Identity Tracking of a Disease as a Causal Chain Kouji Kozaki1,*, Riichiro Mizoguchi1,*, Takeshi Imai2 and Kazuhiko Ohe2 1 ISIR, Osaka University, 8-1 Mihogaoka, Ibaraki, Osaka, Japan 2 Department of Medical Informatics, Graduate School of Medicine, The University of Tokyo, 7-3-1, Hongo, Bunkyo-ku, Tokyo, Japan ABSTRACT those processes. Such an entity (a disease) can change ac- This paper discusses the ontological treatment of identity of a disease cording to its phase while keeping its identity. On the basis based on an ontological model of diseases as causal chains. An individual causal chain changes through its processes such as extending, branching, of this observation, we defined a disease as: fading, etc. When we suppose that an individual disease is constituted by Definition 1: Disease (Mizoguchi et al., 2011) such a causal chain, changes of the causal chain imply changes of the dis- ease. Then, how should we deal with identity of a disease? To answer this, A disease is a dependent continuant constituted of one or we consider identity of a disease as a causal chain based on some ontologi- more causal chains of clinical disorders appearing in a cal theories. As a result, we propose an ontological model for identity human body and initiated by at least one disorder. tracking of a disease as a causal chain. This work should contribute to the ontological treatment of identity of diseases in medical information systems. Note that, although any disease has dynamic flows of the propagation of causality as its internal processes, it is the 1 INTRODUCTION enactor of its external processes, such as branching and ex- Recently, many medical ontologies have been developed for tending its causal chain of disorders. realizing sophisticated medical information systems, such as When we collect individual causal chains belonging to a OGMS (Scheuermann et al., 2009), DOID (Osborne et al., particular disease type (class), we are able to find a common 2009), and IDO (Cowell et al., 2010). They mainly focus on causal chain (partial chain) that appears in all of the instance the ontological definition of disease with related properties. chains. By generalizing such a partial chain, we obtain the We proposed a definition of a disease involving capturing a notion of a core causal chain of a disease as follows: disease as a causal chain of clinical disorders (Mizoguchi et Definition 2: Core causal chain of a disease1 al., 2011). This paper discusses the evolution of individual A sub-chain of the causal chain of a disease whose in- diseases, whereas the main concern in our previous work stances are included in all the individual chains of all in- was capturing static characteristics of diseases. An individu- stances of a particular disease type. It corresponds to the al disease undergoes changes as it evolves, and some of essential property of a disease type. these changes may cause new diseases to appear as a result. In some cases, some symptoms may appear as aftereffects Definition 2 provides a necessary and sufficient condi- lingering beyond the point when the original disease was tion for determining the disease type to which a given causal cured. Even if the disease can be cured completely, it could chain of clinical disorders belongs. That is, when an indi- appear in the patient's anamnesis. Here, the problem is how vidual causal chain of clinical disorders includes instances these changes should be dealt with based on our disease of the core causal chain of a particular disease type, it be- ontology. We discuss it from the viewpoint of identity track- longs to the disease type. We can thus define such a disease ing of a disease based on an ontological model of causal type that includes all possible variations of physical chains chains and a theory of identity (Kozaki et al., 2010). of clinical disorders observed for patients who contract the This paper is organized as follows. The next section out- disease. According to a standard definition of subsumption, lines the definition of a disease in our previous work. In we can introduce an is-a relation between diseases using the Section 3, we describe an ontological theory of identity chain-inclusion relationship between causal chains. tracking of a disease. In Section 4, we summarize some re- Definition 3: Is-a relation between diseases lated work. Finally, we present concluding remarks with a Disease A is a supertype of disease B if the core causal discussion of future work. chain of disease A is included in that of disease B. The in- clusion of nodes (clinical disorders) is judged by taking an 2 A MODEL OF DISEASES is-a relation between the nodes into account, as well as After it begins to exist, a typical disease, as a dependent sameness of the nodes. (Mizoguchi et al., 2011) continuant, enacts extending, branching, and fading pro- Assume, for example, that (non-latent) diabetes and cesses before it disappears. Thanks to these processes, a type-I diabetes are respectively defined as and . Then, we get according to Definition 3. (2) Non-cumulative process: a process that proceeds by There is another way to define is-a relations among dis- completing the current process at every instant in time. ease types. Assume that the core causal chain of arterio- A causal chain is composed of one or more pairs of entities pathic disease is . Then we can define a subtype disease of arterio- where the latter has been caused by the former. The