Expiring while the Doctors are Disputing. Principled ...

Expiring while the Doctors are Disputing. Principled Limits of Medical Knowledge and the Ontological Square Ludger Jansen

1. Introduction “Expires while the Doctors are Disputing” – thus is the title of a trenchant etching by William Hogart (1697–1764; cf. Figure 1). The patient, painfully sitting in her chair, is facing her near death, and another woman is already searching the possessions of the still living for valuables while two well-nourished doctors are merrily discussing which of the drugs of their own making would help the patient best. The message is clear: In the face of death there is no time for discussion about the best medication. It is high time for action. In fact, the pressure to act decreases only slightly if we add more time for clinical reasoning or medical research. There will always be the need to help the patient before one has finished disputation, often even before a clear and precise diagnostic result. For, given the complexity of human physiology and pathology, the time required for a thorough study of the subject matter always outruns the deliberation time provided. Human life is finite, and shortish when compared to the potentially infinite history of the medical discipline. This dilemma has been known ever since Hippocrates; it was briefly expressed by the dictum “Ars longa, vita brevis”. True, both diagnostic and therapeutic processes have to come to an end for any individual patient. One might, however, object that medical research is an intergenerational collective endeavour that can be continued beyond the death of individual researchers. But not only individual lives are finite; any finite concatenation of lives is finite as well. Therefore it is likely that medical knowledge, too, will always be finite and restricted. Fragmentarity is, of course, not limited to medicine; similar arguments apply to virtually all scientific disciplines. Not all problems of medical knowledge, however, are due to the incomplete character of present-day medicine. Some problems persist even under the assumption that, one day, ultimate medicine will have discovered everything there is to know about human health and diseases. These problems are not due to any contingently restricted state of the medical profession, but are caused by the structure of medical reasoning. This paper will discuss several such problems that are all rooted, or so I will argue, in the distinct ontological categories of the entities referred to in medical reasoning. I will bring to bear these categorial distinctions on the question of what medicine as an art or a science (or, as some prefer, as an evidence-based practise) can possibly know and how doctors can apply this knowledge. I will argue that there are a number of principled limits which can be

69

Ludger Jansen

Figure 1: Hogarth, The Harlotqs Progress, Plate V: Expires while the Doctors are Disputing (Source: Wikimedia).

traced back to basic ontological distinctions which can be illustrated by means of the so-called ontological square. These epistemological problems are not new, but to my knowledge they have never been traced back to the ontological square. Before I can name and discuss the epistemological limits in question, I will first have to introduce these basic ontological categories.

2. Two Ontological Distinctions In this paper, I will make use of two crucial ontological distinctions. First, I will make use of the contrast between the universal and the particular. Sometimes this contrast is also expressed by comparing types and their tokens. This contrast can be nicely drawn by considering the number of letters in the name rAnnaq: this name is composed of four letter tokens that belong to two letter types. In this paper, I will talk about universals and types interchangeably and will follow the convention to italicise terms for universals:1 human is a universal and so are brain, 1 Cf., e. g., Arp/Smith/Spear 2015.

70

Expiring while the Doctors are Disputing

cell and polysaccharide, but also health, rhesus factor positive and mitosis. Universals can have instances, indeed some universals have a vast number of instances. The universal human has presently more than 7,000,000,000 instances, the universal brain many more and the universal polysaccharide is instantiated by even more molecules. Second, I will distinguish, in a certain technical sense, between concrete and abstract entities. The material things which we encounter in everyday life are concrete particulars. Concrete particulars are rthickq entities; a thick entity is “thought of as involving its properties”.2 Thick entities come along with a huge variety of properties and a multitude of relations. Infinitely many monadic and relational predicates can truly be predicated of them. For this reason, they cannot have a correct description that is both finite and complete. In other words, they are not finitely describable; any finite list of predications will fail to exhaust all their properties and the relations they stand in. This is traditionally called the rineffabilityq of concrete particulars. If we now choose among this plethora of properties and disregard the others, we end up with an abstract particular. Abstract particulars, in this sense, are dependent entities: They need a bearer in order to exist; they are, as Aristotle puts it, “in another thing”.3 In the medical domain, patients are concrete things, as are organs, cells, and molecules. Health, on the other hand, is an abstract thing; it cannot exist without some concrete organism as its bearer. In combination, these two distinctions yield four very general ontological categories which can be neatly displayed in the so-called ontological square, depicted in Table 1. This table is, of course, inspired by Aristotleqs four general ontological categories in the second section of his Categories. Similar four-category schemes have been contended by Jonathan Lowe and Barry Smith.4 In such four-category schemes, the universal/particular dichotomy is often combined with the essential/accidental dichotomy, i. e. with the contrast between accidental properties on the one hand and essential natures on the other.5 Here, however, I stress the abstract character of properties rather than their contingency. Properties, then, can be considered abstractly (whiteness, dryness etc.) as well as be used to pick out concrete things (white things, dry things etc.). Similarly, natures or essences can be considered both in an abstract way (humanity, doghood etc.) and insofar as they inhere in a concrete particular (human, dog etc.).

2 Armstrong 1989, 95. 3 Aristotle, Categories 2 4 Cf. Lowe 2006, Smith 2005. Ultimately, Smith wants to extend this schema to a sixcategory ontology. Cf. Jansen 2008. For the history of such tables cf. Angelelli 1985. 5 I have used such schemes in Jansen 2007a and 2008.

