While in medicine and healthcare the default assumption is that all patients are different, causation itself is sought as something that is robust throughout different contexts. This leaves us, in effect, to search for same cause and same effect:
Same symptoms, same diagnosis (diagnostics)
Same diagnosis, same intervention (standardised treatment)
Same intervention, same effect (tested though RCTs)
Although individual variations are acknowledged, they are nevertheless not the focus when trying to establish causality. Instead, variations can be used to form more fine-grained classifications or sub-groups, where one again looks at what is the same. In other words, uniqueness is considered an obstacle when one tries to establish causality scientifically.
This contrasts with the dispositionalist framework. No two individuals will have exactly the same combination of causal dispositions or propensities. Even if there are some dispositions that we share, such as gender, age or medical condition, so many other dispositions will be different from one individual to another. Grouping patients into more relevant sub-groups will plausibly tend to give a more appropriate average than broader and unspecific sub-groups. We know, however, that not all pregnant women in their thirties or all men over 60 with hypertension are identical in all their dispositions – or even sufficiently identical. Which of these dispositions are taking part in the single causal process that we are investigating? This is a question that cannot plausibly be answered with certainty. We will therefore never be sure of whether or how precisely a sub-population represents the dispositions in place in the individual process. A frequentist approach, we have said, will either have to overlook this knowledge gap or try to further specify the relevant sub-group (see Rocca, Chap. 3, this book). Eventually, however, one might end up with a sub-group with only one member: the N = 1 group consisting of the single patient. Still, the problem remains how to establish, predict and explain causality for a patient for whom no suitable, or suitable enough, sub-population can be found.
This is one reason why MUS represent a methodological challenge for medicine and healthcare. In the current paradigm, the best way to establish causality is by showing that the same cause makes a difference toward the same effect in sufficiently similar contexts. To make this clear, think back to the principle of randomised controlled studies (RCTs), as explained above in Sect. 3.2. We saw here that RCTs are considered to be the best way to establish causality within the current paradigm of evidence based medicine and practice, and they are designed to test for a type of homogeneity: common causes and common effects. This means that even though there is plenty of individual variation within the clinical study, these variations are not what the RCT is designed to study or establish. On the contrary, such individual variations are supposed to be shielded off through randomisation, so that test group and control group are overall very similar. With RCTs, we look for the overall effect of an intervention in the test group, compared with the overall effect in the control group. The intervention is then the same, and the effect tested is the same.
RCTs thus target same cause (intervention), same effect (outcome). This is completely in line with Hume’s regularity theory of causality (Hume 1739), but it doesn’t acknowledge the dispositionalist perspective that causes as dispositions are intrinsic properties: they tend to manifest, but not always. They tend to make a statistical difference, but not always. They tend to produce one effect, but not always the same (Anjum and Mumford 2018, see also Anjum, Chap. 2, this book). RCTs are great tools to detect manifestations that make a difference at population level, but they are not useful for studying dispositions that remain mostly unmanifested, and which tend to manifest themselves in single and causally unique cases. We see, then, that the problem of MUS is not an isolated one, but one that has its roots in the Humean influence on medical thinking about causality that can be summarised in the following three points:
A and B are observed repeatedly (empiricist criterion: causality must be detected empirically)
Whenever A, B, under some normal or ideal conditions (regularity criterion: same cause, same effect)
B happens because of A (monocausality criterion: one cause, one effect)
If not B, then not A (falsification criterion: a difference in effect must mean that there is a difference in the cause).
For A to be the cause of B, these conditions must be met, according to the Humean notion of causality. Medically unexplained symptoms, however, typically fail to meet one or several of these criteria, which is why we cannot say that a medical cause has been found.
Let us show this by considering the case of unspecific lower back pain. Qualitative studies show that in the clinical dialogue, patients usually associate this condition with an episode such as bending or lifting. Patients mention that they felt sudden pain during a certain activity, and that they have been in pain ever since (Jeffrey and Foster 2012). There is, in the clinical encounter, a deep intuition of a causal link between a certain accident, or event, and the condition. However, this cannot be epidemiologically confirmed. There is much literature on unspecific lower back pain, but no systematic association has been shown with mechanical factors (lifting, standing, walking, postures, bending, twisting, carrying, and manual handling) nor with activity levels, obesity, smoking, mood, or genetic factors (see Eriksen et al. 2013b for a review of the epidemiological evidence). None of these causal factors seem to fulfil the Humean criteria of regularity, repeatability and falsification. Epidemiologically, and according to Humean criteria, therefore, there is no clear cause of unspecific lower back pain, despite decades of research. And yet, single patients tend to be able to indicate a cause, at least as they experience it. This and similar cases of medically unexplained conditions represent a challenge for any attempt at standardisation or universal approach to cure and healthcare.