effect pathic disease which has the core chain by specializing disorder of of multi-pair chains. We define three kinds of causal links artery to disorder of aorta. The specialization of disease from the perspective of “ongoingness”. types in the latter way is always sound because the new spe- Definition 4: Ongoing causal link cialized types include only instances of the original types This is a link that shows that the propagation of causality via chain-inclusion (the former way). On the other hand, from the cause to the effect is ongoing. generalization in the latter way does not always generate Definition 5: Historical causal link sound results via chain-inclusion because what we get will This is a link that shows that the propagation of causality include causal chains that do not go through the core causal from the cause to the effect has terminated. chain of its supertype. Definition 6: Pseudo-simultaneous causal link We understand there is a difficulty in defining main This is a link that shows only a pseudo-simultaneous rela- pathological/etiological condition(s) for each disease due to tion between cause and effect, excluding the temporal in- the variety of disease manifestations. However, at least we formation. It is used mainly for causal events that can be should be able to expect the existence of something com- viewed as having happened at the same time. mon to all instances of a disease type, otherwise we cannot Continuous propagation of causal effect requires that define the type. The core causal chain is such a common both causal and effect processes are ongoing. Therefore, thing that contains something corresponding to the main ongoing links can appear only between ongoing processes, pathological/etiological condition(s). That is, we would like and they correspond to causal relations found in cumulative to claim that the right terminal end (the disorder located at continuous processes; is a good the most downstream position) of the core causal chain of a example of this type of causal link. We believe that many of disease defined in a bottom-up way from a collection of the causal links found in diseases are also of this type. For individual causal chains should be understood as a kind of example, in the chain , both processes are definitely case, the core causal chain of the root disease of an is-a hi- ongoing. The same applies to cial case of a causal chain, and this should correspond to the which is a causal chain in myocardial infarction. On the main pathological/etiological condition(s) of the disease. other hand, might be problematic, since “loss of kidney function” is a past event and hence not ongoing. It is true A disease undergoes change during its course. The cause that “loss of kidney function” is an event, but after the event, may disappear, especially in its latter phase. Such disap- a new process or state “kidney is not functioning” is ongo- pearance might have a negative influence on determination ing, and this causes “increase of waste in the blood”. In of the disease of individual causal chains. Many other fluc- reality, a historical causal link appears between “loss of tuations of disorders might influence the causal chains, too. kidney function” and “kidney is not functioning”, and an Ontology of diseases should address such issues. One of the ongoing link appears between “kidney is not functioning” major issues related to such observations is identity tracking and “increase of waste in the blood”. The same applies to of diseases. To tackle this identity problem, we need to for- “virus invasion”, “steroid treatment”, etc. mulate the identity of a causal chain of clinical disorders. Readers might suspect that we have to write unnecessary Because the disease is a continuant, it can change while processes as follows: “finish traveling” and then “not travel- maintaining its identity. The theory of processes and events ing”. Of course, this is not the case. We are describing caus- adopted in this paper is found in the literature (Galton & al chains of clinical disorders which we believe explain a Mizoguchi 2009), in which the two key notions are (1) pro- disease under consideration, and hence we can select what cesses are intrinsically ongoing/in-progress whereas events to write as a disorder. Thus, we do not have to write pro- are not, and (2) an event is constituted of processes. cesses that are not necessary for explaining diseases. 3.1 Classification of causal links Although many of the causal links appearing in diseases are ongoing, historical causal links also should appear in There are two kinds of processes (Mizoguchi et al., 2011): 2 Identity Tracking of a Disease as a Causal Chain such cases where the causal events have terminated and the minated, then the chain becomes a historical causal chain. causality does not flow currently. A typical example is the Note here that whatever change happens, the identity of a event of a virus invasion that had caused an infection. An- causal chain remains the same as before it becomes a histor- other case is a cured disease consisting of a causal chain of ical causal chain. This topic will be discussed in Section 3.3. past events, which will be discussed in Section 3.2 in detail. Transition from an ongoing process to a terminated 3.3 Identity of a causal chain event can be explained in terms of the change from ongoing We begin our discussion on identity of a causal chain by causal