lord Rodger and averred... that the damage which is the gist of these actions is the mesothelioma and its physical and financial consequences. It is not the risk of contracting mesothelioma. Ibid at para 120. But she nevertheless shared the majority view on the need for proportionate damages. 11 See Richard W. Wright, Causation in tort law, (1985) 73 Cal. L. Rev (Causal) and later his Causation, responsibility, risk, probability, naked statistics, and proof: pruning the bramble bush by clarifying the concepts, (1988) 73 Iowa L. Rev (Bramble 1) and then, Once more into the bramble bush: duty, causal contribution, and the extent of legal responsibility, (2001) 54 Vand. L. Rev (Bramble 2). 12 Above n. 11 (Bramble 1) at 1054.

3 LIABILITY FOR NEGLIGENTLY INCREASED RISK 41 evidence... determining what actually happened because only particularistic evidence can establish or negate the instantiation of the abstract elements in a possibly applicable causal generalisation and its underlying causal law. 13 Stephen Perry, another particularist, bases his opposition to liability for increased risk on an English medical negligence case, Hotson v. East Berkshire Area Health Authority. 14 Since we believe that the Lords ruling in that case made a key contribution to the confusion in personal injury law, we concentrate on Perry s defence of that ruling a defence which is based upon his understanding of the problem of reference classes. 15 To understand one conception of probability, it is necessary to imagine an indefinitely long series of coin tosses, and if the coin is true, then the fraction (1/2) of the total number of tosses that fell heads can be defined as the probability that any toss will result in heads. This is usually referred to as the frequentist version of objective probability objective because it relates to observable events in the real world and not to our beliefs as to the outcome of tossing a true coin. However, it is difficult to understand how we can attach a probability to a single instance: if I toss a coin, either it comes down heads (p(h) = 1) or tails (p(h) = 0), and even if I cannot see the outcome, can I still think of either outcome as having a probability p(h) = p(t ) = 1/2? This question formed the gist of the (subsequently much cited) speech of Lord Mackay in Hotson. On his first admission to hospital, the actual state of the blood vessels in Hotson s fractured hip were such that either he could benefit from prompt treatment (p = 1) or he could not (p = 0). But at first instance (or even earlier) there was no means of telling which category Hotson came within. Imagine (as counsel for the claimant urged their Lordships to do) there are 100 patients who have injured themselves in a manner similar to Hotson: of these, 25 recover when given prompt and appropriate treatment and the remaining 75 do not. In the absence of any further information, 25% is the (frequentist) probability that Hotson might be a member of the minority who can benefit from early diagnosis and treatment. But since the delay in diagnosis robbed Hotson of this 25% chance of avoiding injury, it is this lost chance which, his counsel argued, should be compensated. Perry rejects that idea: suppose that there was some diagnostic test which made it... possible to partition the reference class of 100 persons into the 25 who were treatable and the 75 who were not. 16 The test would be applied to any Hotson-like claimant: if he proves treatable, he secures 100% damages; if the test shows him to be untreatable, he receives nothing. This seems uncontestable; but Perry then makes the curious assertion: [b]ut the claim that the plaintiff suffered risk damage must surely be 13 Ibid. 14 [1987] AC 750. Thirteen-year-old Stephen Hotson fell from a tree injuring his hip, but on first admission to hospital, he was negligently misdiagnosed has having damaged his knee and was sent home. After a week of increasing pain, he returned to hospital whereupon the full extent of his hip injury (avascular necrosis of the neck of the femur) was recognized. The resulting disability could be said to have two causes: (primary) the loss of blood vessels ruptured as an immediate consequence of the fall and (secondary) those lost when the initial bleeding into the hip joint then compressed and blocked some of the vessels that had survived the fall. As in McGhee (above n. 6) the medical witnesses could not reach a consensus: the defendant s argued that the primary loss was alone sufficient to cause the ultimate necrosis; the claimant s maintained that it required those lost after the negligent diagnosis (and hence the failure to offer prompt treatment) for the full injury to arise. The judge at first instance (the then, Mr Justice Simon Brown) took the average of the two rival estimates (0% and 50%) of the probability of avoiding the full harm given prompt diagnosis and treatment and awarded 25% damages. When the case eventually reached the House of Lords, it was decided that because this estimate of increased risk was less than 50%, he could not recover. To succeed, it was argued, a claimant must adduce evidence of a risk increase in excess of 50% because only then does she satisfy the balance of probabilities (BOPs) standard of proof that traditionally applies in civil disputes. 15 S. Perry, Risk, harm, and responsibility in D.G. Owen (Ed) Philosophical Foundations of Tort Law (Oxford: OUP, 1995). 16 Ibid, 334.

