The area of eHealth is constantly evolving in the last years. Health service providers generate and record large volumes of clinical data in electronic or paper format. Most of the time such data is transformed or duplicated in clinical documents manually. In other cases, clinical data is inaccessible due to incompatibility of data types and structures. Therefore, it is a challenge to transform existing models of clinical documents into internationally accepted standards that enable semantic interoperability among the participants in the clinical process. The objective of this paper is to present a model for transformation of such legacy models into a standard reference model that supports semantic interoperability in accordance with the EU approved standard CEN 13606. A realistic case study of information services in Bulgarian healthcare is used to build an UML model and design a corresponding CEN 13606 Archetype model. The results are obtained as part of an eHealth scientific program with the combined research efforts of a team of researchers having professional experience both in medicine and computer science. It allows building a prototype of an information system that enables semantic interoperability among clients in terms of RESTful web services provided by a NoSQL database.
Nowadays patients widely use modern information technology to consume
services offered by the participants in the healthcare system– general practitioners
(GP), labs, hospitals, etc. These services generate and record large volumes of
clinical data in electronic or paper format. For example, in accordance with the
best worldwide practices Bulgarian GPs record in electronic format data for more
than 25, 000, 000 medical exams annually [
The main objective of this paper is to investigate the most frequently used
definitions of clinical data structures and to propose a model for adapting them to
CEN 13606 [
The paper is structured in seven sections. In the following section, we briefly describe the limitations in the information flow of XML documents employed to manage healthcare activities in Bulgaria. This realistic case study makes use of original information sources and provides evidence that the information flow is based on document definitions inconsistent with the CEN 13606 standard for semantic interoperability. It serves as a motivation to formulate in section 3 the problem statement in this paper. Section 4 establishes a common ground for harmonizing the existing document definitions in the national healthcare with the standard reference model of CEN 13606. Section 5 presents the proposed transition from W3C XML schemas used in legacy document definitions into the CEN 13606 reference model. Discussion about the applicability of the obtained results and concluding remarks are provided in sections 6 and 7. 2
Management of health insurance funds and payments to healthcare services providers (HSP) involve gathering and processing a huge amount of information. These tasks are delegated to a specialized institution in Bulgaria – the National Health Insurance Fund (NHIF). NHIF collects account reports in electronic format from HSP like hospitals, GPs, dentists, and medical laboratories for the services they have provided to self-insured patients. The reports are submitted as XML documents prepared in accordance with XML Schema definitions (XSD) designed and officially published by NHIF.
For example, let us consider the structure of such an XSD document employed
to generate the account report in the GP practice [
It is noteworthy that these simple types are direct references to standard W3C XML Schema datatypes such as string, date, boolean or named containers for a value of such datatypes like CodeType10 and Str150. The named containers include validation rules for the contained value in terms of regular expressions (see Fig. 3). These validation rules can be applied just to the value in the named container. For instance, it is impossible to create a validation rule in the XSD to ensure that DateFrom occurs before DateTo.
Moreover, the GP report as an instance of this XSD contains no trace of
the semantic context accompanying the generation of the entries in the AmbList
of this to report to the NHIF. It makes it difficult to trace the results of the
prescribed treatment or even to relate together different sets of medical checks. For
example, it is important to link the prescribed drug to a problem code, illness
history or record the drug code with personalized dosage instructions and codes
of risk factors [
In these circumstances, a patient may face the following problems. Assume the patient has made some expensive and painful medical examinations ordered by his GP. Then, while on vacation or on a duty trip, the patient loses accidentally his Ambulatory paper book and he needs to visit another GP. The newly visited GP has no access to the AmbList records of the patient in the NHIF. In the general case, this GP cannot exchange information with the primary GP of the patient about the reasons he has prescribed the medical examinations or about certain discrepancies in the medical data provided by the patient. Most often in such case, the patient must repeat the medical examinations.
This model of communication in the healthcare system entails duplication of
data and usage of different technologies for data management by each one of the
HSPs. The XSD provided by the NHIF appear to be the only standard for
healthcare data interchange in the national healthcare system. Finally, the absence of
semantic context in the collection of medical data does not allow the application
of modern means for computer-aided support for decision making at all the levels
of the healthcare system [
This case study exposes several information problems in the national healthcare system that can discovered in many other countries as well. First, we note that the exchange of medical information is centralized in a single institution (NHIF) and the information flows unidirectionally to this institution. The scope and content of the electronic records are reduced to serve merely the purposes of an activity-based funding business model for health insurance management.
The W3C XML Schema definitions of these records do not satisfy any
internationally recognized standard for semantic interoperability of health records like
CEN 13606 [
The absence of direct connectivity between HSPs isolates the large volumes
of data they accumulate in executing their everyday activities in the healthcare
domain. Modern eHealth approaches can help in finding reusable, scalable and
knowledge- driven solution based on the implementation of standards and best
practices in healthcare [
Therefore, in this paper we focus our efforts on the content of the XML documents submitted to NHIF on a regular basis by GPs in Bulgaria. Our main research hypothesis is that it is realistic to adapt these documents into the reference model of CEN 13606. In this paper, we prove that this task can be resolved in a typical case study.