71

Ludger Jansen Table 1: The So-Called Ontological Square Abstract

Concrete

Universal Universals of abstract things; e. g., whiteness, humanity

Universals of concrete things; e. g., white thing, human

Particular Abstract particulars; e. g., this particular whiteness, this particular humanity

Concrete particulars; e. g., this white thing, this human

The ontological square is under attack from two sides. On the one hand, many philosophers would like to do with less than these four fields in the table. Often, for example, the abstract and the universal are identified, as well as the particular and the concrete.6 However, some philosophers have made a successful case for the category of individual properties – or abstract particulars – and some even go so far as to insist that abstract particulars (often called rtropesq or rmodesq in the philosophersq jargon7) are the most basic kind of being.8 Questions of basality are, however, beyond the scope of this paper, as they are not relevant to the questions at stake here. On the other hand, other philosophers argue that more categories are needed, first and foremost particular processes and process universals.9 There are important relations between processes and the four edges of the ontological square: concrete particulars participate in particular processes and during a process a concrete particular will be bearer to a succession of different abstract particulars. Again, I will not worry here whether these entities are somehow basic or not. Having laid out these ontological categories, the question now is which of these categories play which roles in medical knowledge? I will first argue that much of medical knowledge that is taught to future doctors is concerned with the level of universals (section 3). I will then proceed with the discussion of the following four problems: (1) the type-token gap between these universals and the individual patient in the application of medical knowledge, leading to the problem of many models (section 4); (2) the problem of competing types which a concrete patient can instantiate (section 5); (3) the abstraction gap between the concreteness of patients and the abstractness of health which is a challenge in causal reasoning (section 6); and, finally, 6 This is, prominently, the position of David Armstrong. Cf. Armstrong 1978 and 1989. 7 These terms are derived from Greek tropos and Latin modus, respectively, both refering to rthe way in which a thing isq. Other meanings of these terms, for example in the jargon of grammar or rhetorics, are unrelated to their present use. 8 Cf. Williams 1953. 9 Cf., e. g., Smith 2005.

72


(4) the micro-macro problem that is due to abstract entities having as bearers concrete entities at different levels (section 4).

3. Medical Knowledge Is About Universals The first question that I proposed to discuss in this paper was what medicine as an art or a science can possibly know. An obvious way to approach this question is to look at the things taught at medical schools. For heuristic purposes, I will assume here that what is considered to be medical knowledge comprises the totality of what can or could be taught or learned at medical schools. From the point of view of many philosophers, this approach is an aberration from the rphilosophically correctq definition of knowledge as justified true belief. However, due to the insurmountable problems of this definition,10 many philosophers ceased to treat knowledge as a necessarily veridical phenomenon. Under my heuristic assumption, what is considered to be knowledge cannot be veridical either: If anything I learned at school was considered knowledge (at least at the time I learned it), much of this will be false.11 It might be oversimplified for didactic reasons, later falsified by new evidence, or thrown out of the curriculum in the vogue of introducing a new paradigm. Hence, if I speak about knowledge in this paper, what is considered to be known need not be true. It should, however, be justified and backed up by evidence to a sufficient degree. Medical schools today often phrase their aim as conveying certain competences to their students, like diagnostic or methodological competences.12 They teach scientific methods, in particular about statistics, and “how to read a paper”.13 Students acquire implicit norms of the medical community and acquire various forms of personal knowledge: how to recognise typical haptic impressions, typical smells and – using the stethoscope – sounds. They acquire, that is, haptic, optic, acoustic and olfactory “knowledge” to recognise certain phenomena. They “learn to see” – and to feel, hear and smell. Part of this personal knowledge will be the ability of Gestalt perception of important phenomena, be it of the patient appearance typical for a certain disease or typical patterns visible through a microscope.14 This kind of implicit knowledge is acquired by imitation and training. 10 Cf. Gettier 1963 and the avalanche of literature triggered by this short paper (as well as the extensive literature predating Gettier). 11 For this reason, Schulz/Jansen 2013 use the phrase rmedical informationq instead, but this does not work here because it would be a strange term to cover rknow howq as well. 12 In Germany, even a national catalogue of such competences has been comiled; cf. MFT 2015. 13 Greenhalg 2010. 14 Cf. Fleck 1935 und 1947.

73

Ludger Jansen

It is rstoredq in the personal skills of people and it can only be disseminated by direct training.15 There is, however, a large corpus of medical knowledge that can be learned from books and journal articles. This is explicit knowledge, or knowledge-that, which has a propositional content and can be expressed in declarative sentences. It consists of loads of tiny little facts which will sometimes be woven into a coherent net by means of medical theories. This kind of knowledge is acquired through research and stored and distributed in publications or databases. Increasing this kind of knowledge is an integral part of scientific progress. Acquisition and dissemination of explicit knowledge is much easier than acquisition of implicit knowledge, which takes a long time to train. In fact, there is something like an explosion of this kind of knowledge: every year more than 10 million new scientific and technical publications are published. It would take several weeks to read a dayqs output of new medical knowledge. Even in a single field like toxicology approximately 120.000 new papers are published per year. To keep up with this field, a doctor would need to read more than 300 papers every single day of the year, which is far from possible.16 Now, what is the content of medical books and journal articles? What do doctors-to-be learn there? Medical students learn, first and foremost, about the existence of certain types of entities: they learn about anatomical parts of the human body, types of diseases, biochemical molecules and pharmaceutical ingredients. They also learn about certain relations between these types: a human has a head and normally two kidneys, a fetus grows in the womb etc. These relations can be of different kinds. Again, I follow convention and use the underscore to connect lexical parts of a relation term:17 – Taxonomic relations (subsumption): human is_a mammal. – Mereological relations: human has_part some head; womb part_of only mammal. – Topological relations: fetus located_in womb; cell lumen location_of some cell nucleus. – Properties: stem cell bearer_of multipotency; lung bearer_of some respiratory function. – Participation relation: mitosis has_participant some cell; fetus participant_of some ontogenesis. In addition, medical schools teach about the existence of causal relations (aetiology): smoking can cause cancer; AIDS is caused by an infection with the HI virus etc. Teaching causal structures involves, on the one hand, imparting knowledge about causal properties (dispositions, tendencies) of the anatomical, 15 Polany 1958; cf. Jung (ed.) 2014 and Johansson 2015. 16 Cf. Gaus 2005, 22. 17 Cf., again, Arp/Smith/Spear 2015.