link to historical causal link as follows: when the talking about identity of entities (Kozaki et al., 2010). There causing process has terminated, the process constitutes a are two types of identity: new event as a past event of the causation, together with a Instance identity: Identity for discussing the sameness of change of the ongoing causal link to a historical causal link. instances. A pseudo-simultaneous causal link is special. Let us as- Class identity of a thing: Identity for discussing the same- sume a series of events associated with a vase, such as “fall ness of the class to which the entity belongs. onto the floor, collision with the floor, break, and spill wa- ter”. It is a common phenomenon and corresponds to a non- In order to talk about identity of a disease instance, we cumulative process. We take two events from it: collision need to establish identity of a causal chain of clinical disor- and breakage events. It seems like the two events happened ders. So, what we should discuss is under what conditions at the same instant in time. Precisely speaking, however, the instance of a causal chain loses identity. these two events need non-zero time intervals, and the colli- After it has been born, an individual causal chain of clin- sion happened before the breakage. In spite of this reality, ical disorders grows, branches, etc. and it may reach another people tend to see that they happened at the same time with disease. Thanks to appropriate remedies, its causal disorders the understanding that the collision caused the breakage. A disappear, whereas some downstream disorders are still on- pseudo-simultaneous causal link is introduced to represent going. During such a change, the disease, that is, the causal such a causal relation. It does not include a non-zero tem- chain of clinical disorders, keeps its identity. For example, poral interval but deals with the two events as if they hap- even if diabetes in a patient causes a cerebral infarction, the pen at the same time while keeping the causal order. It also clinical causal chain remains the same; that is, no new caus- can be interpreted as a degenerated causal link that includes al chain appears because of it. Furthermore, even if an after- the change of an ongoing causal link to a historical causal effect remains after all of the main causes have disappeared, link and is useful for representing causal events composed we should be able to identify it as the same causal chain in a of non-cumulative processes. different state. In order to deal with such changes properly, we employ Identity for replacement (I-Rep) among the four 3.2 Classification of causal chains kinds of identity discussed in our previous work (Kozaki et We introduce three kinds of causal chains to capture the al., 2010). I-Rep is defined as: change of diseases according to the causal links as follows: Identity for replacement (denoted as I-Rep): Definition 7: Strongly connected causal chain Identity wherein an instance of the whole continues to be Any causal chain all of whose causal links are ongoing itself without becoming another thing while its parts are causal links; that is, it is composed of only ongoing pro- being replaced, independently of the kinds and number of cesses. A causal chain composed of one ongoing process replaced parts. is included as a special case. Definition 8: Weakly connected causal chain: A typical example for the use of I-Rep is a bike whose Any causal chain that includes at least one historical or parts are replaced one by one. Whatever parts are replaced pseudo-simultaneous causal link and at least one ongo- and in whatever number, a bike whose parts are replaced ing process. remains the same in terms of I-Rep. Exactly the same ap- Definition 9: Historical causal chain: plies to causal chains of clinical disorders. Some readers Any causal chain all of whose causal links are historical might notice that we could use Identity for essentiality (I- or pseudo-simultaneous causal links. That is, it is a Ess for short) instead of I-Rep. I-Ess is defined as identity causal chain composed only of terminated events. determined by the essential property of an individual. If you use I-Ess with, say, a saddle that you love as its essential Definition 10: Ongoing causal chain: Any causal chain that has at least one ongoing process. property and parts of your bike are being replaced as in the case above, then you would say, “This is not my bike any It subsumes the strongly connected and weakly connect- more,” when the saddle has been replaced with a new one. ed causal chains. In the case of diseases, for example, an individual causal When an ongoing process of a strongly connected causal chain of diabetes has the essential property of an elevated chain has terminated, the whole chain becomes a weakly level of glucose in the blood. If its identity is defined in connected causal chain composed of two strongly connect terms of I-Ess for that property, then it would become an- causal sub-chains. If all of the ongoing processes have ter- other causal chain when the glucose level becomes normal 3 Kozaki, K., Mizoguchi, R., Takeshi, I., and Ohe, K. because it loses its identity. However, this is not appropriate We associate two kinds of identity with each individual because identity of the causal chain itself should be kept causal chain of disorders, namely, instance identity and unchanged as long as some disorders