4 42 C. MILLER independent of knowledge of this kind. 17 If this knowledge exists, there is no need to use risk, chance or any other probabilistic language. And any civil suit (if not settled out of court) would simply weigh in the balance of probabilities (BOP) the claimant s account of the medical negligence and his evidence that he is indeed a member of that partition (of the reference group) which can benefit from treatment. It is not difficult to partition the reference group of workers exposed to industrial dust into two: those who can avoid dermatitis by showering immediately after coming off shift and those for whom dermatitis is inevitable whether the removal of the offending dust is timely or delayed. But when the House of Lords confronted this issue in the seminal case of McGhee, 18 it was clear from the evidence of the rival expert witnesses (consultant dermatologists) that a diagnostic test, which could affect this partition, did not then exist. Quite simply, it was impossible to state categorically that McGhee (or anyone working in such conditions) could or could not have benefitted from prompt showering had it been available. However, the negligent failure to offer a shower at the workplace led to a delay in washing, and they could agree that such a delay increased the risk of the dermatitis, which subsequently arose. The House then decided that liability should attach to this negligent increase in risk of injury. This in essence is the rule in McGhee its application will be discussed further below. Perry invents the term risk damage and then seeks to show that such damage cannot be compensable in tort. But within this term he includes, and contemplates civil liability for, risks which do not eventuate in actual harm. He sees risk damage as harm in itself and distinct from the physical damage (in Hotson, the avascular necrosis). 19 If risk damage can stem, as Perry believes it can, 20 from risks which do not materialize, the use of the word distinct in this context may be meaningful. Perry recognizes that where harm is readily computable (such as the number of fatalities caused by earthquakes or in industrial accidents) and where the probability distribution of the harm is known, the risk can be expressed as the expectation value of the harm. 21 However, it is hard to see injury and its expectation value as distinct concepts. By confining attention (and liability) strictly to eventuated risks, the risk and the injury, although distinguishable concepts, remain intimately tied together. By analogy: I have a body and that body has weight; they are separate but closely related concepts, but it is not meaningful to speak of my body weight as being something distinct from my body. Perry s principal objection to civil liability for what he terms risk damage relates to the problem of multiple partitions of the reference class. Imagine now that there is no diagnostic test and that the (25:75) partition of the hypothetical 100 patients is made a posteriori, namely, based on the hospital records: all 100 received prompt treatment but the records show that only 25 were able, in the event, to avoid disability. Imagine further that records also revealed that age had a perceptible effect on 17 Ibid. 18 Above n Above n. 15, It seems clear, moreover, that the status of agent-imposed risk as harm in itself should not depend on whether threatened physical harm has materialised, ibid. It must be stressed that the present article is concerned exclusively with risks which have materialized in actual injury. The question of civil liability for pure risk, where injury is threatened but through good fortune is avoided, is at best a diversion and one which has been of more interest to academics (see for instance, G. Robinson, Probabilistic causation and compensation for tortious risk, (1985) 14 J. Leg. Stud. 779 and J.G. Fleming, Probabilistic causation in tort law, (1989) 68 La Revue du Barreau Canadien 661) than to judges. Quite apart from near misses, uncertainty over a past event is of a different order from uncertainty over the future; therefore it is also advisable to avoid reference to cases which involve the (negligently increased) risk of feared, but as yet, uneventuated harm, of which Gregg v. Scott [2005] UKHL 2 is of most relevance here, having been quoted extensively by the House in Barker. For a critical analysis of Gregg, see C. Miller, Gregg v Scott: loss of chance revisited, (2005) 4 Law, Probability and Risk a risk is the mathematical expectation of the harm... were it to occur.. Above n. 15, 322.