74


biochemical or pharmaceutical entities involved, and, on the other hand, imparting knowledge about causal mechanisms typically at work in a human body. Of course medical schools also have to teach their students signs and cures of diseases (diagnostics and therapy). Note that such causal, diagnostic and therapeutic knowledge involves entities at different levels, from biomolecules via cells, tissues and organs up to whole organisms and populations. In the medical profession, knowledge about good practices and other procedural knowledge are often made explicit and codified in clinical guidelines and recommendations. Necessarily, however, the content of formerly implicit knowledge changes when being codified, i. e. when being put into a finite number of fixed rules. Implicit knowledge can be much more reactive to context variables while fixed rules are bound to deal with a rather small and fixed number of variables only. As these guidelines are appropriated by various institutions of the medical profession (or even by legal acts), one could presume that medical knowledge is basically social and prescriptive, as some philosophers of medicine claim.18 However, the state of the medical discipline would be very poor if medical knowledge consisted only of such institutionally enacted rules. An integral part of medical knowledge is the rationale for such rules that is not to be found in other rules, but in empirical knowledge about human biology, about chemistry and in the results of experimental studies. Also, conflicts between rules cannot always be solved by reference to other rules, for these could themselves be in conflict, and so on in infinity. Moreover, sometimes it will be the case that the patient can only be healed by breaking the fixed rules in this particular case.19 In the end, that is, the justification of medical practise has to bottom out at the level of explicit propositional knowledge.

4. Bridging the Type-Token Gap The upshot of the last section is that medical schools typically teach knowledge about types of medical entities. The propositions taught typically express relations between such types, but only few of these relations will translate into universal statements about their instances. The statement “organism has_part some cell” can indeed be paraphrased correctly as “Every organism has among his parts some cells”. But “cell has_part some nucleus” cannot be so paraphrased, as not all cells have a nucleus. While it is indeed typical that cells have a nucleus, there are exceptions like, for example, mature red blood cells: for red blood cells it is typical not to have a nucelus. Moreover, many medical statements ascribe only statistical correlations between types. This is due to two related aspects of the complex 18 Cf. Sadegh-Zadeh 2012, as well as his contribution in this issue, Sadegh-Zadeh 2015. Cf. also Jansen 2014. 19 Possible reasons for this are discussed in the next sections.

75

Ludger Jansen

human body system: First, causal mechanisms may be complex and hidden, which makes it difficult to infer the relevant factors. Second, tiny differences between subjects, be they on the level of genetic factors or bodily constitution, can lead to quite different reactions in a medical study. Hence, a uniform behaviour of whole study groups is rarely observed. In combination, these two phenomena explain the prevalence of statistical methods in medical research. Hence, typical medical research findings will be statistical statements like the following: – After 8 weeks of treatment with Lu AA21004 10 mg, there was a significant reduction in HDRS-24 total score compared with placebo in adults with MDD. – Urinary TGF-beta1 increased significantly in the placebo group but did not change significantly in the sulodexide group. – The overall response rate was 63.6 % (95 % confidence interval [CI] 45.1– 79.6 %).20 In order to be useful for medical practise, such a statistical result must fulfil at least two conditions: First, the observed effect must be relevant, i. e. it must be large enough to be of any help. Second, the result must be statistically significant, i. e. the hypothesised result has been observed in a majority of patients. All the twists and turns of statistical reasoning apply here. The important point, however, is that any knowledge gained by such statistical methods is knowledge about groups sharing certain universal features. The goal of the application of medical knowledge, however, is the health of individuals: it is an individual person who is to be cured.21 Hence, a bridge between type-level statements and individuals is needed. This is still the case in the so-called rpersonalised medicineq. In this approach, medication is administered according to the genetic profile of a particular patient. In the background, however, are still typelevel statements of the form “Patients with genetic profile X are best treated with medication Y”. There is still the need to assign the particular patient to the type Bearer of Genetic Profile X. Standard ontological relations that bridge the gap between these levels are instantiation and exemplification.22 Instantiation is a relation between a concrete or abstract particular and a type of concrete or abstract particulars to which the former belongs. For example, Peter instantiates Homo sapiens, and the colour of Peterqs hair may instantiate the colour universal Black. Exemplification, by contrast, is a relation between a concrete particular and a property type such that the concrete particular is the bearer of a property that belongs to that type. For example, Peter may exemplify the property type rhesus factor positive. 20 These examples have been adapted from PubMed abstracts no. 22901346, 25918727 and 26334293, available at http://www.ncbi.nlm.nih.gov/pubmed; all italics are mine. 21 This has already been noted by Aristotle in Metaphysics I 1, 981a 18–20. 22 I adopt the terminology from Lowe 2006, but restrict exemplification to what Lowe would call occurrent exemplification; cf. Lowe 2006, 207 and Jansen 2007c.

76


Universals of concrete particulars

Universals of abstract particulars exemplify

instantiate

Abstract particulars instantiate bearer_of Concrete particulars Figure 2: Important dependence relations between the four fields of the ontological square. Note the central role of concrete particulars.

These bridges are sufficient in many cases. If, for example, Peter exemplifies the property type rhesus factor positive, his blood should only be used for transfusions to patients who also exemplify the rhesus factor positive type. More difficult cases arise from the fact that individual patients will always instantiate a lot of different types of concrete entities, and they will exemplify various types of abstract entities. Depending on which of these types one refers to, one may arrive at contradictory conclusions. Such a situation I call a rtype-type conflictq. I will now elaborate in more detail the challenges imposed on medical deliberations by type-type conflicts.

