-Diseases, risks of pathological processes and predispositions have been formalized as dispositions. The relations between those pathological dispositions, however, remain unclear. We apply here a recently developed theory of mereology and identity among dispositions to analyze such relations. In particular, we show how a framework for the identity of dispositions leads to a disease being realized not only by its disease course but also by each of its pathological process; how it avoids risk multiplicativism; and how a predisposition can be identified with a risk whose estimated probability is higher than the probability of the risk for a reference class. We discuss how this makes a predisposition always relative to a reference class, to a time-frame and to sources of risk estimates; and we clarify the nature of risk factors.
Dispositions are entities such as fragility, solubility, or
vulnerability to poison, which can be triggered by some process,
leading to a realization process. For example, the fragility of a
glass might be triggered by a shock and realized by the process
of this glass breaking. Dispositions may exist even if they are
not realized or even triggered: this glass if fragile even if it never
breaks, or even if it never undergoes any shock. Dispositions are
omnipresent in the clinical domain. An obvious example
concerns predispositions to diseases. But other entities have also
been formalized as dispositions, such as diseases themselves (by
the Ontology for General Medical Science, OGMS [
Reference [
This paper also introduced, for all practical purposes, a
criterion of identity (named here “ID”) acceptable for all
practical purposes in both ONLY and PARTHOOD, which
states that two dispositions d and d’ are identical iff they have
the same categorical basis, the same class of triggers and the
same class of realizations; using relations as defined in [
Reference [
Reference [
We will apply those theories of identity and mereology to several questions: is a disease realized by its disease course, by each of its pathological processes, or by all of those? Are several risks the same entity or different ones? And finally, what are the connections between a predisposition to a disease, the general risk to this disease, and the risk factors of this disease?
Consider Mr. Miller, who is epileptic. Diseases are formalized in OGMS as dispositions realized by various pathological processes that are parts of a disease course. Suppose that over the course of his life, Miller undergoes n epileptic crisis: crisis1, …, crisisn. The epileptic disease course of Miller has as parts all those epileptic crises (and possibly pathological processes of other types, though we will ignore them here): epileptic_dc0 has_part crisisi for every i between 1 and n. Suppose moreover that Miller is a photosensitive epileptic, and that for every i between 1 and n, a flash of light flashi triggers his crisisi. Consider now how each of the two ontological frameworks would formalize this situation.
In ONLY, one could distinguish two dispositions borne by Miller, that might each be identified with the epileptic disease: epilepsy0, which is realized once (over an extended time period) by Miller’s whole disease course, and epilepsy1, which is realized multiple times by each of Miller’s individual epileptic crisis: epilepsy0 has_realization only Epileptic_disease_course epilepsy1 has_realization only Epileptic_crisis More specifically: epilepsy0 has_realization epileptic_dc0 epilepsy1 has_realization crisisi (for i between 1 and n) epilepsy0 and epilepsy1 are different, as they do not have the same realizations. More specifically, epilepsy1 is a proper addpart of epilepsy0. Indeed: • the bearer of epilepsy0 is the same as the bearer of epilepsy1. • any trigger of epilepsy1 (a light flash) is a part of the trigger of epilepsy0 (the mereological sum of all light flashes). • any realization of epilepsy1 (a pathological process of epileptic crisis) is a part of the realization of epilepsy0 (the disease course of the epilepsy).
The problem of this formalization is that it artifically introduces two dispositions epilepsy0 and epilepsy1. These two dispositions have the same categorical basis: the qualities that are responsible for the realization of the epileptic crises are exactly those that are responsible for the realization of the mereological sum of those epileptic crises. However, the medical science does not seem to distinguish those two different dispositions, but would rather identify them. A more economical option would therefore be to introduce an alternative disposition epilepsy2 which is realized by the disease course, but also by each pathological process. We will now show that this is enabled by the second ontological framewok PARTHOOD.
In PARTHOOD, any part of a realization of a disposition d is also a realization of this disposition. Therefore, if we postulate that epilepsy2 has_realization epileptic_dc0, then for every i between 1 and n, crisisi is also a realization of epilepsy2: epilepsy2 has_realization crisisi.