Existing research papers provide evidence that the reference model of this
standard can represent all the medical information contained within an electronic
health record such as an Ambulatory list [
Interoperability refers to the ability two or more communicating parties to exchange information and use the exchanged information. There exist three levels of interoperability, syntactic, functional, and semantic. The levels of interoperability distinguish the degree a machine language can understanding the “meaning” of the exchanged information. Semantic interoperability provides a formal model allowing the receiving system to interpret the shared information by processing of formally defined concepts with a machine language.
The exchange of electronic health records (EHR) is an essential requirement for delivering eHealth services and products. In the general case EHR include patient demographics, his health status data, prescribed therapy, immunizations, laboratory data and other documentation required by the healthcare system. These records contain clinical data such as the GP records discussed in Section 2. Semantic interoperability in exchanging EHR plays an important role in healthcare as an instrument for the implementation of key functional and non-functional requirements that influence the effective management of the healthcare system and the quality of healthcare services.
The CEN 13606 standard is a five-part International standard approved by
the EU for semantic interoperability of EHR. In this paper, we employ the first
part of the standard, namely CEN13606-1:2019, where the EHR reference model
is specified [
The reference model of CEN 13606 follows the hierarchical structure typical for the representation and storage organization of clinical data. Class EXTRACT is the root of a hierarchical structure of an EHR or a part of an EHR. This class identifies the owner of the information extracted from the EHR in terms of demographic data, access policy and version control. For shortness, we will not consider here these properties. Class EXTRACT references a hierarchically structured clinical data defined by classes all of which inherit class RECORD_COMPONENT. It ensures secure exchange of extract components in this hierarchy even in case of duplication of information. Class FOLDER is the container at the top level of the clinical data hierarchy. Its purpose is to divide the clinical data into compartments related to a single condition like individual ambulatory lists of patients as our case study. Class COMPOSITION is model of the information produced by a single data provider. This class describes the data provider (person responsible for executing the healthcare activity or healthcare service structure). Details about clinical action, observation, interpretation, or intention are hierarchically grouped in sections (class SECTION) and entries (class ENTRY), where each component of the hierarchy is referred to by the abstract class CONTENT. Typical examples for instances of ENTRY are “Diagnose”, “Prescribed drug”, while “Subjective symptoms”, “Documents” and “Procedure” are examples for instances of class SECTION. At the lowest level instances of class ENTRY by a hierarchy of clusters (class CLUSTER) and elements (class ELEMENT), where components of this hierarchy are referred to as items (class ITEM). Class ELEMENT is the leaf node in the CEN13606 reference model, and its instances contain a single data value (“Height”, “Body weight”, “Drug name” etc.).
CEN 13606 enables semantic interoperability by separating the
representation of information and knowledge. At the first level of the proposed
methodology, information is structured in terms of the core classes of the reference model
introduced in the previous section. Knowledge is described at an upper level in
terms of a language for defining archetypes specified in the second part of this
standard known as CEN 13606-2:2019 [
The first stage in transforming the information into the reference model of
CEN 13606 is to convert the XSD document into a set of classes. Nowadays there
exist many software tools to complete this task. For example, Java Architecture
for XML Binding (JAXB) is one such software framework that allows mapping
Java classes to XSD documents [
The challenge is to fit the obtained classes at appropriate places into the
framework of the reference model of CEN 13606. First, the transformation of
information must preserve the semantics of the core classes and their relationships
in the reference model. Second, the classes from the here considered healthcare
domain must be mapped to classes in the reference model satisfying best
practices for information management in healthcare [
Let us consider the clinical data contained in an Ambulatory list (AmbList) displayed in Fig. 1. This figure shows the major entities that buildup the information content of an Ambulatory list. Among these entities, there are entities that are of administrative origin and others carry important clinical context. Entities with administrative content such as Practice (identification properties of the GP), Patient (identification properties of the patient), Doctor (identification properties of the GP) and the set of datamembers of the AmbList (the id code of the AmbList, date of issue, payment source etc.) map to the categories of FOLDER and COMPOSITION classes of entities.
It is common entities with semantic context related to healthcare procedures and services to serve as roots of hierarchical structures of documents with specialized content of information. Such documents like Docs (documents for specialized healthcare activities), SIMPList (directions for specialized non-hospital medical care), Procedure (highly specialized healthcare services, laboratory probes) or Profilact (prophylactically administered activities) from Fig. 1 may contain observations for physical, neuro-psychological and general development, temporary labor inability and so on. Therefore, it is appropriate to map these entities to the semantic context of the SECTION class in the reference model. On the other side, there are also entities that carry summary of clinical experience, such as MainDiag and Diag (the Main Diagnose and secondary Diagnosis coded using ICD-10) from Fig. 1. It is common to discover this kind of entities as instances of ENTRY classes of SECTION type in the CEN 13606 reference model. These objects provide content support for the previously considered group of SECTION entities.