5. Type-type Conflicts and the Multi-Model Problem As we have seen, medical knowledge is in general knowledge about universals or types. As patients are particulars, the problem is, of course, how to apply the knowledge about universals to particular cases. So far I have discussed the relation of single universals to single particulars. However, a single concrete particular will always instantiate or exemplify a huge range of universals. Once we take this into account, the picture becomes more complicated. A variety of tricky cases has been discussed in the literature on causation and explanation. A first group of problems is provided by overriding causes. A pet example in the philosophy of causation is the following story: Miller gets a lethal dose of poison. However, before he dies from this cause, he gets shot and he dies because of the bullet wound. Although a lethal dose of poison is a cause of death, and although Miller gets a lethal dose and dies indeed, he does not die from poisoning. The shot and the bullet wound caused by it override the causal chain

77

Ludger Jansen

that starts with the poisoning. It is not an essential bit of this story that the overriding cause (i. e. the shot) is temporally posterior to the cause overridden. This is shown by another popular example: A regular intake of contraceptive pills causes pregnancy prevention, but if Peter takes the contraceptive pill, his intake of the pills is not the cause of his not becoming pregnant. Intake of contraceptive pills only makes a difference for fertile female persons, not for male Peter. Peter, that is, belongs to a type whose instances never become pregnant for anatomical reasons. Hence, intake of contraceptive pills will not change anything here. Or, in medical jargon, contraceptive pills (at least in their traditional varieties) are simply not indicated for male patients. This is because being male overrides the intake of oral contraceptives as the cause of not becoming pregnant. Or, in other words, being an instance of the type active participant of some regular intake of oral contraceptives is overridden by being an instance of the type Male. We have, thus, a conflict between two types which are both instantiated by Peter. While being an instance of the type active participant of some regular intake of oral contraceptives is normally a causal explanation of not becoming pregnant, it is not a good explanation when the type male is instantiated by the same individual. In a similar vein, taking malaria prophylaxis can be a cause for not developing malaria symptoms – which can, however, be overridden by living in, say, Scandinavia or by being immune to malaria because of sickle cell anaemia. Another group of problem cases is the co-occurrence of alternative contraacting causal effects. In medicine, this is infamously illustrated by the side-effects of drugs. For example, pregnancy often causes thrombosis, and as contraceptive pills prevent pregnancy, they also prevent a cause of thrombosis and thus, in turn, are likely to prevent thrombosis. At the same time, the hormones in contraceptive pills increase blood coagulation which, in turn, increases the risk of thrombosis. Now, do contraceptive pills cause or prevent thrombosis?23 While this question on the type-level might be considered purely academic and idle, it is clearly of medical significance on the token-level: if Lucy is going to take oral contraceptives on a regular basis, will this cause or prevent thrombosis? On the level of particular patients, this is clearly a relevant question, though it is hard to answer prognostically. Nevertheless, this is part of everyday medical business. For example, a sanguine complexion may indicate a state of good health, while high blood pressure indicates disease. Now Peter may have a sanguine complexion and high blood pressure at the same time. Should we infer that he is both healthy and diseased? Of course not, as high blood pressure can be an alternative cause for a sanguine complexion. Here high blood pressure overrides the sanguine complexion. We could express this by saying that a sanguine complexion indicates health only if the subject does not have high blood pressure. This way the standard bridges between the type-level and the token-level can be put to work here: if Peter instantiates the type humans with a sanguine complexion, this indicates that 23 The example is taken from Cartwright 1989.

78


Peter exemplifies good health only if Peter does not instantiate the type human with high blood pressure. Other examples, however, are not resolved so neatly. Assume that Peterqs father died from gastric cancer. Assume further that Peter lacks the risk factors for the gastric cancer his father displayed. Has Peter a cancer risk? Persons with a family history of gastric cancer are normally ascribed such a risk. Now a possible explanation for the gastric cancer of Peterqs father is that the father displayed a certain risk factor. If Peter did not inherit this risk factor, he may not have inherited the cancer risk. Even taken together, however, this information is inconclusive with regard to the question of whether Peter has, in fact, a cancer risk: On the one hand, Peter could have inherited other risk factors from either of his parents. On the other hand, he could have non-genetic risk-factors due to, say, his nutrition or smoking habits. The only thing we might reasonably say is that he does not possess one genetic factor that is related to increased cancer risk. Another group of examples involves statistical statements. Assume that 90 out of 100 patients were cured by a certain medication and that 1 out of 100 patients died because of adverse reactions. Assume that Peter is treated with this medication. Will Peter be cured? Will he die? If we do not know more about Peterqs case, all we can say is that there is a high probability for Peter to be cured. To be sure, this would be a mere statistical probability that carries a group index: in different groups, different proportions of patients may be cured. However, as soon as we learn more about the causal mechanisms at work, it becomes possible to ascribe to patients a certain tendency to be cured. If there is no indication of any other relevant factor (i. e. if the proportion of cured patients stays the same in different patient groups), then it is reasonable to ascribe to all patients a tendency to be healed with a probability of 90 % (Model A in Table 2). Table 2: Two models explaining the same statistical data Model A

Share of the population

Model B

Share of the population

90 %-tendency to be cured

100 %

X-pos. = ^ surefire disposition to be cured X-neg. = ^ surefire disposition not to be cured

90 % 10 %

However, the probability distribution may have an alternative causal background, which would justify different property ascriptions. If, for example, we learn that exactly those patients are cured who have genetic factor X, then it is reasonable to ascribe those patients a surefire disposition to be cured – i. e. a disposition that will invariably realise itself if the necessary trigger conditions are given. If we learn

