Causality, Probability, and Medicine

I was delighted to discover this new book by Donald Gillies. I am generally interested in and sympathetic to Gillies’s philosophical perspectives, and the topic of causality and probability in medicine is close to my heart. So with this as my very positive starting point, I read the book cover to cover.

I will share some of the premises that Gillies sets for the aims and scope of discussion. By being so explicit early on, he helps the reader understand what to expect and what he asks us to accept without argument. I found this useful because it also gave me a clear idea of what would not be covered.

His first qualification concerns the scope of causality. He does not deal with all types of causality, but only with general causal claims of the form ‘A causes B’. General causality covers many cases, while single causal claims only covers one case: a caused b. So ‘causality’ refers only to general causality (p. 1).

Second, he specifies the scope of ‘medicine’, as the restriction to general causal claims necessarily affects the scope. In theoretical medicine, one is mainly interested in such general causal claims, while the focus in clinical medicine is on single patients, symptoms, causes, and treatments. In this book, ‘medicine’ refers to theoretical medicine, not clinical medicine (p. 1).

Third, a distinction is made between two types of causality. Deterministic causality here means that the cause is sufficient for the effect (p. 199), so that whenever A occurs, it is followed by B, ceteris paribus (p. 2). This is the classical notion of causality. Indeterministic causality, in contrast, is when the cause is not always followed by B. Instead, B might follow A in a certain percentage of cases. Smoking causes lung cancer, for instance, but only in 5%. So it might be, Gillies notes, that most medical causes would fall under the indeterministic cases (pp. 2–3). Although the division between deterministic and indeterministic causality is quite common in philosophy, I suspect that some would think of the latter as an epistemological rather than ontological matter. For instance, it might be that we lack the knowledge about the ceteris paribus conditions under which lung cancer is produced in 100% of the smokers. My own take is to see causality as tendencies that are neither deterministic nor indeterministic, but I suspect that Gillies’s position requires far less work to convince the general reader.

Fourth, he makes a commitment to Russell’s thesis. I found it interesting to see that Gillies explicitly accepts Russell’s ([1994]) claim that causality is not relevant for theoretical physics, with causal laws replaced by non-causal mathematical laws (p. 21). But recently, he says, Jon Williamson convinced him that this thesis does not hold generally for all sciences and definitely not for medicine (p. 21). Medicine, he says, is all about causality, attempting to find the causes of disease and to develop ways to prevent or treat disease (pp. 20–1). This book, therefore, considers causality in medicine, not science in general.

I would probably have preferred it if Gillies had made fewer qualifications, opting instead for a more uncompromising line: discussing all of causality and all of medicine, with no separation between deterministic and indeterministic causality, and a full rejection of Russell’s thesis. But I can also appreciate that in being explicit about what he chooses to discuss and what he simply assumes, he saves a great deal of time and space that would otherwise be spent on internal philosophical debate. This allows him to direct his focus on the practical application of and implications of his theory to medicine. A clear strength of the book is exactly that the philosophical points are continuously illustrated with a number of detailed examples from medicine. The examples are also used to challenge or support philosophical positions, thus avoiding the philosophical pitfall of getting lost in abstractions. This is important, especially since the book is targeted both at philosophers and at readers with a background in the health sciences. A helpful glossary of medical terms is included, and perhaps a similar glossary for philosophical terms would have been equally useful for the non-philosophers.

Now, the structure: The book has three parts. In Part 1, Gillies develops his own version of an AIM (action, intervention, manipulation) theory of causality for deterministic causality, which he calls an ‘action-related theory of causality’. Part 2 discusses the relationship between causality and mechanisms. This part also includes an in-depth discussion of the Russo–Williamson thesis ([2007]) (namely, both statistical and mechanistic evidence is required to establish a causal hypothesis), which is then modified by Gillies to overcome some empirical counterexamples. Part 3 concerns the connection between probability and causality, primarily for indeterministic causality. Here, Gillies presents propensity theory as his preferred theory of probability. Propensity theory treats probability as objective, and Gillies opts for a long-run propensity theory where the magnitude of a propensity can be measured using statistical frequencies. This part includes a discussion of indeterministic causality, causal networks, and a comparison between Gillies’s theory of probability and Pearl’s position (that probabilities should be interpreted subjectively as degrees of belief).

I won’t be saying much about the discussion on probability here, except for this: I found the discussion on whether a cause raises the probability of its effect interesting but also technically difficult. In contrast to the more conceptual discussion of propensities in (Popper [1959], [1990]) and (Bohm [1957]), Gillies engages less with the conceptual and more with mathematical and statistical discussions of probability theory.