5 LIABILITY FOR NEGLIGENTLY INCREASED RISK 43 FIG. 1. Partitioning the reference class. the chance of recovery (see Fig. 1): among those who did benefit, 20 were found to be aged 14 or younger. Stephen Hotson, aged 13 at the time of his accident, now belongs to the reference class of under-fourteens and hence his chance of benefiting from prompt treatment is now 20/35 or 57%. Since this figure is greater than 50%, or more likely than not, an orthodox approach to causation would now suggest that he should receive full damages. But it is not difficult to imagine further partitions (e.g. gender, ethnicity, social class). If the hospital records also show that, within those 20 under-fourteens who did benefit, girls formed a clear (16:4 in Fig. 1) majority, a 13-year-old boy like Stephen Hotson now finds himself in the reference group which has a probability of benefiting of 4/12 (i.e. <50%), so now the orthodoxy would deny him compensation. If the partition produces a zero in either cell, then this is equivalent to Perry s hypothetical diagnostic test. Failing that outcome, it is this oscillation in the estimate of the risk damage (and the associated verdict) which persuades Perry that risk damage can never be compensable. If, instead of oscillating, further partitions lead to no change in the proportions in each cell, we have what Papineau 22 labels a homogeneous partition of the reference class and that suggests that all relevant variables have been taken into account. However, the possibility that some relevant variable has been overlooked can never be excluded. In addition, the greater the number of partitions, the smaller the numbers in each cell which, in turn, increases the confidence interval on the risk estimate. Nevertheless, it is possible to hypothesize an epidemiological study over a sufficiently large population to overcome these objections, but the question still remains: should a remedy in tort be confined to those claimants who could take advantage of the results of such a study? In the absence of a study boasting high statistical power, the fact-finder must look to the narrowest 23 partition that employs all the plausibly relevant information available. If a young, male claimant in our hypothetical case had sought to rely on the first partition, namely, age and the favourable 20/35 probability, an alert defendant would insist that the fact-finder should also consider the data 22 See David Papineau, Probabilities and causes, (1985) 82 Journal of Philosophy 57 at As suggested by Hans Reichenbach, The Theory of Probability (1949, University of California Press, Berkeley).

6 44 C. MILLER on the gender partition. More generally, if a fact-finder suspects that an irrelevant variable had been employed to produce a partition which then gave the claimant a favourable probability (that her injury was caused by defendant negligence), then he can take that into account when gauging the credibility of the claimant s overall account. 24 As a child answers the doctor s successive questions about the nature of his stomach ache, the more satisfied she becomes that it is attributable to eating unripe apples rather than to acute appendicitis. The emergence, shortly after an innocent man has gone to the gallows, of the information which confirms his alibi has not been confined to the world of fiction. All verdicts depend on information: information reduces uncertainty and probability is a measure of uncertainty. But epidemiological evidence where every effort is (or should be) made to establish and justify the minimum inference that the uncertainty will allow would be excluded by the particularists. Risk is the product of probability and harm, and the oscillation which disturbs Perry is in the probability not in the harm. But he has invented this further notion of risk damage which obscures this difference. Curiously he is prepared to concede that there is a very plausible argument that the lost opportunity [his italics] to receive proper medical treatment is a distinct form of harm in its own right. 25 He continues: [t]he 25 lost chance... might then be regarded, not as damage in itself, but as evidence bearing on the valuation of the lost opportunity. 26 More recently, Perry has averred that [t]he idea that this kind of lost chance is harm, and the idea that imposing an increased risk of harm is likewise harm, are morally and conceptually equivalent. 27 Richard Wright also accepts liability when tortious conduct reduced the victim s chance of avoiding an injury that has actually occurred, 28 with damages factored by the estimate of reduced chance. But he stipulates that whatever it is that suffers a tortiously reduced chance must be recognized as a legally protected interest, 29 and avoidance of further injury, by seeking medical treatment, must surely qualify as such an interest. Thus, Perry and Wright appear to approve of Hotson leaving court with 25% damages for the negligent loss of an opportunity to avoid his disability, but they are opposed to such an award if it is preceded by an argument couched in terms of increased risk. But surely opportunity, if not exactly a synonym of chance, is still a probabilistic term. If I am selected for the England cricket team, I have been given an opportunity to score a century in my debut test match; if I am excluded, that opportunity goes to someone else whether I use the word opportunity, chance or possibility seems a matter of linguistic choice. 30 But whatever term is employed, if it were denied for me because of an injury resulting from medical negligence, my sense of grievance would be palpable and the loss of this opportunity would undoubtedly diminish my well-being. Exactly why Perry believes that Hotson s... lost chance of avoiding avascular necrosis.... cannot be construed as an aspect of the well-being of an individual is not made clear. 24 The point that BOP is the criterion against which the fact-finder judges the credibility of the claimant s overall account of causation is pursued further below, see text at n Above n. 15 at Ibid. 27 S. Perry, Risk, harm, interests, and rights in Tim Lewens (Ed) Risk: Theoretical Perspectives (2007, Routledge) endnote 10, p Above n. 11 (Causal), Ibid. 30 The word risk tends to be applied to negative outcomes, e.g. earthquakes, accidents and avalanches. To refer to the risk of winning the lottery would sound odd to most English speakers. Whereas chance, likelihood and, even more so, probability are outcome-neutral terms. 31 Above n. 15, 335.