79

Ludger Jansen

that exactly those patients lacking the factor X are not cured, this is a good reason to ascribe the disposition not to be cured to these patients. On these assumptions, the probability of 90 % can also be explained if the study group consists of 90 % patients with factor X and 10 % of patients lacking factor X (model B in Table 2). This is, of course, only one out of infinitely many possible alternative models. That means that there is always a multitude of possible alternative causal stories that could explain a given statistical distribution, and we need a lot of causal assumptions to decide between these alternative options. This would allow us to be more informative when answering the question of whether Peter will be cured: if Peter possesses the factor X, then he will be cured; not so if he lacks factor X. Would we have ascribed tendencies instead of sure-fire dispositions to the various subgroups, the result would have been less determined, but still more reliable. This way we can combine knowledge about the dispositions and tendencies of an individual with knowledge about the properties it exemplifies to infer the causal mechanisms that are relevant for this particular case. It is an open question whether all statistical correlations can be reduced to model-B-like stories.24 There is a strong case that this is not possible in quantum mechanics where there is strong evidence for irreducibly probabilistic dispositions. This effect may or may not carry over to medicine. In any case, there are often plenty of models ready to lend a hand to explain certain phenomena.

6. Causal Inference and the Concrete/Abstract Distinction A crucial problem both in the generation and application of medical knowledge is the contrast between concrete and abstract entities: patients are concrete things; their health, however, is one of their properties. That means that, according to the ontological square introduced in § 2, health is an abstract thing. This gap between abstract and concrete things is relevant in many cases of causal reasoning. This will become clear when we look at Kochqs postulates which are a typical token of causal reasoning in medicine. Koch postulates that a certain infectious disease is caused by instances of a certain kind of microorganisms if instances of this kind be found in all diseased people, but not in healthy ones, and if the disease can be artificially induced if such microorganisms are introduced into previously healthy organisms. On closer inspection, Kochqs postulates turn out to be a combination of two of John Stuart Millqs “experimental methods”25 – which are, in fact, general strategies for causal inferences which are regularly used in both scientific and everyday reasoning. Let us start with a close look at Millqs first method, the method of agreement. “If two or more instances of the phenomenon under investigation have only one circumstance in common,” Mill writes, “the 24 Cf. Jansen 2007b. 25 Cf. Mill, System of Logic, Book III, ch. 8.

80


circumstance in which alone all the instances agree, is the cause (or effect) of the given phenomenon.”26 That is, if we observe that a combination of circumstances A, B and X is followed by an effect E, and also that a combination of circumstances C, D and X is followed by E, we conclude that X is the cause of E and not A, B, C or D. For example, in all of Kochqs patients the microorganisms in question were found in abundance while they were normally not found in healthy persons. Mind that in light of what has been said in § 2, Millqs circumstances are universals and that ragreeing in circumstancesq is a short way to express that in some way or other instances of the same universal are involved: either relevant concrete particulars are bearers of abstract particulars that instantiate the same universal or relevant concrete particulars participate in events of the same type etc. The first problem is that Mill postulates that X is to be the “one circumstance in common” (my italics). Kochqs patients had, of course, a lot of features in common. To start with, they were all humans. Moreover, Millqs description suggests that A, B, C, D and X are all the circumstances present. This is, of course, also false. Situations are concrete, i. e. they come along with infinitely many abstract circumstances. Not all of these are causally relevant for X. Normally we invest previously existing causal knowledge in order to preselect potentially relevant circumstances. Nevertheless, we need to acknowledge the infinity of circumstances, for we may be wrong in our selection, missing the right causes for lack of creativity. Also, it is important to describe the circumstances with an appropriate specificity. An additional source for causal knowledge is provided by Millqs second method, the method of difference. Mill describes this inference thus: “If an instance in which the phenomenon under investigation occurs, and an instance in which it does not occur, have every circumstance save one in common, that one occurring only in the former; the circumstance in which alone the two instances differ, is the effect, or cause, or a necessary part of the cause, of the phenomenon.”27 If we observe, thus, that a combination of circumstances A, B, C and X is followed by an effect E, but that a combination of circumstances A, B and C, but without X, is not followed by E, we conclude that X is the cause of E and not A, B and C. Again, Mill postulates that X is to be “the circumstance in which alone the two instances differ” (my italics). As a rule, however, there is always more than one circumstance in which two situations differ. Of course we could try to combine descriptions of circumstances in a Boolean way in order to construct a unified description of all the circumstances in which the two situations differ. In a way, this would be the one (complex) circumstance in which the two differ. It would, however, not be satisfactory to point to this complex circumstance as a causal explanation of the phenomenon, as this complex circumstance contains a lot of irrelevant aspects of the situation and does not yield to generalisation. 26 Mill, System of Logic, Vol. 1. 1843. p. 454. 27 Mill, System of Logic, Vol. 1. 1843. p. 455.

81

Ludger Jansen Table 3: A fictive RCT example

Effect occurs Effect does not occur

Treatment (n = 500)

Control group (n = 500)

391 (78,2 %) 109 (21,8 %)

117 (23,4 %) 383 (76,6 %)