caused by the diabetes class identity. At first glance, class identity seems to be use- remain. Furthermore, even if only an aftereffect is left, it ful to talk about when a disease is cured. However, this is should be identified as the same causal chain in a different not the case, because a causal chain keeps the same instance state. This is why we employ I-Rep instead of I-Ess. identity and remains being the same disease as long as there Class identity, on the other hand, is used to talk about to are still ongoing downstream disorders/symptoms even what type an individual causal chain belongs. As stated in when all the disorders included in its core causal chain have Definitions 1 and 2, an individual causal chain of clinical been cured. When all disorders included in a causal chain disorders Ci is identified as belonging to a disease type DA are cured, the ongoing causal chain disappears and a new when it includes an instance of the core causal chain of DA historical causal chain is born. As discussed already, be- as a part, and then, it gains class identity for DA. This Ci is cause the newly born historical causal chain is a different said to be an ongoing disease DA if it is an ongoing causal individual from the corresponding ongoing causal chain, it chain. Ci can get another identity for a different disease should have a different instance identity. However, its class when it also includes an individual core causal chain of an- identity should be the same as that of the original causal other disease, say, DB. Class identity is kept until the ongo- chain, since a previous disease that appeared in the patient's ing causal chain becomes a historical causal chain. There- anamnesis must be the same as the disease of the corre- fore, the extinction condition of causal chains in terms of I- sponding ongoing one. In summary, the historical causal Rep and I-Ess is the same. chain should have the same class identity as the original Let us discuss conditions for extinction of a causal chain disease. Considering various cases concerning ongo- in terms of I-Rep. Any individual causal chain, Ci, is born as ing/historical states, there can be several states for a causal a single ongoing process, that is, as a strongly connected chain. We can identify the following three major states: causal chain. It never changes its I-Rep identity by replace- ment of its parts. Neither growth nor branching can change 1) Ongoing: an individual causal chain Ci of disease DA its identity. When an intermediate disorder terminates to whose core causal chain is still ongoing. become an event, it becomes a weakly connected causal 2) Curing: Ci as a whole is ongoing but the core causal chain, keeping its I-Rep identity. In other words, it does not chain of disease DA of Ci is inactive (has become a change its I-Rep identity if it has at least one ongoing disor- historical causal chain). der. Now, we discuss what happens when all the ongoing 2.1 A few partial causal chains in the downstream processes have terminated and hence all the links have be- are still ongoing. come historical causal links. According to previous work 2.2 No ongoing causal chains but a few long-term (Galton and Mizoguchi, 2009), an event is constituted of remaining disorders that have no influence on processes, and when all associated processes have terminat- others (aftereffects). ed, it finally appears as a whole as the result of constitution 2.3 Although disease DA seems to have been cured of past (terminated) processes. Because processes and at a certain time, it might recur in the future. events are different categories, the newly created event 2.4 In a case of cancer, it might metastasize to other should have different identity from that of the processes. regions, etc. When all the links have become historical causal links, 3) Cured: Ci as a whole is a historical causal chain. therefore, the resulting historical causal chain must be a different instance (chain) from the ongoing causal chain The first and last cases are obvious. The second case is because all of its nodes and links are different from the orig- vague in its nature. It may not be clearly defined, especially inal chain (chain before the change). That is, when all disor- when doctors consider aftereffects, recurrence, metastasis ders included in an ongoing causal chain have gone (been etc. We would like to cover such cases as states in which the successfully treated), the chain loses its identity and be- core causal chain becomes a historical causal chain while comes a historical causal chain of different identity, which the chain as a whole is ongoing. However, there is room to corresponds to curing the disease, and the resulting histori- investigate more-detailed classification of these cases. cal causal chain represents the anamnesis of the patient. 3.4 Identity tracking of a disease Those past processes constituting the events represent how As a summary of the discussion made thus far, we discuss the previous disease appeared in the patient's anamnesis examples of the life of an individual disease as a causal and how the disease was cured. chain of clinical disorders, as shown in Fig. 1. Note here Note here that a change of part of the ongoing causal that we are not trying to predict the future state of a particu- chain does not influence the identity of the whole chain lar disease of a particular patient, but to exemplify possible thanks to the I-Rep identity; that is, an ongoing causal chain changes of diseases in the real world represented as a causal remains the same chain after curing a few disorders in the chain of clinical disorders in terms of its I-Rep identity and chain before all of the disorders have been cured. type membership. 