4.2.1 The Patient Context: What Was There Before
From a healthcare perspective, one does not expect that the same cause will give the same effect in different individuals. Individual responses depend on what else was there already, as part of the patient’s own context. A person who is at a vulnerable stage in life might be more disposed to an infection than a person who is at a more robust stage, for instance. This is, one can say, elementary clinical knowledge. Still, this real-life complexity becomes a problem for causal understanding when we try to analyse it using the Humean criteria. Dispositionalism instead acknowledges complexity and context-sensitivity as basic features of causality.
We can represent the different impact of one causal disposition in different patients with the vector model of causality (Mumford and Anjum 2011), where each vector represents one causal disposition in place, and the line T represents the threshold for the manifestation of an effect, as explained in Chap. 2. In the vulnerable patient (Fig. 4.1), the situation is much closer to the threshold of illness than in the robust case (Fig. 4.2). This means that even a minor burden on health can have a major impact, because it pushes the situation over a threshold. This is a well-known phenomenon. We often speak of the straw that broke the camel’s back, which was simply the final straw adding to the already heavy burden. A cause might then be simply what tips the situation over the threshold, which seems far too insignificant if we ignore what was already there before it.
Let us say that the two patients get affected by influenza, and after that only one of them develops a chronic burnout syndrome, while the other recovers normally. From a Humean point of view, this does not tell us much about the causal role of the influenza for the onset of burnout. Instead, we would need to check whether there are other patients as similar as possible to the patient who develops burnout symptoms after getting influenza (same cause – same effect, all else being equal).
From a dispositionalist perspective, however, looking further into cases of individual variation and context-sensitivity represents a chance for understanding something about the underlying causal story. When the same cause gives different effects in two different contexts, we might learn something new. Clearly, to do that, it does not help to focus on the single cause or the single effect. Instead, one should try to understand what was already there in the two different contexts, disposing toward or away from health and illness. This type of reasoning needs to be qualitative and explanatory, in order to be fruitful. By trying to understand all the causal dispositions in place, and the way they interact with each other, we build a causal explanation – a hypothesis – for how and why things went the way they did.
Note that the causal explanation, or causal mechanism, although being based on empirical evidence, is not itself something we can observe directly. This is why Hume’s empiricist approach does not include causal theories or explanations but sticks to what can be observed and counted. On the contrary, the search for evidence of a plausible causal explanation (which dispositions are present, how they interact and manifest) is at the core of the dispositional approach. It is also crucial for any scientific theory, including in medicine and healthcare.
Humean and empiricist influence have been strong, not only in research, but also in the clinic. There tends to be the expectation, at least in the implementation of some health policy and clinical guidelines, that patients with the same diagnosis should respond similarly to the same treatment. Personalised medicine and system medicine have been rising trends and can be seen as attempts toward a more dispositionalist approach: aiming to fit the treatment to the patient’s own dispositions. However, these approaches are mainly focussed on genetic or molecular dispositions and have less focus on psycho-social or ecological complexity (Vogt et al. 2014). This will be discussed in Chap. 5 when we look at the biomedical model of medicine.
Allowing the features of uniqueness and complexity to guide the clinical encounter, we should focus less on what is the same and more on what is different and unique for this particular patient, also for causal matters.
Humeanism refers to David Hume’s regularity theory of causality, which emphasises features such as empiricism, observable features (data), mono-causality, repetition and same cause – same effect. Probabilities are understood as generated statistically (frequentism).
4.2.2 Qualitative and Quantitative Approaches to Causal Inquiry
What exactly do we mean by a qualitative analysis? And how should such analysis help us look for causal explanations? How does this approach contrast with quantitative analysis and the search for same cause – same effect? In our philosophical framework, we have a particular take on what should count as qualitative and quantitative approaches to causal inquiry. Qualitative approaches will be concerned with the investigation of many types of information in few tokens, and with how these relate, and under which conditions. In contrast, quantitative methods will look for few types of information that are in common for many tokens (see also Anjum and Mumford 2018: 106). Notice that from the dispositionalist perspective, qualitative research can advance causal understanding and theory, and is not limited to the purposes of meaning and lived experience (see Sect. 4.4). Qualitative research, in our definition, encompasses scientific enquiry of a phenomenon, as long as such enquiry aims to understand a causal process, while quantitative research aims to identify the numerical relationship between variables. A qualitative approach, in our understanding, might involve numerical values, but is always process-oriented, aims to generate theoretical understanding, it is adapted to the most relevant context of application of the research, and it happens in-situ, often in a participatory way.
An example might help illustrate this distinction.
For example, a recent large study compared the whole-genome sequences of participants with food allergy to peanuts, egg or milk with non-allergic participants (in total almost 3.000 individuals were included)[…] The results showed statistically significant DNA modifications in specific loci of the genome, indicating that these loci are probably part of the genetic component of the food allergy. Other information about the participants were age, sex, ancestry (European or non-European), results of food allergy tests, and presence of other allergy-related disease. While such a horizontal analysis has big statistical power, it relies on the preliminary selection of a limited amount of variables to compare. The selection is informed by existing knowledge and working hypothesis (in the case of this study, that allergies have a genetic component). Additionally, it is dictated by practical considerations since these studies include a large amount of participants. While results are statistically robust, their contribution is limited to a small part of the picture. In fact, genetic predisposition is only one of many actors for the onset of a condition.
