Heterogeneity of the Concept of “Cause” in Medicine1
Eterogeneità del concetto di “causa” in medicina
Paolo Vineis
University of Torino and Institute for Scientific Interchange Foundation, Torino
The concept of cause has been the object of a number of analyses, and is one of the
most important subjects for discussion in philosophy. It is possible to identify several
different traditions on causality in the history of philosophy, and each of these
corresponds to a different conception of causes in medicine.
1. A first, deeply-rooted tradition is represented by the aristotelian point of view, based
on the central idea of material and efficient cause. This tradition had its maximum
development in the medicine of the early nineteenth century, when Bichat stated that an
autopsy can shed more light on disease than hundreds of physical examinations of
patients. This tradition is founded (a) on morbid anatomy, that is on the identification in
the dead body of characteristic lesions left by the cause of disease, and (b), especially
after Virchow’s contribution, on the identification of the cell as the main actor in the
pathogenetic process. Another extension of this traditional paradigm is microbiology,
with its search for agents that meet Henle-Koch’s criteria, that is a single, necessary and
sufficient cause. The cause is an agent that is always identifiable in the lesions, can be
observed, it is absent in healthy subjects, and causes the disease by inoculation into
animals. The experimental attitude becomes so extreme that Pasteur makes brave and
sometimes ethically questionable experiments, like the vaccination of Joseph Meister,
that Pasteur’s disciple Roux refused as premature and dangerous.
A typical example of the conception we are considering is the German school of
physiology, that included Carl Ludwig, Helmholtz and Du Bois-Reymond, the latter
being the main representative of medical positivism.
2. To such a tradition, that dominates science in the nineteenth century, a reaction arises
in the first years of this century, aiming to substitute the concept of functional “law” to
that of material cause. The most important expression of this tendency can be found in
Mach’s philosophy – who proposed some typical themes of Hume’s philosophy while
criticizing a naive and material conception of causes -, and subsequently in Bertrand
Russell, who claimed that the concept of cause would have soon become out-of-date
like the monarchy. Also this conception finds a correspondence in medicine, in the
school of Virchow’s disciples. Students such as Martius and Hueppe, in fact, developed
causal models of a multifactorial kind, that we find in the current probabilistic
interpretation of causality. They insisted on the fact that we should get rid of a
unilateral conception of causality, aiming to search for necessary causes. According to
Mach’s functionalist philosophy, Martius interpreted the “disposition” (internal cause)
and the “stimulus” (external cause) as variables subject to a mathematical treatment.
1
Acknowledgment: this research has been made possible by a grant of the Compagnia di S. Paolo
(Torino) to the ISI Foundation.
According to the even more radical school of “conditionalists”, the word “cause” itself
had to be completely abandoned and replaced by “determining condition”: “all the
conditions in a process are equally important”, that is no hierarchy can be established.
Hanemann claimed that the monocausal point of view was promoted in medicine
because of a prevailing interest in cure, but from a strictly scientific perspective
monocausality should be abandoned (1).
3. A third recent stream is represented by the development of modal logics and
particularly of “counterfactual conditionals”. The basic idea is that of the cause as a
“conditio sine qua non” for the occurrence of the effect; thus, the background is still that
of the necessary cause. One the most representative philosophers in this stream of
thought is Mackie, who suggests the concept of cause as INUS (Insufficient Nonreduntant component of an Unnecessary Sufficient Complex) (2). In other words,
Mackie claims that there are not separate causes, but causal complexes; the different
complexes are sufficient to induce the effect, but none of them is in itself necessary.
Within such complexes, however, “non-redundant” (i.e. necessary) components can be
isolated, and these represent the true causes.
This style of reasoning is very common in experimental medicine: a large part of the
knowledge we have in endocrinology, for example, comes from experiments aiming to
show the effects of the elimination of a certain gland. However, this approach is not
very effective to describe probabilistic inferences in epidemiology, when these consider
the individual subject: lung cancer can arise in the single individual even in the absence
of cigarette smoking, which is not a necessary cause. The counterfactual reasoning,
though, is applicable at the population level: without cigarette smoking there would not
have been the long-lasting epidemic of lung cancer that has characterized developed
countries and is now spreading to developing countries.
4. One of the interesting aspects of modal logic is that to reason about counterfactual
conditionals we have to refer not only to the current situation, but also to one or more
alternative, possible situations (or, according to Lewis and Stalnaker, “possible
worlds”). The use of counterfactual conditionals is accompanied by a hypotheticdeductive style of reasoning, and to induction “by elimination”. A typical example is
represented by the famous description, due to Hempel, of Ignatz Semmelweis’ work (3,
4). Hempel describes Semmelweis’ inferential procedures as an example of “trial and
error” approach: after the early epidemiologic observations, that suggested the
transmission of “particles” from dead bodies, he started seeking for evidence that would
falsify his theory, and for alternative explanations. For example, he ruled out that
puerperal fever had been contracted before hospitalization due to the women’s living
conditions. He also excluded, partly by the means of experiments, several other
potential confounders or sources of bias.