These problems also affect Millqs other methods, the method of residue and the method of concomitant variations. The problems, however, are neither peculiar to Millqs methods nor to the medical domain, but a general feature of all causal reasoning. In particular, they also affect Randomised Controlled Trials (RCTs), which are presently the method of choice for much medical research. An RCT is basically the attempt to distribute all other causally relevant factors by chance equally among the treatment and the control group. According to the law of large numbers, the probability of a successful equal distribution will increase with the size of the groups.28 A fictive example is detailed in Table 3: out of a cohort of a thousand people, 500 have been randomly assigned to the treatment group and the control group, respectively. People in the treatment group receive the treatment to be tested; people in the control group do not receive this treatment. In socalled “double-blind” studies, neither the study subjects nor the participating doctors know whether they deal with the real treatment to be researched or a mere placebo. This is meant to bar confounding causal factors, for example to secure that participants are not biased and that subjects in both study groups receive the same amount of medical attention, comforting words and confidence. In the example, 109 people receive the treatment, but the hypothesised effect does not occur. The vast majority of treatment cases, however, display the expected effect. Very probably, not all of these outcomes can be ascribed to the treatment to be tested because the hypothesised effect does also occur in a considerable amount of control cases that did not receive the treatment. These cases indicate that the treatment is not necessary for the effect. The results in the treatment group, however, show that the effect is very probable given the treatment. For this reason, a study with such an outcome would be good (though defeasible) evidence for the hypothesis that has been tested, namely that the treatment tested causes the hypothesised effect because it is the common element to which all people in the treatment group have been subjected. An RCT is, thus, a statistical variant of the method of agreement and it inherits all the problems connected with this first of Millqs methods.29 The only help is to repeatedly and critically use several of 28 For philosophical evaluations of RCT studies cf. Cartwright 1989 and Johansson/ Lynøe 2008, 191–200. 29 Millqs non-statistical version of the method of agreement does not, of course, allow to draw any conclusion from these data. Millqs method of difference would, in fact, hinder us from identifying the treatment as a relevant cause.

82


these methods in combination, and iteratively, to select the right circumstances as candidate causes.30

7. Bridging the Micro-Macro Gap: The Multi-Level Problem In the previous section, I discussed the gap between the patient as a concrete particular organism on the one hand and health as an abstract entity on the other. I will now introduce an additional twist to the problem: a lot of the type-related facts taught in medical schools are not at all directly about organims, but belong to the molecular level. Medical students learn about genes and molecules, enzymes and inhibitors. In contrast, the goal of medicine is the health of patients and this is situated on the level of organisms. What these facts do is to relate health to a wide range of properties – and only some of these inhere directly in organisms while others are properties of organs, tissues, cells or molecules. Health is, of course, a complex and difficult phenomenon and there has been a long debate in the philosophy of medicine about what precisely health is. I will not recapitulate this debate in this paper, nor will I be able to do justice to all problems connected with health. For sure, health combines objective and subjective elements: while a person can be diseased without knowing it, she cannot be perfectly healthy when she permanently feels ill. (One could put it thus: the hypochondriac person is not falsely imagining that she is diseased, rather she is merely in error about the nature of her disease, for hypochondria itself is a disease – and probably one of few diseases that is never object of a hypochondriac illusion.) The problem at stake here is not, however, this subjective component of health as such, but rather the fact that the goal of medicine is to restore or preserve health at the organismic level. This is not to say that the predicate “… is healthy” cannot be applied to things at sub-organismic level. To the contrary, a doctor may well say such things like “Your liver is perfectly healthy” or “Your muscles are in a good state of health”. This does not, however, imply that organs and bodily tissues are healthy in the same way as whole organisms are. As was notably observed by Aristotle, doctors may also say things like “Low-fat food is healthy” or “Your urine is healthy”. The predicate “… is healthy” can thus also be applied to things contributing to the health of organisms (like food or drugs) or indicating it (like body temperature or chemical profiles of bodily fluids).31 Similar things can, obviously, be said about the health of entities at the sub-organismic level: certain states of organs or tissues indicate the health of the organism, and they certainly 30 In the end, RCTs can lead to the ascriptions of dispositions or tendencies. That there is no direct route from RCTs to tendency ascriptions should be clear for two reasons: First, the numerical results of a particular study hinges on many contingent factors; other studies will show different numbers. Second, the multi-model problem applies (§ 5). 31 Cf. Aristotle, Metaphysics IV 2 (to hygieinon).

83

Ludger Jansen

contribute to it. In both cases, a statement of the form “The sub-organismic entity X is healthy” can mean, roughly, “The sub-organismic entity X is in such a state as we would expect if X is (or was) part of a healthy organism”. Thus, health attributions at the sub-organismic level have to be explained with reference to health at the organismic level. Health is, thus, primarily situated at the organismic level. For we cannot explain health by simply enumerating necessary requirements situated at the sub-organismic level. Healthy persons need not share all their sub-organismic properties. There is no single combination of sub-organismic properties that is necessary for a healthy organism.32 Health, that is, can be rmultiply realisedq in terms of suborganismic states. But how, then, does knowledge about micro-level entities contribute to knowledge about health? In order to account for this, we are in need of bridges between the micro-level and the macro-level. Faced with this situation, it would not be a good idea simply to dispense with the more complex higher-up levels. For if we give up talk about health of organisms, we would deprive ourselves of the one unifying goal guiding medical research and practise. Nor can we identify health with a specific pattern of molecules or molecular processes in our bodies. One reason for this is that healthy patients may display a wide variety of such patterns, depending, for example, on genetic factors, nutritional input and age. Again, health is multiply realisable, and that forbids reducing health to a certain type of molecular pattern. It is, however, plausible that any change in health status implies some change in the molecular pattern of the body. To adopt David Lewisq phrase,33 there is no difference in health without a difference in the molecular pattern of the body. Health, that is, locally supervenes on molecular patterns. It could be asked whether extrinsic factors like the social and biological environment might be constitutive for health. Whoever wants to include extrinsic factors for health should rephrase the last sentence in terms of global supervenience. Supervenience claims of any variety are, however, not very informative. Remember that the supervenience of mental states on brain states would be both compatible with a variant of mind-brain identity as well as with the prestabilised harmony of Leibnizian dualism.34 If we want to give a more informative account of the relations between different levels of granularity, we need to describe causal mechanisms on the one 32 Some even argue that no combination of sub-organismic properties is sufficient for being healthy because judgements about health often involve comparisons with the statistical average within a certain reference group. Health, then, would not be an intrinsic, but an extrinsic property, at least in part. The downside of this approach is, of course, that someone can become healthy or diseased without any intrinsic change through a fluctuation of the average in the reference group. 33 Cf. Lewis 1986, 14: “no difference of one sort without differences of another sort”. 34 Cf. Weber 2005, 80–81.