7 LIABILITY FOR NEGLIGENTLY INCREASED RISK 45 In a recent article, 32 Allen and Pardo argue that the problem of reference classes limits the role of mathematical models in illuminating the probative value of legal evidence. These limitations throw doubt across the wide range of legal concerns, from DNA profiling to toxic tort and others involving epidemiology, which rely heavily on statistical evidence. Their illustration of oscillating probability estimates with different partitions of the reference class involves the famous blue bus hypothetical : 33 the victim of a road traffic accident believes he was knocked down by a bus but is unsure of its colour; the Blue Bus Company owns 75% of the buses in the town, but only 25% of the buses in the county and 75% in the state. The choice of reference group (town, county or state) clearly affects the a priori likelihood that a bus owned by the Blue Company was responsible for this accident. Tillers has observed that when such general-to-individual inferences amount to racial stereotyping, they tend to be abjured on moral grounds. But Tillers 34 also points out that attempting to remove, from the judicial process, all inferences from statistics is doomed to fail. In particular, we can never evaluate the behaviour of an individual without calling upon prior knowledge of similar behaviour in other people. How, for example, can we judge the behaviour of a mother who has harmed her week-old baby without referring to our accumulated knowledge of puerperal psychosis? It is far easier to insist upon specific or particularistic evidence than it is to define it or to demonstrate that any inference we draw from it is entirely untainted by assumptions stemming from prior experience of similar instances of that evidence. This is not the place to attempt a comprehensive review of the reference class problem. 35 The first aim of the second part of this article is to show that the problem has not proven a barrier to the English common law attaching civil liability to negligent acts or omissions which have increased the risk of a harm which has indeed eventuated. If a mesothelioma victim belongs to the reference class of employees in firms where asbestos was negligently controlled, then that is sufficient to secure him or her compensation. In addition, the claimant in the leading English case involving epidemiological evidence (on vaccine damage) failed to secure compensation, not because she relied on statistical evidence on causation, but because the evidence was insufficiently robust. 36 The path to Barker may not have been blocked by philosophical objections, but it was still very tortuous. Below we point to three cases (one Scottish and two from England) which might, and arguably should, have been decided differently, had the respective judges not shown an antipathy to statistical evidence. It is therefore neither surprising nor coincidental that progress has come from cases in which the claimants did not venture a numerical estimate of the increased (and eventuated) risk to which they were exposed. The second aim of the remainder of this article is to explain why the issue of liability for increased risk in English common law remains far from resolved. 32 R. Allen and M. Pardo, The problematic value of mathematical models of evidence, (2007) 36 Journal of Legal Studies M. Finkelstein and W. Fairley, A Bayesian approach to identification evidence, (1970) 83 Harvard Law Review Any inferential strategy that purports to abjure all inferences about an individual that rest on events or conditions that are not generated by the decisions or behaviour of that individual is bankrupt. P. Tillers, If wishes were horses: discursive comments on attempts to prevent. Individuals from being unfairly burdened by their reference classes (2005) 4 Law, Probability and Risk 33, See M. Pardo, Reference classes and legal evidence (2007) 11 IJ of Evidence & Proof 255 and the other articles in this special issue. 36 Loveday v. Renton [1990] 1 Med LR 117. She failed because the estimate of her increased risk was not statistically significant at the 95% level; in other words, the chance that there was no increase was greater than 1 in 20.