5. Still another peculiar and recent concept of cause has been introduced by Wesley
Salmon through the idea of the “propagation – or trasmission – of a mark over time”
(5). In fact Salmon proposes two different models for causality, which are supposed to
integrate each other. The first model had been suggested by Reichenbach many years
before, and it was simply based on probabilistic computation (the “positive relevance”
criterion). The basic idea is that two events (A and B) have a common cause C if we can
show (a) that the joint probability of A and B is greater than the product of their
separate probabilities, i.e. p(AB)>p(A)p(B); (b) that the introduction of C into the
equation entirely or at least partially explains the association between A and B.
Condition (a) simply expresses the concept that the probability that two events occur
jointly is the product of the individual probabilities if the two events are independent,
while it is greater than the product if they are not independent. Condition (b) means that
the joint occurrence of A and B can be explained by a common third event: for example,
subjects A and B both have an angiosarcoma of the liver, and they were both exposed to
vinyl chloride (VC) in the same plant. We compute that the probability of such joint
events by simple chance is very low (condition a). In addition, hypothesizing that the
common exposure is VC, we observe that all the excess risk (observed/expected cases)
is explained by such exposure (condition b).
However, according to Salmon this line of reasoning is not sufficient, because it does
not allow us to distinguish genuine causes from confounders. He suggests that a way to
establish whether an event is a genuine cause, in addition to the statistical considerations
above, is to include it in a process, and establish whether along the process there is what
he calls the propagation of a “mark”. In other words, a genuine causal process is one in
which you can follow a “mark” that propagates in the course of time, precisely because
the causal events are able to induce a structural change that becomes a part of the
effect.
According to a classical example, yellow fingers are a risk indicator for lung cancer:
lung cancer patients, in fact, have yellow fingers more frequently than population
controls, thus fulfilling Reichenbach’s definition of causality (positive relevance).
However, clearly yellow fingers are just an indirect marker of exposure to tobacco
smoke, and not a genuine cause of lung cancer. The reason is because they do not fulfill
Salmon’s second criterion of causality, the propagation of a mark along the process. In
fact, while you can identify p53 mutations in the lung cancer that are characteristic of
tobacco carcinogens, thus following the structural changes left by tobacco into the lung,
the same does not hold true with yellow fingers.
We go back, in this way, to the nineteenth century idea of a “localization” of disease,
that is the cause is an alteration that can be identified in the tissue and is characteristic
of a certain agent (for example the tubercle for tuberculosis). The cause is, therefore, a
kind of signature that persists over time. This is an extremely common concept in the
history of medicine, now widespread thanks to the developments of molecular
medicine: the “cause of cancer”, for example, is identifed with a genetic alteration such
as a mutation in the p53 gene, the so called “guardian of the genome”. According to the
theory of molecular “fingerprints”, a carcinogenic chemical would be able to leave a
characteristic signature in a specific cancer gene, and this fingerprint would be
transmitted through generations of cells until it is observable within the tumour. This is
at the moment still a theoretical hypothesis with some observational basis (6). Salmon’s
theory, incidentally, does not escape the difficulties of other strong, non-probabilistic
views of causality.
6. There is still a problem that deserves consideration, that is the fact that both the idea
of cause and the definition of disease have become “blurred” in the course of time. Let
us consider the following Table:
Table 1: Disease manifestations
Definition of
causal agent
Monothetic
Polythetic
Polythetic
Monothetic
Tuberculosis
(I)
Smallpox
(II)
Bulimia and Anorexia
(III)
Lung cancer
(IV)
According to the table, for groups I and II the causal agent is single and well-known (a
virus for smallpox, a bacterium for tuberculosis). However, even in such cases of
monocausality, disease manifestations can be polythetic, i.e. based on more than one
criterion; in the case of tuberculosis, the disease is not always univocally recognizable,
and complex clinical and immunological criteria have to be used. In other words, while
for smallpox clinical diagnosis is unequivocal, tuberculosis is a complex constellation
from a symptomatologic point of view, and the only way to recognize its different
manifestations as belonging to a single entity is to refer to the causal agent. For many
infectious diseases (another example is syphilis), the complex and polythetic clinical
features can be unified in a single entity only because the causal agent can be identified
or its presence can be inferred by immunological means.
A univocally defined disease (on the basis of histology) such as lung cancer has several
causes (in fact, different sorts of “INUS” in which no necessary component, according
to current knowledge, is identifiable). Lung cancer cases have a certain histologic
appearance in common, but they cannot be unified on the basis of their etiology.
