Medical records are crucial resources for every aspect of healthcare practice. The amount and complexity of the information they bear require the use of automation. In this paper we propose a method for separating and classifying the information available in medical records, drawing on Karl Popper philosophical theories. We test this method by using descriptions of clinical cases within the scope of a biomedical project that deals with the human T cell lymphotropic virus. Our goal is to come up with a framework that allows for the organization and sharing of information in knowledge representation ontologies according to their epistemological or ontological nature.
Some kinds of information, which are relevant for the medical field, cannot be
properly represented following realistic guidelines. Examples are claims about the
characteristics of signs and symptoms, which we name “epistemological information”
[
In previous papers [
We have been conducting the investigation that is the object of this paper,
written within the scope of a biomedical project, specifically focused on human blood.
The goal of this biomedical project is the development of a knowledge base for
scientific and educational applications related to the human T cell lymphotropic virus
(HTLV). The basis of infection by HTLV is not well-established [
The remainder of this paper is organized as follows: section 2 describes the theoretical basis of our investigation, presenting Popper´s theories. Section 3 contains our strategies for analyzing real data and the methodological steps taken. Section 4 presents the results of the application of our framework to real medical records. Finally, in section 5 we present a discussion and future works.
In this section, we present the theoretical background employed as a basis for our investigation. Section 2.1 describes, briefly, the theory of three worlds of the Karl Popper and the section 2.2 his theory about truthlikeness and fallibilism.
A useful approach combining reality, cognition and representations was proposed by
Popper in his theory of three worlds. Popper proposes a pluralist view of the universe
that recognizes at least three different but interactive worlds [
According to Popper, there is a world that consists of physical bodies, such as stones, plants and animals, which is called world 1. World 1 can be divided into the world of non-living physical objects and the world of living things or biological objects. There is the mental and psychological world, called world 2, which includes thoughts, perceptions and observations, that is, the mental and psychological processes and subjective experiences. In world 2 we can distinguish conscious experiences from dreams, or distinguish human consciousness from animal consciousness. There is also another world, called world 3, which includes all content of world 2 mental processes, such as languages, scientific theories, mathematical constructions, symphonies and sculptures. While a block of marble pertains to world 1, the creation by an artist of a sculpture using this block is a manifestation in world 3. From an ontological perspective, one can claim that world 2 and world 3 are evolutionary products of world 1.
Popper´s three worlds theory has been applied to investigations in health
information science [
While Popper´s ontological view allows one to better understand the relation
between entities pertaining to the world, his epistemological view proposes that every
conceptualization reveals mismatches between reality and theories about reality.
Though Popper’s theories have been criticized [
In addition to the three world’s theory, Popper is also known for his falsifiability criterion and for his advocacy of fallibilism. According to the falsifiability criterion, scientific hypotheses are falsifiable and, therefore, scientists are able to state what empirical findings make such hypotheses false. Fallibilism is the view that no presumed knowledge, not even scientific knowledge, is absolute certain.
In this line of thought, epistemological searches are fallible. As human
knowledge is incomplete, probable, and conjectural, one should seek truth but expect
truthlikeness. Truthlikeness is a qualitative measure of how a theory can be more or less
close to truth [
The medical practice is still heavily grounded in the study of signs and
symptoms, which are interpreted by a physician. Medical reasoning is a sum of different
cognitive practices including induction, abduction and deduction [
In order to develop better possibilities for medical record representation, we need to organize the kinds of information they contain. The method we propose here is composed of the following four steps.
First, we develop an analysis framework, which draws on inputs from Popper´s
three worlds and we also researched by recent medical ontologies, namely, the Basic
Formal Ontology (BFO) [
As a second step, we are testing such framework on real medical records under
evaluation in a biomedical project about blood diseases [Almeida, Proietti and Smith
2011]. In this paper, we will present as an example a clinical case description, which is
considerably clearer than real medical records, while still requiring proper
representation of the full range of medical entities. We use a generic clinical case
available at
In order to identify propositions within the clinical case, a domain expert
transcribed the records into sentential fragments that make sense to him. The domain
expert was asked to identify the reason for recording those entities and the information
that is being conveyed by the representation. The transcription draws upon principles of
logic and controlled languages [
As a third step, we consider an alternative for measuring truthlikeness, in order
to classify epistemological information that came from the selected records. We took
the position that epistemological information relevant in the context of medical practice
cannot be registered in an ontology as a universal, following the tenets of the adopted
realistic methodology [
Finally, as fourth step, we organize the information from the medical records into four types, which are then employed in order to recommend both a data arrangement and a scenario for collaboration among different representations.
In this section, we present the analysis framework created to organize information present in a medical record (section 4.1) and, in the section 4.2, we conduct a preliminary test of the framework by analyzing individual information entities contained in examples of the medical records.