84


hand and constitution relations on the other. Causal mechanisms are networks of possible interactions between, e. g., molecules in the cells of the body which can be described as typical patterns of bearers of relevant dispositions or functions.35 As long as cause and effect remain at the micro-level, however, the description is not complete. In addition, we need a bridge between the micro-level and the macrolevel. Such a bridge between two states of different levels is provided by the socalled constitution relation: roughly, an entity of type G is constituted by an entity of type F if any entity of type F in G-favourable circumstances spatially coincides with an entity of type G, while it is possible that an entity of type F does not coincide with an entity of type G.36 (This is compatible with the possibility that other entities of type G are constituted by entities that are not of type F.) For example, the ancient characterisation of health as a harmony or natural equilibrium of elements in an organism would follow the pattern that a certain mixture of the four elements constitutes health if situated in a living body. Today, of course, no account as simple as that can be given of the material constituents of health. In a modern account, the presence of certain molecules in the body may rather constitute a risk factor for the organism. Theoretically challenging is the fact that even non-existing entities may be important in constitution relations. For example, it is not the presence, but rather the lack of certain enzymes on the molecular level that constitutes lactose intolerance on the organismic level. In sum, we need to combine the description of causal mechanisms with the relevant information about which micro-states constitute which macro-states. Only together will they allow bridging the micro-macro gap in medicine.

8. The Limits of Medical Knowledge The aim of this paper was to point out principled problems of medicine based on ontological analysis. I started from two basic ontological dichotomies, i. e. the distinction between the universal and the particular on the one hand and the distinction between the concrete and abstract on the other hand. The first bundle of problems arose from the contrast between the universal and the particular: the type-character of medical knowledge brings with it the challenge to apply it to particular cases. But patients are particulars instantiating a wide range of types and a certain patient may, indeed, instantiate types that lead to contradictory claims about his health status. Of course it will never be possible to consider all types under which a phenomenon can be subsumed, which is one of the causes of the fragmentation of our medical knowledge. A second bundle of problems arose from the contrast between the concrete and the abstract. Concrete patients are bearers of infinitely many abstract properties. 35 Cf. Röhl 2012 and the references given there. 36 Cf. Baker 2000, 95 and Jansen 2009.

85

Ludger Jansen

Again, medical practitioners have to choose a small set of types that they consider being relevant – and they may be in error both about the causal or diagnostic relevance of the chosen property types as well as the irrelevance of the types not chosen. A further challenge is that medical knowledge covers various levels of granularity, from molecules and cells via tissues and organs to organisms and populations. Health, in particular, is an abstract particular, a property primarily of organisms. Much medical knowledge, however, is knowledge about types of suborganismic level. Researchers as well as practitioners have to relate these distinct levels of granularity in order to apply this knowledge for the benefit of the patient. Again, not all levels of granularity can be used in a decision making process; practitioners do not only need to decide on their choice of granularity, but also about the precision necessary in the case in question. Based on an ontological analysis, I have pointed to some reasons why medical science and medical practise is so difficult. These problems affect the generation of knowledge in medical research as well as the application of medical knowledge in decision making processes by practitioners. Much of what has been said in this paper is, of course, not only relevant for medicine, but of general relevance. The problems of causal reasoning, for example, affect all areas of (non-formal) knowledge and practise. The problems are, however, of particular relevance for medicine because there it is a matter of life and death for the patient, to not expire while the doctors are disputing.37

References Angelelli, I. 1985. En torno al cuadrado ontol|gico, in: Annuario Filosvfico 18, 23–32. Aristotle, Categories, in: The Complete Works of Aristotle. The Revised Oxford Translation, ed. by J. Barnes, Princeton 1984, vol. 1, pp. 3–24. Aristotle, Metaphysics, in: The Complete Works of Aristotle. The Revised Oxford Translation, ed. by J. Barnes, Princeton 1984, vol. 2, pp. 1552–1728. Armstrong, D. M. 1978. Universals and Scientific Realism, 2 vols., Cambridge. Armstrong, D. M. 1989. Universals. An Opinionated Introduction, Boulder/San Francisco/London. Baker, L. R. 2000. Persons and Bodies. A Constitution View, Cambridge. Cartwright, N. 1989. Naturems Capacities and Their Measurement, Oxford. Fleck, L. 1935. Entstehung und Entwicklung einer wissenschaftlichen Tatsache. Einführung in die Lehre vom Denkstil und Denkkollektiv, Basel; repr. Frankfurt am Main 1980. 37 This paper originated in a talk I gave at the workshop “Health, Life, and Experts. Knowledge in the Life-Sciences” in Rostock in 2011. For helpful comments and discussion on versions I would like to thank Martin Boeker, Georg Fuellen, Hilte GeerdesFenge, Niels Grewe, Ingvar Johansson, Jürgen Rötker, and Petter Sandstad.