Finally, many psychiatric diseases are polythetic, both clinically and etiologically. For
example, bulimia cannot be univocally distinguished from other conditions
characterized by compulsive eating or obesity. As for causal hypotheses, they are still
extremely vague and apparently at the border between psychiatry, genetics and social
medicine.
In this way, we can group diseases in four groups, which express, so to say, the
epidemiologic transition from univocally defined, monocausal conditions to
etiologically polythetic diseases such as lung cancer, down to the emerging conditions
such as bulimia and other psychiatric or socially-determined disorders. It should be
stressed that situations such as smallpox are an exception, while the rule is represented
by “blurred” borders among different diseases, both on the side of symptoms and on the
side of causal pathways.
7. From the analysis I have conducted it is clear that in the history of medicine a variety
of concepts of cause have been used, that fit different purposes. Current medicine is
probably the most evident expression of the intertwining of different traditions. On one
side we have a rigorously empirical concept of proof, the one that dominates in the
clinical trials on the efficacy of drugs: in this case the “gold standard” is the randomized
trial, that is a strictly experimental approach based on the empirical comparison among
different groups of patients with a heavy use of the probability theory. In this case
attention is drawn towards a single cause because of the eminently practical (curative)
goal. Epidemiology itself is largely depending on probabilistic relationships, but it does
not refer any longer to single causes (not to speak of necessary causes), and in this sense
it is related to historical “conditionalism”. On the opposite, many types of research in
physiology and physiopathology are based on a counterfactual experimental logic, and
apply a conception of cause not very different from the one proposed by Mackie.
Finally, molecular research almost literally refers to Salmon’s idea of a a mark that can
be followed along time.
This heterogeneity does not imply conceptual confusion. Rather, I have the impression
that different causal models be used according to the context. When the goal is a
therapy, we wish to be as much effective as possible (partly for ethical reasons), and
therefore we stick to rigorous empirical proof and controlled experiments. The
epidemiologic approach to risk factors, although rigorous, is multicausal as a
consequence of the complexities that characterize the relationships among individuals
and their physical, biological and social environment.
Rather than facing incompatible models of causality, I suspect we are simply
considering different levels of reality. This is consistent with the theoretical framework
that has been offered by Schaffner for biomedical sciences (7). According to Schaffner,
biology is not characterized by “universal laws” (with exceptions such the genetic
code), but by “middle-range” theories, that is theories that are intermediate between the
simple observation of empirical regularities and universal statements on Nature.
Middle-range theories have the peculiarity of being based on a reciprocal reinforcement
among different types of observation, coming from different levels of reality and
including a reference to the basic laws of Nature. The two main characteristics of
middle-range theories are that they are expressed by time-related models (that is they
refer to processes such as carcinogenesis), and by “overlapping interlevel models”, i.e.
they aim at connecting different layers of reality. This is very clear if we consider the
evidence in favour of the relationship between smoking and lung cancer. In addition to
sound epidemiologic observations, we have several types of evidence referring to
different layers of reality. For example, molecular studies tend to show that point
mutations occurring in cancer genes (p53 or ras) can be specific for both the type of
cancer and the type of environmental carcinogen (the hypothesis of fingerprints as an
example of Salmon’s theory of the “mark”). In the mean time, the hypothesis that was
provocatively advanced by Sir Ronald Fisher, that genetic predisposition on one side
induces people to smoke, and on the other side increases their risk of cancer (thus
creating a spurious association between smoking and cancer) has been clearly refuted on
the basis of the observation of twins. This is a counterfactual conditional; the questions
is: what would happen if we eliminate the role of genetics by studying monozygotic
twins? In fact, the answer is simply that between two monozygotic twins, the one who
gets lung cancer is the one who smokes.
In conclusion, heterogeneity of the concept of cause in medicine should not be
surprising, but it should rather persuade us that the word cause itself changes its
meaning according to the different uses and contexts.
References
Mackie JL (1965) Causes and Conditions. American Philosophical Quarterly; 2, 245255.
von Engelhardt D. Causality and Conditionality in Medicine around 1900. In:
Delkeskamp-Hayes C., Gardell Cutter M.A. (Eds.) (1993) Science, Technology, and
the Art of Medicine, Kluwer, Dordecht.
Hempel C G (1966) Philosophy of Natural Science, Prentice-Hall, Englewood Cliffs,
NJ.
Nuland SB Doctors. The Biography of Medicine. Vintage Books, New York
W. Salmon (1984) Scientific Explanation and the Causal Structure of the World,
Princeton University Press.
Vineis P et al (1999) Human Cancer, Carcinogenic Exposures and Mutation Spectra.
Mutation Res 436, 185-194.
Schaffner KF (1993) Discovery and Explanation in Biology and Medicine, Chicago
University Press, Chicago.