The goal of Health Level 7 is to develop standards that permit structured health care information to be exchanged between computer applications in such a way that meaning is preserved. Recent analyses of the HL7 RIM, which forms the heart of the new Version 3 of HL7, have given rise to concerns that this objective can not be achieved by means of the paradigm currently proposed. In this panel, the floor is given to experts who, on the one hand are involved in the development and implementation efforts of the HL7 RIM, and, on the other hand, have published serious critiques both concerning problems with the RIM itself, and also with its documentation. The panel will address the question as to what extent these critiques have a sound basis and thus point to serious defects in the HL7 V3 project, or whether they are, rather, based on misinterpretations and misunderstandings. The panel will seek to conclude with an agreement as to what actions should be taken in the immediate future in order to remediate mutually acknowledged problems.
The Health Level 7 (HL7) Reference Information Model (RIM) was introduced as an object oriented information model to harmonize the definition of HL7 messages across different application domains. On the heels of HL7’s successful version 2, the new version 3, including the RIM, which forms its centerpiece, has received significant attention, but it has in turn also been subjected to criticism, which has addressed the following questions:
- Questions about implementation: The decision by HL7 to adopt the new RIM-based methodology was adopted in 1996; have the promised benefits of interoperability yet been realized?
- Questions about the usability in specialist domains: The RIM defines ‘normative’ classes such as Act, Role, Entity, etc. each of which is associated with a rich stock of attributes. When the RIM is applied to each new domain (for example pharmacy, clinical genomics), one then needs to select from these attributes. Does the RIM yield a coherent basis for constructing well-designed software artifacts for functions like the EHR or computerized decision support?
- Questions about scope: The RIM has a small number of ‘foundation classes’, under which everything formulated in its terms must somehow fall. Of these, ‘Acts’ (roughly: actions which need to be described and communicated, healthcare ‘transactions’) play a central role, so that only the class Entity (defined as: ‘A physical thing, group of physical things or an organization capable of participating in Acts, while in a role’) is left over for those things which are not Acts. How, on this basis can the RIM deal transparently with information about, say, disease processes, drug interactions, wounds, accidents, bodily organs, and many other phenomena central to healthcare?
- Questions about the documentation: The he RIM is designed to serve as backbone of the V3 methodology; unfortunately, documentation of the RIM appears to contain numerous inconsistencies and/or incomplete definitions.. How, then, can the RIM be used as reference information model by the many disparate communities it is designed to serve?
- Questions of learnability: Can HL7 V3 be taught, and therefore engaged with and used by a wider public, given what many people feel are a substantial number of special cases, frequent amendments, and other complex hurdles that must be overcome in creating a message?
- Questions about marketing: Are the marketing claims made on behalf of V3 as ‘the data standard for biomedical informatics’ justifiable, given what seem to be many still unresolved technical problems confronting those engaged in implementation.
Persons working closely with the HL7 RIM and V3 methodology acknowledge the less-than-perfect quality of its documentation. However, they argue that the formal and technical criticisms recently advanced, including criticisms of the conceptual design of the RIM, are rooted in misunderstandings as to the purpose and scope of the RIM and V3, and more importantly in fundamental disagreements as to how a reference model for semantic and pragmatic interoperability should be created. This panel is designed to move the debate forward in the interests of both better understanding and of better tools to meet the healthcare information needs of the future.
PANEL STRUCTURE AND ORGANIZATION
The panel will take the form of four brief presentations, followed by a debate on the Future of HL7 chaired by Professor Don Detmer and involving all panelists. The presentations will cover the following topics:
Barry Smith: The Ontological Analysis of HL7’s Reference Information Model
The HL7 Version 3 is an ambitious attempt to address certain problems created through the very successes of its predecessor Version 2: it is designed to facilitate consistent sharing and usage of data across multiple local contexts. The vision of HL7 V3 is to minimize the possibility for local dialects to be formed by providing a normative repertoire of ‘classes’ in terms of which all healthcare messages would be defined. But what are the ‘classes’ of which the HL7 ‘information model’ is comprised? In some places we are told that they are information objects; in other places we are told that they are the entities in reality to which such information objects refer. (So that persons, for example, could be ‘created in error’ and therefore ‘nullified’ by being deleted from a database.) We shall argue that this apparent confusion is not merely a product of poor documentation, but that there is in fact no stable interpretation of ‘class’ in the RIM which will enable it to meet the goals it is designed to achieve.