We propose the analysis framework depicted in Fig. 1, which was created to organize
information present in a medical record according to the best possibility for
representation. This framework is divided into two sub-frameworks, the first one
organizing the kinds of general information present in a medical record based on the
three world’s theory (slightly modified from Andrade and
In this framework, everything begins at the level of cognitive representations
when a physician observes the reality at the patient side (arrow 1). Each of these entities
are filtered by cognition and represented by artifacts (arrow 2). Ontological entities
(entities O) are analyzed according to strict philosophical tenets, and are based on
reality itself rather than on physician´s mental representations. Examples of ontological
entities are cells, anatomical features and chemical substances. These entities are
directly considered in world 1 because in the realistic methodology adopted here
[
Other physicians can constantly interpret records and reality (arrows 4), resulting in new cognitive representations. Finally, the physicians involved in healthcare make judgments and process past and current information. Some of this processing of information (arrow 5) follows medical training rules, which determine the likelihood of a diagnosis and the correct interpretation of an exam result, to mention but a few. The representation of this reasoning process is also required for care continuation, which is a complementary part of the record (entity R). Examples of this include rules for interpreting lab data, as hemoglobin level < 12 g/dl means “low hemoglobin level”; and relevant negative information such as “lack of bowel alteration during episodes”.
When performing this sort of analysis, we distinguish ontological information from epistemological information, the latter represented as entities in Popper´s world 3, which is equivalent in Fig. 1 to the concretizations of cognitive representations level. Within this sub-framework we recognize at least four kinds of information to be separated according to their suitability for information systems: i) information that represents aspects of reality; ii) information that represents useful constructs for the medical practice that are not empirically verifiable; iii) information that represents observations about the reality, not the reality itself; iv) information that represents observations about the physician´s understanding of the clinical situation, not about the reality.
According to the aforementioned approach, only information that represents aspects of reality can be properly represented by universals. The other three sorts identified are epistemological information. It´s worth mentioning the link between belonging to one of the three worlds and the degree of truthlikeness. The information that pertains to worlds 2 and 3 is epistemological information and it will be classified according to a degree of truthlikeness. We don´t use the notion of truthlikeness to deal with ontological information pertaining to world 1.
It is clear that (ii) and (iii) are closely related to reality, with (ii) being a surrogate for a defined state of things on the side of the patient, and (iii) an objective account of its measures. Relations that allow for proper interpretation of those statements are particular to each domain. For instance, the examination of the color of the sclera may indicate jaundice (yellow color, surrogate for liver problems) or anemia (blue color, surrogate for iron deficiency anemia). The interpretation of what such signs mean depends on training, cultural practices and subjective characteristics. There are also specific relationships with regard to lab tests, since statements like “the total bilirubin level in the blood of patient X is high” requires knowledge of the method of sampling and analysis, knowledge of the probabilistic distribution of bilirubin concentration in the normal population, consideration of measurement errors and confusion factors and understanding of the meaning of measurement units. The last category (iv) requires more attention, since medical reasoning practices include both ontological relations and ad hoc heuristic rules that are not guaranteed to hold true in the world. We consider that the information in (iv) will eventually be registered in medical records as part of a learning process.
Our proposal also includes a way of characterizing epistemological information
based on its likeness to the truth. Following semantic approaches distinct from Popper´s
account, such as
In this sense, a simple propositional framework with three primitives (h, r, w) and the correspondent logical spaces are depicted in Fig. 2 as an example. The sentences from the associated propositional language are taken to express propositions within these logical spaces. This framework can be useful for characterizing information and scientific findings around a virus that has been studied only in the last few years, such as HTLV.
Here we conduct a preliminary test of the framework by analyzing individual
information entities contained in medical records. As an example, we present a small
extract of the clinical case available at
After we obtain a sentential fragment from an evaluation by a domain expert, we then isolate what could be represented in realism-based ontologies following the rationale of the BFO, OGMS and IAO. After that, we arrange the information according to the sub-frameworks mentioned in section 4.1. The final results systematize the information contained in a medical record based on either their ontological or epistemological nature. To the second kind, that of an epistemological nature, we add a classification based on the level of truth.