Let us imagine using a larger filter to evaluate which contextual variables to include in the analysis. We might then consider a complete range of clinical factors, blood levels, present and former state of health, dietary habits, lifestyle, polypharmacy, psychological health, addictions, traumas, including as much information as possible about the unique context that was exposed to the allergen. This would necessarily restrict the comparison to a limited number n of patients […].
The experiment would then be a qualitative, rather than quantitative, analysis. It would have a different aim: the aim of identifying not a single element that is frequently involved, but enough elements to suggest a pattern, or offer an explanation that is valid in this specific instance. Such explanation might fall outside the boundaries of existing knowledge and suggest an advance in the overall understanding of causal mechanisms. Finding out whether these hypotheses are generalizable and to which extent, belongs to a subsequent stage of research. (Rocca 2017: 117)
In our framework we propose that qualitative approaches to causal inquiry collect many types of information in few tokens and look for a theoretical understanding of how these relate causally in a particular context. Quantitative approaches, instead, look for few types of information in many tokens, and aim to identify numerical correlations among them.
4.2.3 Dispositional Take On Perfect Regularity: Is It Causality or Something Else Entirely?
A dispositionalist denies that causality is something that produces perfect regularity between cause and effect. Instead, causality is understood as tendencies, where the cause A only tends or disposes toward the effect B. So even if A is present, B can still be counteracted by adding an interferer I (Mumford and Anjum 2011). All of medicine is premised on this idea. Even if one has not yet been able to find a treatment, the expectation is still that if one can understand the causal mechanisms of the disease, it should be possible to counteract or interfere with the causal process in one way or another.
This has a surprising consequence. If there were to be a perfect correlation of A and B, where no changes in context could influence the situation in any way, a dispositionalist should become suspicious. Is this a case of causality after all? Or could it be a case of classification or identity? For instance, all humans are mortal. And although scientists are still working on ways to counteract and delay death, one could still argue that any immortal being could not be human. So even if there were a human-like immortal species in the future, we might say it’s not the same as a human.
Now take a medical case. It is said that Down syndrome is a genetic disorder caused by the presence of all or part of a third copy of chromosome 21. But is this the right way to phrase it? As long as a person has the extra chromosome, they will be diagnosed with Down syndrome. If Down syndrome is then defined as the condition of having the third copy of chromosome 21, then, of course, there will be a perfect regularity between A and B. But the reason why we have the situation that whenever A then B, is that A is defined as B. In that case, A = B. This does not mean that there is no causality going on here. The causal relationship would then be between the extra chromosome and the expression of the condition, which vary in degree from individual to individual. The symptoms of Down syndrome are then caused by the extra chromosome, and will be manifested in different ways in different individuals. Whether someone has the syndrome will correlate perfectly with whether they have the extra chromosome, without any individual variations. This suggests that it is an identity relation, not a causal one.
Perfect regularities, on the dispositionalist perspective, could be produced by other types of truths than causal ones, such as classification (all humans are mammals), stipulation (all electrons are negatively charged), identity (bachelors are unmarried men) or essence (humans are mortal). In contrast to these types of claims that have the categorical form ‘All As are Bs’, causal claims are about what happens under certain conditions. A causal claim is therefore a hypothetical or conditional matter: ‘If we do x to y, would z follow?’. To say that ‘All As are Bs’ is to make a statement about how to categorise A and B with respect to one another. In contrast, when we ask whether A is a cause of B, we want to know whether and to what degree A is able to bring about B or at least contribute to the production of B.
Note, however, that if A is indeed a cause of B, a dispositionalist should not expect that all instances of A will actually and successfully produce B. Causes, as dispositions, are irreducibly tendential. There is never more than a tendency of A to produce B. As discussed in Chap. 2, a dispositional tendency can be stronger or weaker. Someone can be more or less vulnerable, more or less violent and more or less allergic to peanuts, for instance. We also saw that dispositional tendencies give rise to individual propensities, rather than statistical frequencies (see Rocca, Chap. 3, this book). The degree of tendency does not determine how often a disposition will manifest, but only how strong the intrinsic disposition is in this individual situation. For instance, if we want to know how fertile someone is, one should do a sperm count rather than counting the number of offspring. The higher the sperm count, the stronger the disposition of fertility. It does not follow from the strong fertility that one will eventually have a lot of children. It also does not mean that other people with the same sperm count will have many children.
We see, then, that a dispositionalist should not expect perfect regularity of cause and effect. Instead, a dispositionalist should be sceptical if there is a perfect correlation that is insensitive to contextual change. Could it be a case of identity, classification or essence instead? Or have we already stipulated some ideal conditions or idealised model under which the cause would always produce the effect? Either way, we cannot expect that causality will manifest itself in perfect correlation in a real-life situation such as what we encounter in the clinic. The only way we can expect that the same cause will always produce the same effect, is by stipulating some average, normal or ideal patient with average, normal or ideal responses. In the clinic, however, such encounters are rare.