8 46 C. MILLER 3. The Fairchild exception When coining the term Fairchild exception and labelling McGhee as an application avant la lettre, 37 Lord Hoffmann perhaps had the benefit of having read Jane Stapleton s case analysis 38 in which she listed six characteristics which are sufficiently common to these two rulings that it is possible to speak of a McGhee/Fairchild principle. The principle specifies the circumstances in which a claimant can leap an evidentiary gap and secure damages against a defendant who increases the risk of an injury which then arises. The language of risk is employed because uncertainty in the evidence prevents an identification of the precise causal connection between the negligent act or omission and the eventuated injury which the common law has traditionally required. This article will concentrate on two of Stapleton s six characteristics: 39 the (substantial) similarity of the two or more agents of risk and the extent to which the sources of risk are under the control of the defendants. Thus in Fig. 2, only cases (cell A) where both these conditions are met can qualify for the Fairchild exception. McGhee is the founder member of cell A: the dust encountered non-negligently during the working shift and the dust which, through the employer s negligent omission to provide showers, remained upon claimant James McGhee s skin were identical (more than substantially similar) and were both under the control of his employer. Fitting the circumstances of Fairchild into the McGhee mould required a good deal of judicial ingenuity. The key issues were the assumption that all cases of mesothelioma require at least some small exposure to asbestos fibres and the far less defensible assumption that, for these claimants, their exposures arose only from the neglect of appropriate control procedures by their various employers. As Stapleton is at pains to point out, 40 given the number of people known to have died of mesothelioma but who never worked in a setting where asbestos was obviously present, environmental exposure can rarely be excluded in Britain or in any comparable industrial society. The claim of Wilsher v. Essex Area Health Authority 41 to be the definitive member of cell D (in Fig. 2) is clear: there were five, dissimilar, possible causes of the claimant s blindness; four stemmed from hazards inherent in premature birth and it was a failure adequately to regulate the air supply that constituted the (medical negligence) fifth. It was not possible to ascertain which of these five possible causes was actually responsible, and even though the tort could be interpreted as increasing the risk of an injury which did indeed eventuate, the shape 42 of this increased risk was not the shape which the Fairchild exception has subsequently come to recognize. 37 Above n. 1 at para J. Stapleton, Lords a leaping evidentiary gaps (2002) 10 Torts Law Journal Essentially, she distils Lord Bingham s six (above n. 2 at para 2) key issues which could make Fairchild a special case alongside the five significant features of that case as listed by Lord Hoffmann (above n. 2 at para 61). Stapleton s list of the same characteristics includes the acceptance that all relevant exposures to risk occurred in the victims workplace (above n. 38 at 285). We shall interpret that requirement as all sources of risk being under the control [my emphasis] of the defendant, as Stapleton does (Ibid, 287). It is not possible to exclude the inference that the Fairchild exception, as currently formulated, can benefit only victims of occupational injury. In Barker (above n. 1 at para 114), Lord Walker is even more restrictive and argues that the principle must... be restricted to mesothelioma induced by inhalation of asbestos fibres, and other conditions having the same distinctive aetiology and prognosis (such as the dermatitis caused by brick dust in McGhee... ). This very restrictive stance is undermined by his strange belief that mesothelioma and dermatitis share a distinctive aetiology and prognosis. 40 Above n. 38 at [1988] 1 AC Shape is the term Lord Justice Mustill used in the Court of Appeal to denote the particular circumstances which led to increased (and eventuated) risk. And before applying McGhee, he opined that the shape taken by the enhancement of the risk ought not to be of crucial significance. Wilsher v. Essex Area Health Authority [1987] QB 730, 772.

9 LIABILITY FOR NEGLIGENTLY INCREASED RISK 47 FIG. 2. Two conditions of the Fairchild exception. The problem of dissimilar agents of risks but where all are under the defendant s control (cell B) was raised but not answered by Lord Hoffmann in Fairchild. 43 Imagine a rare form of cancer known to be caused only by two different toxic substances X and Y ; an employee is negligently exposed to both at work and contracts this particular cancer. Does the burden of proof include having to demonstrate which agent, X or Y, was the actual cause on this occasion? With the hindsight of the sequence of litigation ending (for the moment) in Barker, it is difficult to see the courts withholding liability simply because there is no biopsy test which can distinguish between an X-induced cancer and one attributable to Y, when it is not disputed that but for employer negligence the cancer would not have arisen. The position is less clear when (cell C in Fig. 2) there are multiple sources of similar or identical risk agents but not all are under the control of identifiable defendants. If someone fell victim to a disease of which a certain toxic substance is the only cause currently known to science, she would be hard put to secure damages if both her favourite leisure pursuit and her employment entailed regular exposure to that substance. It is not clear whether a British equivalent of the Californian case of Sindell v Abbott Labs 44 damages for injuries caused by a defective drug were divided between various companies in proportion to the their respective shares of the U.S. market for that particular drug would fall within cell A or cell C. Claimants would have to identify, and be able to sue, those companies whose aggregate share of the U.K. market could be said to amount to control. Curiously, Barker could itself be interpreted as a departure from the control (negligently exercised by the defendants over asbestos exposures) in Fairchild some of the exposure was under the control of the claimant during his period of self-employment. We now consider three, evidentially very different, personal injury cases which illustrate the prevailing confusion in the law of causation. 3.1 Medical negligence: Hotson v East Berkshire Area Health Authority 45 The conceptual robustness of the Fairchild exception can usefully be explored by revisiting Hotson which, Lord Scott averred in Barker, 46 had been correctly decided in 1987 when the Lords denied the claimant the benefit of the rule in McGhee. 43 Above n. 2 at para Cal.3d Above n Above n. 1 at 64. His reasoning appears to be concerned more with single agent than with control ; he believes that the single agent requirement was not met in Hotson.