• epilepsy2 has_trigger flashi • epilepsy2 has_realization crisisi • epilepsy2 has_realization epileptic_dc0
We can also add that Miller’s epileptic disease course is a
maximal realization of epilepsy2. These relations involving
particulars can be generalized into relations between classes as
represented in the figure 2 below (using relations defined in
[
Note that in PARTHOOD, one cannot say that ‘Epilepsy has_trigger only Flash’ – because any process that has a Flash as a part is also a trigger of the epileptic disease; however, we can state an axiom concerning a subclass of Flash named
Such an ontological framework seems preferable: we do not have to duplicate artificially the disease into a disposition realized by the disease course and a disposition realized by each pathological process.
In [
RiskAF,6m,Stroke was left open. We will analyze it in the two frameworks of identity of dispositions.
The framework ONLY raises two issues for the
formalization of risks developed in [
However, it is not the case that everything that happens in the body of a person has a causal influence on the possible realization of his risk. Therefore, some proper parts of instances of H-P12m (the ones that are causally relevant) should also be triggers of RiskAF,12m,Stroke.
RiskAF,12m,Stroke cannot be identical with each others. RiskAF,Stroke
is triggered by all history-parts of its bearer (that is, all parts of
his history – which is the mereological sum of all processes
taking place in or on his body [
Consider a patient Jones who has AF, and his risks
riskJones,Stroke, riskJones,12m,Stroke, and riskJones,6m,Stroke (as defined
by [
Moreover, both riskJones,6m,Stroke and riskJones,12m,Stroke are mod-parts of riskJones,Stroke (that is, Jones’ 6-months risk of stroke and 12-months risk of stroke are two of the many modes of realization of his general risk of stroke); and indeed: • the bearer of riskJones,Stroke, riskJones,12m,Stroke, and riskJones,6m,Stroke is the same, namely Jones. • any trigger of riskJones,12m,Stroke or riskJones,6m,Stroke is a trigger of riskJones,Stroke. • any realization of riskJones,12m,Stroke or riskJones,6m,Stroke is a realization of riskJones,Stroke.
Finally, one can introduce various risk estimates, such as risk_estimateJones,12m,Stroke, which is_about riskJones,12m,Stroke (alternatively, we can introduce the relation object_of inverse of is_about, and state that riskJones,12m,Stroke is the object_of risk_estimateJones,12m,Stroke).
Thus, the relations between the various risks can be described by the theory of mereology of dispositions. However, this leads to risk multiplicativism: for any class of time intervals with a given length of time t, there is a corresponding disposition riskJones,t,Stroke.
The framework PARTHOOD helps to solve the two issues mentioned above. As a matter of fact, in PARTHOOD, a disposition does not necessarily have only one class of trigger (or one class of realization). Thus, it does not exclude that proper parts of instances of H-P12m (in particular, the causally relevant parts) would be triggers of RiskAF,12m,Stroke. Moreover, in PARTHOOD, any process that has as part a trigger is also a trigger. Thus, if a particular 6-months-long history part h-p6 is a trigger of a disposition, then a particular 12-months-long history part h-p12 that has h-p6 as part is also a trigger of this disposition. As a consequence, we cannot define as earlier entities such as riskJones,6m,Stroke or riskJones,12m,Stroke. However, we can still define riskJones,Stroke, which can be triggered (in particular, but not only) by history-parts of Jones of any time length. To relate risk_estimateJones,6m,Stroke with its 6-months time frame, we can assert it as being not only about riskJones,Stroke, but also about the set of triggers H-PJones,6m. That is: riskJones,Stroke object_of risk_estimateJones,6m,Stroke
and History-partsJones,12m: riskJones,Stroke object_of risk_estimateJones,12m,Stroke
This ontological framework is more economical than the former one, as it introduces only one risk of stroke for Jones (rather than one for each time-length).
Note that in PARTHOOD, the relations between riskJones,Stroke, StrokeJones and H-PJones are different from the relations that hold between those entities in ONLY. For example, in ONLY, riskJones,Stroke would be realized only by instances of StrokeJones; whereas in PARTHOOD, riskJones,Stroke would be realized by all processes that are parts of a maximal realization, and the latter always has as part an instance of
In OGMS, a predisposition to disease of type X is defined as “a disposition in an organism that constitutes an increased risk of the organism’s subsequently developing the disease X.” This definition introduces three dispositions: the predisposition itself, the risk of the organism to develop disease X, and the disease X. Which of those are identical?
First, a predisposition to X is not identical with a disease X: one can have a predisposition to X without ever having a disease X. For the same reason, a risk to X is not identical with a disease X. But what is this relation of “constitution” between a predisposition and a risk? To answer this, we will analyze a specific example.