This way we have demonstrated our approach for transforming an existing clinical document into the CEN 13606 standard reference model. According to this approach: • entities of administrative origin having the same values in each document instance are mapped to FOLDER and COMPOSITION classes of entities; • entities of clinical semantic origin that require more detailed classification of content with different values for each instance are mapped to SECTION types; and • entities that appear with different values for each instance of a SECTION type are mapped to ENTRY classes.
This transformation process obligatory requires practical experience in healthcare information management and consultation with a clinical professional.
The final step but not the least important in the transformation process is the mapping of ISO 21090:2011 compatible datatypes to the datatypes used in the XSD representation of the existing clinical document. In the here considered use case we discover user-defined datatypes like CodeType10 and Str150 (see Fig. 2) as well as simple types in XML such as string, date or boolean. ISO 21090:2011 supports directly simple XML types like string, date or boolean. Fig. 4 shows that these types can be wrapped in classes that are more complex, respectively ST, TS, and BL, if desired. The same refers to value containers such as CodeType10 and Str150.
The thus discovered information entities in XSD must be structured in a
reference model that is compatible with the CEN13606 standard. Therefore, the
second stage of the proposed information transformation involves extending core
classes in the reference model with classes discovered in the XSD document.
This way we to preserve the standard content in the core classes and the
relationships among them as per the standard specifications. It ensures that the obtained
extended reference model is compatible in terms of object-oriented design with
the base implementation of that model. A similar approach is followed in [
The extended reference model of a GP report is displayed in Fig. 5, where the shaded blocks represent the objects inheriting some core class in the base definition of CEN 13606. The state of the objects is determined by the values of the data members of the respective class. For simplicity of presentation core classes CLUSTER and ELEMENT as well as some, the discovered information entities in the XSD are not shown in Fig. 5. The proposed inheritance relationships allow these objects to be referenced in any software implementation that makes use of the core classes in this standard. The greatest challenge in developing the extended reference model is the proper selection of the base class among the standard core classes for each one of the classes obtained from the XSD model. ISO 21090:2011 compatible datatypes. 6
The Reference Model shown in Fig. 5 satisfies the CEN 13606; however, it is
not ready for practical implementations. One of its drawbacks is that it does not
reflect well the business process followed by GPs to collect clinical information,
evaluate, and prescribe a therapy to a patient in need. Another drawback is that
this model is difficult to share and implement readily in software applications.
For this purpose, we convert the Information model shown in Fig. 5 into the
Archetype model of CEN 13606 [
Unlike the Reference model shown in Fig. 4 the Archetype model transforms the data entities in the XML Schema definition of a GP report (see Fig. 1) into a set of four major SECTIONs that are intuitively understood by a GP. These SECTIONs represent the four major stages of a visit to a GP (Observation, Evaluation, Instruction and Action).
Initially a GP performs an OBSERVATION of the vital signs of the patient and records the measured values as well as gets familiar with the history of the illness or other clinical evidence. Next, the GP makes an EVALUATION of the clinical information in the semantic context of the overall health status of the patient. This allows the GP to assign a therapy for the illness in terms of a set of INSTRUCTIONs and propose a set of ACTIONs the patient must execute.
It is noteworthy that the proposed grouping of concepts based on their
semantic context in the business process a GP follows in executing his activities is
not clearly defined in the CEN 13606 standard. Similarly, to the openEHR open
source specification [
This archetype model itself is expressed in an Archetype Description Language (ADL) that is based on XML. This model is inherently extensible and scalable with respect to the semantic context we want to include in clinical data exchange. It also allows representing any clinical document in an EU standardbased model that can be shared among application developers as well as used by GPs to exchange clinical data.
The Archetype model of CEN 13606 provides several features that are required for achieving semantic interoperability. For example, it allows imposing constraints on the range of attributes of primitive types, constraints on the cardinalities of these attributes as well as introducing constraints on complex objects. The ontology section (see Fig. 6) of that model allows describing and binding information entities to clinical terminologies like SNOMED CT. For example, the entries of MainDiag and Diag in AmbList are bound to the ICD-10 terminology. It helps to embed semantic context in easy to discover patterns related to Procedures ENTRY concepts of health-related activities (SECTION-PROCEDURE) or to Observable ENTRY concepts (SECTION-ACTION), where specific values identify a health-related finding of the GP.
The Archetype model of the clinical documents has been exported in a W3C
XML schema. It allowed managing instances of this model in a uniform way on
a noSQL database like eXist-db [
This research is supported by the National Scientific Program еHealth in Bulgaria.