4 Identity Tracking of a Disease as a Causal Chain We do not discuss cases where pseudo-simultaneous (6) When a partial chain of Ci corresponding to the core causal links appear, since these should be dealt with in a causal chain of diabetes and diabetic retinopathy be- similar way to the case of historical causal links. comes a historical causal chain (the associated disorders (1) A causal chain of clinical disorders, Ci, is born as a have disappeared), we might say that the patient is strongly connected causal chain. At the same time, it cured of the disease. However, in some cases, a few gains its instance identity (Iid_1). Ci can extend, branch, disorders/symptoms might still remain, and hence Ci etc. while keeping its identity. Before it establishes it- might still be ongoing. For example, some destroyed self as an instance of a particular disease type, its class nerve tissue might be left after that change. identity remains undetermined. (7) When all of the processes in Ci have terminated and (2) When Ci extends and includes an instance of the core become events, and hence all of the causal links be- causal chain of the disease diabetes as its partial chain, come historical links, Ci becomes a historical causal it becomes an instance of diabetes and gets class identi- chain as a new individual causal chain, all of whose ty Cid_A for diabetes. nodes are events constituted of related past processes. (3) Once Ci gets class identity for a disease type, it never loses it until Ci disappears, and the class identity re- 4 CONCLUSION mains unchanged independently of its extension and/or We have discussed and proposed an ontological account of branching because it still includes the instance of the the evolution of a disease as a causal chain of clinical disor- core causal chain of the disease. ders. As far as the authors know, there is little activity re- (4) Ci can get more than one class identity, say, Cins_B for garding ontology for the evolution of diseases in DOID and diabetic retinopathy. Ci may extend further and cause OGMS, whose main focuses are definitions of related types. serious disorders/symptoms, which could lead to the pa- Although OGMS defines “disease course”, it is dealt with as tient's death. a process, whereas we discussed it in terms of a causal chain (5) After some changes of the disorders as well as the ter- as a dependent continuant. We believe that, together with mination of the ongoing process to become an event, re- our previous paper (Mizoguchi et al., 2011), we have made lated causal links of Ci change from ongoing links to a step toward a comprehensive ontological account of dis- historical links. Then, Ci changes from a strongly con- eases, covering their evolution. Other researchers (Galton et nected causal chain to a weakly connected causal chain. al., 2005) have discussed causal relations amongst events, Nevertheless, Ci keeps its instance identity as an ongo- states, and process. Although our notions about the kinds of ing causal chain. causal links share some ideas with theirs, the definition of Strongly-connected causal chain identity of a causal chain is a unique feature of our theory. Legend (1) Iid_1 Disorder (ongoing) Strongly-connected causal chain ACKNOWLEDGEMENTS Diabetes Iid_1 Disorder (terminated) A part of this research is supported by the Japan Society for (2) Cid_A the Promotion of Science (JSPS) through its “Funding Pro- Strong causal link Deficiency Elevated level of (ongoing) gram for World-Leading Innovative R&D on Science and of insulin glucose in the blood Weak causal link I _1 Strongly-connected causal chain ins (has been caused by) Technology (FIRST Program)." Cid_A Core causal chain (each color represents a (3) … disease) REFERENCES Strongly-connected causal chain Iid_1 Scheuermann, R. H., Ceusters, W., and Smith, B. (2009) Toward an Onto- Cid_A Cid_B logical Treatment of Disease and Diagnosis. Proc. of the 2009 AMIA Diabetic retinopathy Summit on Translational Bioinformatics, 116-120, San Francisco, CA. (4) … Galton, A. et al. (2005) Processes and Events in Dynamic Geo-Networks. Proc. of GeoS 2005, LNCS 3799, 45–59, Mexico City, Mexico. Weakly-connected causal chain Iid_1 Cid_A Cid_B Galton, A. and Mizoguchi, R. (2009) The water falls but the waterfall does not fall: New perspectives on objects, processes and events. Applied (5) … Ontology 4(2), 71-107. Weakly-connected causal chain Iid_1 Cid_A Cid_B Osborne, J. D., et al. (2009) Annotating the human genome with Disease Ontology. BMC Genomics 10(1):S6. (6) Cowell, L. G. and Smith, B (2010) Infectious Disease Ontology. Infectious … Disease Informatics, Chapter 19, Sintchenko V., 373-395. Destroyed nerve tissue Mizoguchi, R., et al. (2011) River Flow Model of Diseases, Proc. of Historical causal chain ICBO2011, 63-70, Buffalo, USA. Historical causal chain Iid_２ Constituted by Kozaki, K., Endo, S., and Mizoguchi, R. (2010) Practical Considerations on Identity for Instance Management in Ontological Investigation, Proc. (7) of EKAW2010, LNAI6317,16-30, Lisbon, Portugal. Fig. 1. An example of the life of an individual dis- ease as a causal chain of clinical disorders. 5