10 48 C. MILLER Reference has already been made to the influential speech which Lord Mackay made in Hotson. There he made reference to certain passages in the U.S. medical negligence case, Herskovits v. Group Health Cooperative of Puget Sound, 47 which still serves to expose and clarify the key issues in probabilistic causation. In the minority opinion of Brachtenbach J, Lord Mackay found support for his antipathy to naked statistics. In particular, he cited Brachtenbach J s example of the town with two cab companies: one operated three cabs painted blue and the other one cab painted yellow. In the event of an injury caused by an unidentified cab, the blue company, with its 75% probability of being the culprit, stands to incur liability (on BOP) each time. For Brachtenbach (and by implication Lord Mackay), this probability was both relevant and admissible but it was not sufficient to prove that the blue cab company more probably than not committed the act. 48 In Hotson, Lord Mackay did not refer to the majority opinion in Herskovits. But this was no less germane: there was a consensus among the medical experts that the (6-month) negligent delay in diagnosis had reduced the plaintiff s chance of surviving lung cancer from 39% to 25%. And the 3/2 majority held that this reduction in chance was sufficient evidence of causation to go before the jury. The defendant doctor (who prescribed cough medicine for Herskovits coughing and chest pains) was clearly not responsible for whatever had initiated the tumour well before the first consultation. Unlike the circumstances which obtained in McGhee, control passed to the defendant only after his negligence (namely, misdiagnosis). (But would the single-agent criterion have been satisfied in Herskovits? The case report does not go into details but it could well be that, during the 6-month delay, there arose other consequences (metastases, for example,) of the original cancer, which increased the probability that it would (as it did) become terminal.) The language of risk was used in Hotson simply because of scientific uncertainty: in the absence of a more precise account of the aetiology of the injury, 25% represented the best (albeit crude) estimate of the increased risk of injury attributable to the defendant s negligence. So why did not Hotson receive full damages under the rule in McGhee? Although this was not explicitly discussed by the Lords in Hotson the single-agent criterion was first articulated 15 years later in Fairchild it could be that the two mechanisms 49 (one ante-tort and the other post-tort) which damaged blood vessels were deemed too dissimilar to satisfy what was then a tacitly understood condition of liability. To the layperson, a blood vessel not functioning as nature intended is simply that regardless of the origin of the non-function. But in an adversarial system, rival medical experts have an incentive to stress the differences that exist between what, to those without medical training, appear to be similar pathological states. 50 This dilemma was not absent in McGhee but there the House was resolved that this form of evidential uncertainty (dissensus among medical witnesses) should not become a barrier to the claimant. But what of control? In McGhee, the defendant exercised complete control (over the workplace dust which was undoubtedly the single agent of dermatitis). That degree of control is not present in a negligent diagnosis case like Herskovits or Hotson. In the latter, the hospital s negligence meant that they did not recognize and therefore failed to control (or, rather, intervene and hence forestall) the secondary loss of blood vessels. But they had no control over the primary loss (i.e. that occurring between the P.2d Ibid, 490. Like Allen and Pardo (above n. 32), Brachtenbach J relied on Finkelstein and Fairley (above n. 33) for his blue cab illustration. 49 The facts of the case are summarized above in n I explore this question of who decides what s similar in more depth in Causation in personal injury: legal or epidemiological common sense, (2006) 26 LS 544.