Consider Mrs. Roy, who has an increased risk of breast cancer because of a BRCA1 mutation. We will distinguish here a causal risk factor from a general predisposition. First, one may say that Mrs. Roy has a causal risk factor to breast cancer because of her BRCA1 mutation. However, despite this mutation, she might have other features (maybe other genetic mutations) that protect her against breast cancer, and therefore she might have overall a lower risk of breast cancer than the average women of the same age group. In such a case, one would not say that she has a general predisposition to breast cancer, taking into account all her features. However, if she has overall a higher risk of breast cancer than other women of the same age group, then she has such a general predisposition to breast cancer relatively to those women. We will analyze in turn the nature of her general predisposition and of her risk factor.
Let’s consider first Mrs. Roy general predisposition to breast
cancer, and show that it can be identified with the risk of breast
cancer for Mrs. Roy noted riskRoy,BC. Following [
More generally, consider PredBRCA1,BC the class of predispositions of women with the genetic mutation BRCA1 to get a breast cancer. By the same reasoning as above, it can be identified with RiskBRCA1,BC, the class of risks of women with the genetic mutation BRCA1 to get a breast cancer.
But when is a risk a predisposition? Not all risks are predispositions: for example, people with drepanocytosis have – like everyone – a risk to get malaria if bitten by a malariainfected mosquito, but their risk is lower than the risk of people who do not have drepanocytosis; thus, their risk to get malaria cannot be called a “predisposition”. And indeed, OGMS suggests that a predisposition is “A disposition in an organism that constitutes an increased risk of the organism’s subsequently developing the disease D.” Thus, it is important to capture this notion of “increased” risk. A natural proposal is to consider that a person p has an increased risk to disease D if she has a higher probability to get D than the probability to get D for a class of persons People1 of which p is an instance (although this is not the only option: uncertainty might be quantified with other tools than probability).
Probabilities can be analyzed through an epistemic account,
as proposed by [
We can compare this risk estimate to the risk estimate assigned to a class People1 of which Mrs. Roy is an instance. For example, if Mrs. Roy is a 54 years-old Quebec woman in 2018, People1 might be the (defined) class of women between 50 and 60 years old in Quebec in 2017. A risk estimate might then estimate a class of risks borne by the class of persons People1:
Then, we can say that Mrs. Roy has a predisposition to breast cancer if pRoy,BC > pPeople1,BC.
Note that to estimate riskRoy,BC, we would usually also use a risk estimate assigned to a class of which Mrs. Roy is an instance. Suppose for example that Mrs. Roy has not only a mutation BRCA1, but also another genetic protective factor against breast cancer called X. Consider People2 the subclass of People1 who have BRCA1 mutation; and People3 the subclass of People2 who also have the protective factor X. Then, we might provide an estimate of riskRoy,BC with a probability equal to the probability of a risk estimate of riskPeople3,BC (that is, pRoy,BC = pPeople3,BC). This shows that whether a risk is a predisposition or not depends on the choice of a reference class: Mrs. Roy has a predisposition to breast cancer relatively to the class People1 if pRoy,BC > pPeople1,BC, but not relatively to the class
Additionally, even when a reference class such as People1 is chosen, the status of predisposition of a risk depends on the risk estimates that warrants the probabilities such that pRoy,BC > pPeople1,BC. One could imagine a scenario with two different risk estimates of the risk of breast cancer for People1, and two different risk estimates of the probability of breast cancer for Mrs. Roy as follows: • risk_estimatePeople1,BC1 has_value pPeople1,BC1 • risk_estimateRoy,BC1 has_value pRoy,BC1 • risk_estimatePeople1,BC2 has_value pPeople1,BC2 • risk_estimateRoy,BC2 has_value pRoy,BC2 and such that pPeople1,BC1 < pRoy,BC1, pPeople1,BC2 > pRoy,BC2: in such a case, Mrs. Roy has a predisposition to breast cancer relatively to People1 and the two first risk estimates, and has no such predisposition relatively to People1 and the two last risk estimates.
For example, risk_estimatePeople1,BC1 might be drawn from a first journal article and risk_estimatePeople1,BC2 from a second journal article; and risk_estimateRoy,BC1 might be made by a first doctor who examined Mrs. Roy, and risk_estimateRoy,BC2 by a second doctor who examined Mrs. Roy. Alternatively, it might be the same doctor who drew two risk estimates risk_estimateRoy,BC1 and risk_estimateRoy,BC2, by basing them on two different classes, e.g. basing risk_estimateRoy,BC1 on risk_estimatePeople3,BC and risk_estimateRoy,BC2 on risk_estimatePeople2,BC (although the doctor would presumably rely in priority on risk_ estimateRoy,BC1, as it is based on the most specific reference class People3).