86

Expiring while the Doctors are Disputing Fleck, L. 1947. Patrzec´, widziec´, wiedziec´, Problemy, no. 2, 74–84; English translation: To Look, To See, To Know in: Cognition and Fact. Materials on Ludwik Fleck, ed. by R. S. Cohen and Th. Schnelle, Dordrecht 1986, 129–152; German translation: Schauen, sehen, wissen, in: L. Fleck, Erfahrung und Tatsache. Gesammelte Aufsätze, ed. by Lothar Schäfer and Thomas Schnelle, Frankfurt M. 1983, 147–174. Gaus, W. 2005. Dokumentations- und Ordnungslehre. Theorie und Praxis des Information Retrieval, 5th revised edition, Berlin et al. Gettier, E. 1963. Is Justified True Belief Knowledge? in: Analysis, 23, 121–123. Greenhalg, T. 2010. How to Read a Paper. Basics in Evidence-Based Medicine, 4th edition, Oxford. Jansen, L. 2007a. Aristotleqs Categories [Untimely Review], in: Topoi 26 (2007) 151–158. Jansen, L. 2007b. Tendencies and other Realizables in Medical Information Sciences, in: The Monist 90/4, 534–555. Jansen, L. 2007c. Dispositions, Laws, and Categories. A Critical Study of E. J. Loweqs The Four-Category Ontology, in: Metaphysica 8, 210–220. Jansen, L. 2008. Kategorien: Die top level Ontologie, in: L. Jansen, B. Smith (eds.), Biomedizinische Ontologie. Wissen strukturieren für den Informatik-Einsatz, Zürich: vdf 2008, 85–112; English translation: Categories. The Top-Level Ontology, in: K. Munn, B. Smith (eds.), Applied Ontology. An Introduction, Frankfurt 2008, 173– 196. Jansen, L. 2009. Unity and Constitution of Social Entities, in: L. Honnefelder, E. Runggaldier (Hgg.), Unity and Time in Metaphysics, Berlin/New York: de Gruyter 2009, 15–45. Jansen L. 2014. Review of Kazem Sadegh-Zadeh, Handbook of Analytic Philosophy of Medicine, in: Medicine, Health Care and Philosophy 17, 161–162. Johansson, I., Lynøe, N. 2008. Medicine and Philosophy. A Twenty-First Century Introduction, Frankfurt: Ontos. Johansson, I. 2015. Tacit knowledge revisited, in: N. Juth, G. Helgesson (eds.), Patients, Values, and Medicine: Hommage r Niels Lynøe, Stockholm, 69–81. Jung, E. (ed.) 2014. Jenseits der Sprache. Interdisziplinäre Beiträge zur Wissenstheorie Michael Polanyis, Münster. Lewis, D. K. 1986. On the Plurality of Worlds, Oxford. Lowe, E. J. 2006. The Four Category Ontology, Oxford. Mill, J. St. 1843. A System of Logic Ratiocinative and Inductive, London; repr. in: The Collected Works of John Stuart Mill, vols. 7–8, Toronto 1973–74. MFT Medizinischer Fakultätentag der Bundesrepublik Deutschland 2015. Nationaler kompetenzbasierter Lernzielkatalog Medizin, Berlin 2015; available at http:// www.nklm.de. Polanyi, M. 1958. Personal Knowledge: Towards a Post-Critical Philosophy, Chicago. Röhl, J. 2012. Mechanisms in Biomedical Ontology, in: Journal of Biomedical Semantics 3 (Suppl 2), S9. Sadegh-Zadeh, K. 2012. Handbook of Analytic Philosophy of Medicine (Philosophy of Medicine series, vol. 113), Dordrecht et al. Schulz S., Jansen L. 2013. Formal ontologies in biomedical knowledge representation, in: Yearbook of Medical Informatics 8, 132–146.

87

Ludger Jansen Smith, B. 2005. Against Fantology, in: J. C. Marek, M. E. Reicher (eds.), Experience and Analysis, Wien, 153–170. Weber, M. 2005. Supervenienz und Physikalismus, in: U. Krohs, G. Toepfer (eds.), Philosophie der Biologie. Eine Einführung, Frankfurt am Main. Williams, D. C. 1953. On the Elements of Being, in: Review of Metaphysics 7, 3–18 and 171–192.

PD Dr. Ludger Jansen, Institute of Philosophy, University of Rostock, D-18051 Rostock.

88

Angewandte Philosophie. Eine internationale Zeitschrift/ Applied Philosophy. An International Journal Heft/Volume 1|2015 herausgegeben von/edited by Ludger Jansen, Jörg Hardy

Medizinische Erkenntnistheorie / Medical Epistemology

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar. ISSN 2198-8404 ISBN 978-3-8471-0505-3 ISBN 978-3-8470-0505-6 (E-Book) ISBN 978-3-7370-0505-0 (V&R eLibrary) Weitere Ausgaben und Online-Angebote sind erhältlich unter: www.v-r.de Gedruckt mit freundlicher Unterstützung der Stiftung „Menschenwürde weltweit“ in der Verwaltung der Deutschen Stiftungsagentur GmbH, Neuss. © 2015, V&R unipress GmbH, Robert-Bosch-Breite 6, 37079 Göttingen / www.v-r.de Alle Rechte vorbehalten. Das Werk und seine Teile sind urheberrechtlich geschützt. Jede Verwertung in anderen als den gesetzlich zugelassenen Fällen bedarf der vorherigen schriftlichen Einwilligung des Verlages. Printed in Germany. Druck und Bindung: CPI buchbuecher.de GmbH, Zum Alten Berg 24, 96158 Birkach Gedruckt auf alterungsbeständigem Papier.

Inhalt Themenschwerpunkt: Medizinische Erkenntnistheorie / Medical Epistemology Ludger Jansen Einleitung: Was weiß die Medizin? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

Kazem Sadegh-Zadeh Die Medizin ist eine deontische Disziplin . . . . . . . . . . . . . . . . . . . . . . . . . .

10

Alex Broadbent Is Stability a Stable Category in Medical Epistemology? . . . . . . . . . . . . .

24

Heiner Fangerau, Michael Martin Medizinische Diagnostik und das Problem der Darstellung: Methoden der Evidenzerzeugung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

Ludger Jansen Expiring while the Doctors are Disputing. Principled Limits of Medical Knowledge and the Ontological Square . . . . . . . . . . . . . . . . . . . . . . . . . . .

69

Marco Stier Im Dickicht der Kategorienfehler, oder: Was weiß die biologische Psychiatrie? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

89

Diskussion Rainer Enskat Gettier und kein Ende? Das doxastische Dogma der modernen Wissenstheorie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115