11 LIABILITY FOR NEGLIGENTLY INCREASED RISK 49 fall and the first admission). The following scenario endows Hotson with a degree of defendant control comparable with that which obtained in McGhee. Imagine that a young Hotson2 is about to be discharged from hospital following a minor operation (such as tonsillectomy). Bored with waiting for his parents to collect him, he goes into the grounds to find a tree to climb he falls and is taken to the casualty department of this hospital, whereupon history repeats itself. At the trial, it is accepted that the fall did not entail any breach of duty by the hospital and thus the primary loss of blood vessels was (again) non-tortious. But since the fall occurred in hospital grounds, the control has arguably been restored and so, assuming his counsel successfully counters any similar agent challenge, Hotson2 should now benefit from the Fairchild exception (although post-barker, his damages should be factored by 25% or any other estimate of increased risk agreed at first instance). Should the location or ownership of a tree be a material issue in a medical negligence case? And recall: the ownership and location of the tree (as well as the similarity between two sets of defunct blood vessels) only become an issue when the estimate of risk (increased by the negligently late diagnosis) is either absent or less than 50%. 3.2 Medical negligence: the vcjd straddlers litigation It is possible to point to litigation, beginning in 1997 in the High Court, 51 in which certain claimants in an increased risk case failed to recover, even though the single-agent and defendant control requirements were as clearly met as in McGhee itself. A number of children with a growth disorder were being treated, under the aegis of the Medical Research Council (MRC), with human growth hormone prepared from pituitary glands removed from human corpses. Once it became apparent that this procedure exposed these patients to the risk of new variant Creutzfeldt Jakob Disease (vcjd), it was decided that this form of treatment should cease. But the MRC was negligently slow in advising the doctors administering the series of injections to the various patients to discontinue. In the subsequent litigation by those who had contracted vcjd (or incurred psychiatric harm from the fear of incurring this invariably fatal condition), a claimant fell into one of four distinct groups: group A, where treatment only began after the threshold date (TD) when, it was accepted in court, that the warning to discontinue should have been heeded; group B, whose treatment was completed before that date and two groups S1 and S2 whose treatment straddled the TD. Group A, whose treatment lay entirely in the tortious period recovered full damages, and for equally uncontroversial reasons group B received nothing. S1 recovered because, with more than half their treatment coming after the date at which continued treatment became tortious, the probability that negligent treatment caused their injury was >0.5, whereas for S2, the equivalent probability was <0.5 and this group 52 received nothing. For both straddler groups, the tortious (post-td) and non-tortious (ante-td) injections of contaminated hormone were identical and therefore constituted a single agent of risk and there was no doubt that all the injections were under the control of the defendant (MRC). So why could S2 not enjoy the benefit of the rule in McGhee? When the Court of Appeal considered the straddler 51 Newman and others v. Secretary of State for Health (No. 1) 54 BMLR 8, which was partially (the straddlers issue) overturned by Newman and others v. Secretary of State for Health (No. 1) 54 BMLR 85. Some of the subsequent reconsideration is reported in Newman and others v. Secretary of State for Health (No. 2) 54 BMLR It is reported that two claimants were denied compensation on these grounds; Mark Mildred The Human Growth Hormone (Creutzfeldt Jakob Disease) Litigation [1998] JPIL 251, 256.

12 50 C. MILLER issue, McGhee does not appear, from the report, 53 to have been cited. And when the case was subsequently reconsidered by the High Court, Mr Justice Morland made no direct reference to McGhee 54. However, he does quote at length part of the speech of Lord Bridge in Hotson, where it is argued that McGhee provided no justification for departing from the orthodox by discounting damages in proportion to the estimated chance that avascular necrosis might well still have developed even if prompt treatment had been given. 55 And later Morland J cited a passage from the speech of Lord Ackner (again in Hotson) 56 in which Simon Brown J was criticized for his earlier decision to discount damages for a non-tortious contribution (5%) to an eventuated risk a decision which, with the hindsight of Barker, now seems uncontroversial. It is hardly surprising that the ruling in McGhee did not engage with the question of discounting damages in line with an estimate of the counterfactual chance of avoiding injury no such estimate was ventured in that case. When a numerical estimate of risk has been offered (in the vcjd litigation and earlier in Hotson), a comparison of this estimate against the 50% yardstick of BOP is treated as the end of the matter. No further thought is given to whether a (p < 50%) claimant might still be eligible to benefit from the rule in McGhee (as it was understood in 1997, namely, after Wilsher but before Fairchild). But in the vcjd case, at least the S1 group did receive damages. The claimant in McTear v. Imperial Tobacco Industries 57 did not; an obiter remark made by Lord Nimmo Smith in this case points to a deeper confusion over and above the difficulties inherent in the notions of control and similar agents of risk. 3.3 Tobacco litigation: McTear v Imperial Tobacco Industries When considering epidemiological evidence in a 2002 defective drug trial, 58 the High Court interpreted the BOP standard of proof as requiring the claimant to adduce evidence of a tortiously increased risk >0.5. If tortious increase in risk tortious increase in risk + background risk > 0.5, then it is more likely than not that it was the tortious, rather than the background, risk which eventuated in the harm suffered by the claimant. A probability of p > 0.5 is equivalent to the epidemiologists relative risk (RR) being greater than 2.0; but the RR of the lung cancer suffered by Mr McTear and others with a comparable tobacco dependence is, on the basis of the accepted epidemiological evidence, RR > 9.0. Once a claimant can point to an RR >2.0, and this statistical association is taken as reflecting an underlying causal mechanism, 59 BOP is exceeded and the claimant has discharged 53 Above n The Creutzfeldt-Jacob Disease Litigation, Straddlers Groups A and C v. Secretary of State for Health 54 BMLR 104, Above n. 14 at The decision of Simon Brown J in the subsequent case of Bagley v. North Herts Health Authority (1986) NLJ 1014, in which he discounted an award for a stillbirth because there was a 5% risk that the plaintiff would have had a stillborn child even if the hospital had not been negligent, was clearly wrong. Ibid at [2005] CSOH 69. The principal reason for his rejection of McTear s claim was that he could not accept, on the evidence presented in court, the general thesis that smoking increases the risk of lung cancer. 58 XYZ v. Schering Health Care Ltd. [2002] EWHC In the UK, the attribution of a causal status to an observed statistical association usually entails reference to the Bradford Hill criteria ; for a discussion, see C. Miller Causation in personal injury: legal or epidemiological common sense?, above n. 50.