When we compare the risk estimate for Mrs. Roy and the risk estimate for a class she is an instance of, both risk estimates need to be made on the same time-frame; that is, both risk estimates need to be about history-parts of the same time-length. Indeed, Mrs. Roy certainly has a higher probability to get breast cancer over 30 years than an unspecified woman would have over one day, but this does not prove that she has a predisposition to breast cancer. It is at least theoretically possible that Mrs. Roy would have a predisposition on a 12-month frame to get breast cancer relatively to the class of women in Quebec between 50 and 60 years old, but would have no such predisposition on a 6-month frame relatively to the same group (if, for example, she has a mild protective factor during the first 6 months that eventually disappear and is replaced by a strong risk factor during the following 6 months).
Note also that not all classes of people would typically be the object of a risk estimate. While a risk estimate might be attributed to the breast cancer risk for the very general class of Women, its clinical use would be quite limited given the high variability of individual risks within this class, especially across various ages.
Thus, we can define a predisposition of o to D as “A risk in an organism o of developing a disease D about which there is a risk estimate with a probability higher than the probability of a risk estimate on the same time-frame about the risk of developing D for a class of organisms of which o is an instance.”
We can now turn to causal risk factors, as introduced earlier. There is indeed a holistic aspect in predRoy,BC: it is the general predisposition of Roy to have breast cancer, to which causally contribute all her relevant causal risk factors and causal protective factors (we specify “causal” here, because the definition of “risk factor” sometimes encompass correlational factor that have no causal influence). However, as mentioned earlier, we might want to represent those risk factors. Suppose now that Mrs. Roy has actually two gene mutations increasing her risk of breast cancer: BRCA1 and BRCA2. Then, we could introduce the risk factor rfRoy,BC,BRCA1 of Mrs. Roy to have breast cancer because of her BRCA1 mutation, as well as her risk factor rfRoy,BC,BRCA2 to have breast cancer because of her BRCA2 mutation. Both are instances of RFRoy,BC,BRCA2 such that:
rfRoy,BC,BRCA1 has_basis BRCA1Roy rfRoy,BC,BRCA2 has_basis BRCA2Roy
We can quantify the effect of such risk factors as e.g. the
difference between the probability assigned to a class with this
risk factor and the probability assigned to a class without this
risk factor (that is, the “absolute risk increase” [
OGMS defines a “genetic predisposition to disease of type D” as “A predisposition to disease of type D whose physical basis is a constitutional abnormality in an organism’s genome.” However, this definition rather fits with what we defined as a risk factor. Thus, we suggest to define risk factors and protective factors as follows:
that causally contributes to increase the probability of a risk of an organism to develop the disease D.”
disposition that causally contributes to decrease the probability of a risk of an organism to develop the disease D.” To qualify as a reference class for a risk r, R must be a class whose r is an instance (e.g. on Fig. 5, RiskPeople1,BC for riskRoy,BC).
We see that introducing the framework for the identity of dispositions PARTHOOD enables a more economical ontological framework for representing diseases, and suggests a more economical framework for the representation of risks. Moreover, the criterion of identity ID suggests to identify the general predisposition to a disease with the risk to this disease, when this risk is estimated to have a higher probability than the risk associated with a reference class. We have contrasted general predispositions with causal risk factors.
Note that a good understanding of the nature of risk factors,
protective factors and general predispositions are of crucial
importance for precision medicine. Indeed, a trend sometimes
criticized in medicine is to ground prevention or therapy on risk
•
•
•
•
factors, rather than on a general risk level [
This analysis also points that the term “predisposition” may have a wider definition than the one presupposed by its regular use. A person is often deemed to be “predisposed” to a disease if she has a risk with a higher probability than the risk of a group to which she belongs of the same age-group and nationality. However, one could imagine to state that a person is “predisposed” to have a disease relatively to, say, younger persons, or persons from a different national group.
Finally, risk factors and protective factors can interact in multiple ways to contribute to a general risk (that might be – or not – a predisposition). Future works should analyze how they can combine in various specific examples to illustrate the large diversity of possible interactions.
We thank Ludger Jansen and Alan Ruttenberg for useful discussions on dispositions, as well as anonymous reviewers for their comments. Adrien Barton and Olivier Grenier thank the CIHR-funded “Quebec SPOR Support Unit” for funding.