“A 62-year-old woman presented to the urgent care clinic with gingival bleeding after periodontal scaling of her lower-right second molar. She had undergone the procedure 5 hours before presentation, and the bleeding has persisted despite the application of pressure and ice. [...] The patient recalled a similar episode that had occurred 6 months earlier, also after a periodontal procedure, in which bleeding had stopped only after firm pressure had been applied and held for 6 hours. [...] She was otherwise in her usual state of good health. She reported no easy bruising, epistaxis, rectal bleeding, hematuria, weakness, fatigue, arthralgia, dyspnea, jaundice, abdominal pain, back pain, rash, confusion...” [Connors and Britton 2009]
In the Fig. 3, hereafter, we present samples of data obtained from the medical record in Fig. 4 and classified it according to the kinds proposed in section 3. Data representing aspects of the
reality Physician Woman 62 years-old
Data that represent useful constructs for the medical practice State of good health Former smoker
Data that represents observations about the
reality Heart rate: 80 bpm Blood pressure: 128/76 mmHg Patient report
No prior episodes of
White-cell count = 6,200
Data that represents observations about the physicians understanding Patient class: "Emergency patient" Bleeding had persisted despite the application of pressure and ice
Bleeding had stopped only Time of bleeding Time between episodes Aspirin Aspirin taken daily (rule) Thiazide diuretic Physical exam finding of that encounter unpredictable bleeding No allergies … … … … …
Lymphocytes = 37 Platelet-count = 352,000 Creatinine = 1.4 Albumin = 3.9 Prothrombin time = 13 sec … after firm pressure had been applied and held for 6 hours … The timing of bleeding after vascular trauma is different The patients presentation suggests platelet disorder … Patient class: "Emergency patient" Bleeding had persisted despite the application of pressure and ice
This data classification was based on both the levels of representation provided in section 4.1. From the empirical assessment by physicians, the categories suggested in figure 3 were created. The relation between the proposed framework and the organization of data from medical records can be summarized as follows: a) b) c) d)
“Data representing aspects of reality” (column 1) were mapped from processes (1) and (2) to entities (O) (Fig. 1) - only this information that can be directly used to populate realist ontologies, since terms in ontologies refer to universals; “Data that represent useful constructs for the medical practice” (column 2) were mapped from the process (1) and (2) to entities (A) (Fig. 1);
“Data that represents observations about the reality” (column 3) were mapped from process (3) to entities (I) (Fig. 1);
“Data that represents observations about the physicians understanding” (column 4) were from processes (4) mapped to entities (R) (Fig. 1).
Already the information classified in (b), (c) and (d) can to be use to support the building sets of sentence. For both, we define a set of true sentences about a blood disease following the orientation of experts. In context of the existence of the HTLV virus in a patient, a set of related sentences would be: i) HTLV cause neoplastic manifestation on human being infected by it, which we call proposition n; ii) HTLV cause inflammatory manifestation on human being infected by it, which we call proposition f; iii) HTLV cause infectious manifestation on human being infected by it, which we call proposition i. We can then consider that, in context of HTLV prevention and treatment, in a patient infected with the virus that presents both a neoplastic, an inflammatory and an infectious manifestation, those manifestations were cause by the HTLV virus. This complex situation is considered a true equivalent to the actual world we name w1. Table 1 depicts combinations of propositions ranging from w2 to w8, according to the relative closeness to the truth. ~n ~n ~n ~n ~n ~f f f ~f ~f
The truthlikeness gives us an objective criteria to evaluate the consequences of inclusion of such rules of thumb (weakness in HTLV infection is a neurologic complication) will behave in ontologies. Using this general rationale we can create “n” systems of spheres representing the situation considered real and other situations standing a logical distance from the actual world that is the truth (Fig. 4).
In this paper we presented a framework that aims to clarify the distinctions between reality, medical understanding and the recording of it, while maintaining the medical record as the main information source. Besides, we propose a way of dealing with epistemological information based on the notion of truthlikeness.
It is now well established that ontologies are an important resource to explicitly
define the meaning of terms, especially when coupled with advances in description
logics. Description logic is a powerful logic for describing the world, but is susceptible
to inconsistencies, particularly when dealing with instance data. We advocate that realist
ontologies provide a robust way of representing entities in reality, ensuring
interoperability and safe inferences. This is possible because epistemological
information, which can cause inconsistencies in inference processes, is not used in the
ontology. Interoperabiliy is favored by the use of the top-level ontology which is the
basis of the methodology adopted in this paper [
However, as we have shown, many entities in medical records do not have a
referent in the world, being representations of epistemological evaluations by physicians
and patient or measurements about real world entities. Ontologies, as pointed by Schulz
in
Popper’s theories provide a useful perspective on the different levels of reality and the relationship between theories (the ontology artifact being one of these theories) and reality per se. Our treatment of epistemological information seems to be an alternative to dealing with uncertainty common in the medical practice, from a logical point of view. Popper´s initiative in this regard, while essentially syntactic, entails the idea that no false theory is closer to the truth than any other. Other authors [Hilpinen 1976] [Volpe 1995] follow a semantic-oriented approach in looking for a plausible theory of distance between the semantic content of sentences. We believe that this latter approach fits well to the needs of a still-evolving subject that has to be captured in ontologies.
The proposed method has been tested in sentences obtained from real medical
records, but the partial results have suggested the need for refinement. The test
presented in this paper deals with a very small number of sentences and the feasibility
of the approach has to be tested in more complex situations. The possibility of dealing
with more complex cases is presented, for example, in
In future works, we intend to create clear rules for dividing kinds of information in a semi-automatic fashion. It will then be possible to test our approach against a greater sample. In doing so, we aim to explore the best characteristics of different systems and which representations suitable for each sort of system.