13 LIABILITY FOR NEGLIGENTLY INCREASED RISK 51 her burden of proof. And there should be no need for her to establish her right to enjoy the Fairchild exception. But Lord Nimmo Smith was anxious to avoid the fallacy [sic] of applying statistical probability to individual causation 60 and he argued that it was still necessary for the claimant to demonstrate that his was not one of the (1 in 10) minority of lung cancers which are believed to be unrelated to smoking. He quoted at length from various speeches in Fairchild and McGhee and he concluded that the action before him fell outside the remit of those cases and therefore the but for test of causation was not suspended. 61 But the fact-finder in a civil trial is not required to be totally convinced that but for the tortious act or omission, the harm would not have arisen; they need only be satisfied on the BOP and that, on the authority of the leading English case 62 relying on epidemiological evidence in defective drug litigation, is equivalent to requiring RR > 2.0 in cases where epidemiologists offer evidence of the cause of injury. There may indeed have been perfectly sound reasons for dismissing McTear s claim; the fact that there was a 10% chance that his cancer was not caused by smoking cannot be one of them. Perhaps Lord Nimmo Smith s (obiter) remark was an aberration; if so, where did it originate? It stemmed from the suspicion that has attached to the rule in McGhee since the Lords decision in Wilsher. Notwithstanding all the recent efforts of the most eminent British judges first to restore McGhee and then to restrict its application, they have yet to give explicit advice on what would happen if a McGhee-like claimant offers a numerical estimate (<50%) of his increased risk. 4. The balance of (and other) probabilities Lord Hoffmann, in Barker, declares his wish to avoid the fictions employed in earlier discussions of risk and causation and he now clearly recognizes that in... a case in which science can deal only in probabilities, the law should accept that position and attribute liability according to probabilities. 63 However, the law must also recognize that the probability that quantifies an increased risk of an eventuated injury is not necessarily the one that is weighed in the BOP. BOP relates to the credibility or the degree of belief that the fact-finder ascribes to the claimant s overall attempt to attribute her harm to the defendant s negligence. If there is uncertainty in the scientific evidence, that attribution may have to be expressed in terms of risk, chance, odds, likelihood or other probabilistic terminology. But the fact-finder must still gauge the overall credibility of that account thus, he may be 90% sure that the increased risk of the claimant s injury is less than 30%. 64 The philosopher David Lewis has explained why it is meaningful to speak of credence in chance ; 65 in other words, we can have a subjective or epistemic degree of belief in an objective probability. By objective probability, I mean a measure of the relative frequency of a series of repeated events (typically, the tossing of a coin). Actuarial data perhaps provide the best illustration: from the observed age at death of a large number of white, U.K. males, it is possible to derive the 60 Above n. 57 at and the word fallacious appears in the Conclusions (on individual causation) at The present case did not fall outside the type of case to which the normal but for test was applicable. Insofar as it might in any sense be a special case, it fell within the reasoning of Wilsher. Above n. 57 at para Above n Above n. 1 at para 43; compare this view with the comment of Lord Nimmo Smith s epidemiology cannot provide information on the likelihood that an exposure caused an individual s condition, above n. 57 at Within the vast literature on causation, perhaps the first to emphasize the need for a clear distinction between BOP and any estimate of increased risk was the article by S. Gold Causation in toxic torts: burdens of proof, standards of persuasion, and statistical evidence (1986) 96 Yale LJ D. Lewis, Philosophical Papers Volume II (1986, Oxford UP) 83.