I will now share some of what I consider highlights of the book. Although I read the medical examples with great interest, there were some parts of Gillies’s philosophical theory that I particularly enjoyed. The first highlight is from Chapter 4 in Part 2 on causality and mechanisms. The standard mechanistic theory of causality says that A causes B if there is a mechanism, M, linking A to B (p. 71). In what Gillies refers to as the Dowe–Salmon theory of causality, however, he notes that such mechanisms are primarily physical, involving conserved quantities such as energy, linear momentum, and electric charge. This is not particularly useful, he thinks, for the purpose of developing a theory of causality that is suitable for medicine—unless, of course, one assumes that medicine can ultimately be reduced to physics. Gillies does not accept any such reductionist premise, of course, since he already accepted Russell’s thesis that theoretical physics is non-causal. Still, reductionism seems to be the default assumption in philosophy of science, and I was happy to see Gillies dismissing it so explicitly and categorically:

The problem with the theory that everything can be reduced to physics is that it remains a philosophical speculation, which has not been established in any detail. A few simple results of chemistry have indeed been reduced to physics, but that really is as far as it goes at present. This is a far cry from reducing the whole of medicine to physics. I would argue that it is wrong to make an analysis of the notion of causality depend on such a highly speculative philosophical claim as reductionism. (p. 73)

Gillies then goes through the mechanistic theories of causality proposed by Glennan ([1996], [2002], [2017]) and by Machamer et al. ([2000]), but remains unconvinced that they manage to avoid circularity when attempting to define mechanisms without any use of causal terms (p. 77). He agrees that there is a close relationship between causality and mechanism, but not one where mechanism is used to define causality. Here Gillies makes a turn that I thought was quite ingenious. He suggests that the relationship between causality and mechanism is reversed: ‘we should try to define mechanisms in terms of causality. To put it another way, instead of a mechanistic theory of causality, we should try to produce a causal theory of mechanisms’ (p. 78).

Gillies agrees with the general idea of the mechanistic theory of causality, that if A causes, prevents, or cures a disease D (all versions of causality), it does so by a causal mechanism, M. He thus accepts that A → M → D, where ‘→’ means ‘causes’. But then he defines a mechanism as a sequence of causes, C₁ → C₂ → C₃ → … → C_n (p. 79).

My own reason for being excited by this move is that I think of causality as primitive and irreducible. Would Gillies agree? Or does he see actions as primitive and causality as derived? ‘Causality itself is no longer defined or characterized in terms of mechanism, but rather in terms of human action, using the action-related theory of causality developed in Part I’ (p. 78). I guess it comes down to the question of whether action, intervention, and manipulation (AIM) are thought of as definitive of causality or simply linked to causality. If linked, I would think that the most plausible explanation is that to act, intervene, and manipulate are all causal verbs, and that an action is a manifestation of the agent’s own causal powers. Gillies does not mention causal powers or dispositions in this book, as far as I could see, although he does commit explicitly to propensities in Part 3, on indeterministic causality. Still, to engage in this particular philosophical discussion would come at the expense of the main focus of this book, namely, causality and probability within the domain of theoretical medicine and general causal claims.

In terms of medicine at least, he argues that the reason for being interested in causality is that we can use the causal knowledge to bring about, block, or interfere with causal processes (pp. 24–5). We can produce A, for instance, which then produces B. Gillies calls this ‘productive action’. Discoveries are then used either for the prevention of disease (for example, quit smoking to avoid lung cancer) or for developing a cure for disease (medical intervention). He calls this ‘avoidance action’. The latter is crucial in medicine, since avoidance or prevention of disease is the reason we are interested in causal knowledge in the first place. One example of avoidance action that Gillies uses involves the prevention of cervical cancer through vaccination against the human papilloma virus (HPV). HPV is a necessary but not sufficient condition for cervical cancer, which is why we can avoid the cancer by avoiding HPV infection (p. 199).

This points to the practical importance of understanding causal mechanisms in medicine, and Gillies spends Chapters 8 to 10 discussing the Russo–Williamson thesis. Since this thesis has been subject to extensive debate already, I did not expect this part to end up as one of my highlights. The discussion concerns two types of evidence in medicine: evidence of mechanism (Chapter 5) and statistical evidence (Chapter 6). Gillies has stakes in the mechanism side of the debate because of his AIM theory of causality and the related principle of interventional evidence introduced in Part 1 (p. 24): ‘It does indeed follow from the action-related theory of causality (and other AIM theories of causality) that causality in medicine cannot be established without some interventional evidence’ (p. 67). In Chapter 7, he discusses the evidential principle of ‘strength through combining’ from (Illari [2011]). This principle says that when trying to establish causal hypotheses, it is better to look at different types of sources of evidence than to focus only on one type (p. 130). Unlike the Russo–Williamson thesis, however, the strength through combining principle does not specify which types of evidence are required to establish causality. This is why Gillies opts for the stronger claim, ending up with a modified version of the Russo–Williamson thesis.

I will not go through the examples and arguments in this discussion, but instead give a brief overview of the conclusions. The Russo–Williamson thesis stipulates that to establish a causal hypothesis, one needs both statistical and mechanistic evidence. Gillies modifies this to say that one needs only a plausible mechanism, and that it might not yet be possible to establish the mechanism firmly (p. 141). Besides, causal mechanisms are complex and it is always possible to add more detail to them. This means that mechanistic knowledge would never be complete (p. 181). Second, when it comes to establishing causality for medical interventions to prevent or cure a disease, one should not have to wait for the mechanism. Instead, one should accept low-risk intervention if one has good statistical evidence that it works, while continuing to search for a causal mechanism (p. 176).

To sum up, Gillies sees mechanistic knowledge as crucial in medicine—for establishing causal hypotheses, explaining disease, developing cures, and evaluating the safety of treatment. For all these points, he presents convincing medical examples in rich detail and gives the reader an insight into the history and development of medicine along the way. I was convinced; but regardless of whether one agrees with Gillies’s take, this book offers valuable and original contributions to the philosophy of causality and causal evidence, the philosophy of medicine, and the philosophy of probability. From where I stand, that is pretty cool.