Although promising, the synergies between declarative and procedural knowledge have been little explored in Medicine [1]. To tackle this problem, this paper proposes an approach that combines declarative and procedural knowledge. The approach uses ontologies as a knowledge artifact that expresses medical declarative knowledge. This ontology is used as a starting point for an approach that derives a BPMN procedural specification from the knowledge expressed in the ontology. The approach is illustrated in a infectious disease scenario, in particular, it uses the Basic Formal Ontology (BFO) [2] and the Infectious Disease Ontology (IDO) [3].
Expert systems have been used in Medicine successfully since 1990s [
Another strain of research in medical informatics considers the formalization of narrative
clinical guidelines into computer interpretable clinical guidelines (CIG) using formalisms such as
document models, decision trees, probabilistic models, task-network models [
Although promising, the synergies between declarative and procedural knowledge have been
little explored in both medical informatics and BPM communities [
The rest of the paper is structured as follows: Section 2 briefly introduces the BFO and IDO ontologies and procedural models in the medical sector, briefly introducing BPMN syntax. Section 3 presents the approach for deriving a BPMN procedural model starting from the ontologies, illustrating it in the scenario of infectious diseases. Section 5 concludes the paper and outlines future work.
Ontologies are a knowledge representation formalism that represent (or strive to represent)
reality in such way a group of stakeholders understand the terms that compose a certain domain
of discourse, and can thus learn about such domain [
In this paper, (medical) declarative knowledge is represented as ontologies. The Basic
Foundational Ontology (BFO) [
The BPM discipline is concerned with the formalization and analysis of the activities conducted
by an enterprise to produce goods and services to its customers [
Both types of processes can be typically captured by a procedural process language, such as the Business Process Modeling Notation (BPMN). Fig. 1 depicts a knowledge-intensive business process in healthcare. Rounded rectangles denote activities (e.g., verify the existence of infectious process), while circles represent events (patient’s temperature > 38 degrees). Diamonds represent decisions (presence or absence of infectious process?), while arrows denote diferent results of decisions.
This section depicts the approach of building process models based on the knowledge extracted from ontologies. As ontologies are commonly structured into foundational ontologies ℱ and domain ontology , our approach starts from a set of ontologies = ⟨ℱ , ⟩, where ℱ is the set of foundational ontologies and the set of domain ontologies. The approach is composed by the following steps: Step 1. The knowledge engineer starts by investigating the concepts of for a preliminary understanding of the domain.
Step 2. As the size of ontologies may be significant, the knowledge engineer selects the subset ′ from that is relevant for her modeling purposes. ′ may be sub-ontologies or even portions of .
Step 3. The knowledge engineer explores the semantics of each particular concept to grasp about the domain.
Step 4. The knowledge engineer builds the procedural representation. To perform such step, s/he considers the following sub-steps (not necessarily in this order): • One must take into account that activities are performed to react to events happening in the world. The hint is to look for the relevant events {1, 2, ..., } taking place within the domain • To express such events in the procedural specification, consider that the occurrence of events is captured by changes in the states of entities of the domain (concepts of the ontology). By searching for the concepts in the ontology and which changes in the state of these concepts are relevant, one can express the relevant events. • Order the relevant events. Include the events and the activity that has to be performed to address the event (change), including them all the procedural specification.
Step 5. As procedural and declarative knowledge present a complementary view of reality, it is possible that all knowledge required to build the procedural specification is not found in the declarative specification (and vice-versa). In this case, the knowledge engineer will have to complement the procedural specification with the help of domain experts.
Next section illustrates our approach in a infectious disease scenario.
This section illustrates the approach from Sec. 3 applied in a infectious disease scenario [
Step 1. Starting the approach, the set of ontogies related to the IDO have been investigated for a basic understanding of the infectious disease domain. Three ontologies (BFO, OGMS and IDO) have been identified. In this way we have = ⟨ℬℱ , ⟩, where = ⟨ℐ, ℳ⟩. Step 2. Multiple possibilities for selecting parts ′ of exist, given that IDO captures many distinct dimensions (biological scales, disciplinary perspectives and diferent phases). The disciplinary perspective of epidemiology has been chosen.
Step 3. The semantics of each particular concept relative to epidemiology has been investigated
at Table 5 and section "Epidemiology and surveillance" from [
Step 4. Fig. 2 depicts the BPMN specification built in our approach (the three traces inside the first activity denotes that the activity is performed multiple times, in this case, for each geographic region).
This specification has been built by analyzing the semantics of three IDO concepts ( infectious disease incidence, infectious disease pandemic and infectious disease epidemic) that are highlighted in boldface in BPMN. On top of identifying the concepts, we have identified the relevant events.
After understanding the semantics of the three concepts, following with the identification of the events, if the incidence of an infectious disease in a population (in a certain geographic region) is above a certain threshold1 (infection disease incidence > threshold1), this may indicate the existence of an epidemics in that region. With that, the public administrator has to check the existence of epidemics in other regions as well. If other regions have not surpass the acceptable threshold1 of infectious disease, then the number of regions with infectious disease is below threshold2 (#regions < threshold2), no signal of epidemics is found and the surveillance process ifnishes.
On the contrary, if a number of regions that population has infectious disease surpass a certain threshold (#regions > threshold2), this indicates a pandemics. In this case, the public administrator proceeds with the monitoring and two situations may happen. One is when we continuosly have a pandemics (#regions > threshold3). The second one happens when the number of regions drop and we no longer have a pandemics, but we may still have an epidemics (#regions < threshold3).
Step 5. Notice that the ontology captures only knowledge about the domain, but not the real-time data relative to the incidence of infectious diseases in population or the thresholds for epidemics and pandemics. In this case, public admnistrators have to provide such numbers.
This paper proposed a preliminary approach to derive a BPMN procedural specification from the knowledge expressed in an ontology. The approach is illustrated in an infectious disease scenario, using BFO (top-level ontology) and IDO (domain specific ontology). As the integration between declarative and procedural knowledge is generally challenging in many domains (not only Medicine), I hope that this preliminary approach provides an initial insight on how to solve this problem.
In particular, an advantage of this approach in the infectious disease scenario is the possibility
of integrating data to the overall approach presented in this paper. Two types of data may
be integrated, together with two distinct monitoring perspectives. One perspective considers
the usage of the values of incidence of infectious diseases in population and the thresholds
for epidemics and pandemics. With these values, it is possible to use epidemic simulators in
public health monitoring [
As future work, I consider this integration with data, together with the usage of more concepts from the ontology to derive more information for the procedural specification. Further, I also want to explore the synergy between declarative and procedural knowledge in the other way round, such as, can we derive the ontology from a procedural specification?