Charlie Mead: The Role of the RIM and V3 in Computable Semantic Interoperability
The HL7 Reference Information Model is a cornerstone of HL7’s commitment to facilitate computable semantic interoperability in the healthcare and clinical research domains. In particular, the RIM – in concert with the overarching V3 methodology – defines a collection of concepts, datatypes, attributes, and relationships which collectively specify the cross-domain semantics of a set of ‘common structures.’ In order to express a particular semantic utterance in a particular domain, RIM structures must be bound to domain-specific terminologies and contexts. The resulting RIM-derived data structures form the basis for both V3 ‘messages’ (including the Clinical Data Architecture specification) and application transactions (e.g. Oracle’s Healthcare Transaction Base uses the RIM to build its persistence layer.) The RIM and V3 are ‘necessary but not sufficient’ pieces in the overall solution of attaining computable semantic interoperability (CSI). Dr. Mead will discuss the application and scope of the RIM in the larger context of CSI.
Werner Ceusters: Referent Tracking in HL7
HL7 played an important pioneer role in recognizing the need for a system of unique patient IDs as an indispensable part of any Electronic Health Record system. The goal of referent tracking is to extend this idea beyond the patient, to create an ever-growing pool of data relating to other sorts of entities existing in concrete spatiotemporal reality and relevant to patient care. In the context of EHRs the relevant concrete entities are not only particular patients but also their parts, diseases, therapies, lesions, and so forth, insofar as these are salient to diagnosis and treatment. Within a referent tracking system, all such entities would be referred to directly and explicitly. In this presentation, we will present our analysis of the HL7 RIM in light of its capacity to deal with the requirements put forward under this new paradigm.
Christopher G. Chute: HL7 RIM in the Context of US and International Standards
The HL7 Reference Information Model represents the most successful globally coordinated effort to evolve a shared model about the health care process. It has demonstrated substantial change over time, and no doubt will continue to evolve. Robust debate on its merits and gaps must be welcomed, however this must be balanced with what has evolved and the profound transformation that has occurred within standards organization who now recognize more clearly the importance of high-level interoperability and semantics. The RIM has had focused thinking on these issues, and if properly engaged – will continue to do so.
Barry Smith (organizer)
Christopher G. Chute
Don E. Detmer (moderator)
Barry Smith is Julian Park Distinguished Professor of Philosophy in the University at Buffalo and Director of the Institute for Formal Ontology and Medical Information Science in Saarbrücken, Germany. His research focus is ontology and its applications in biomedicine and biomedical informatics, including electronic health records.
Werner Ceusters studied medicine, neuropsychiatry, informatics, and knowledge engineering, and directed a series of international research projects in medical natural language processing under the Third, Fourth and Fifth Research Frameworks of the European Commission. He is Professor in Health Sciences and Director of the Ontology Research Group, Center of Excellence in Bioinformatics and Life Sciences, University at Buffalo.
Charles Mead has been involved with HL7 for more than 10 years. He has served on the HL7 Board of Directors (2001-05) and is currently the Methodology and Modeling Facilitator for the Regulated Clinical Information Management (RCRIM) Technical Committee. He now works for Booze Allen Hamilton providing CTO-level support to the National Cancer Institute’s caBIG, CRIX, and CTWS projects.
Christopher G. Chute is Professor of Medical Informatics and Associate Professor of Epidemiology at the Mayo Medical School, Rochester, MN. He is Vice-chair of the ANSI Health Information Standards Board, chair-elect of the US delegation to ISO TC215 for Health Informatics, co-chair of the HL7 Terminology Committee and a past member of the NIH Medical Informatics Study Section.
Don E. Detmer is President and CEO of the American Medical Informatics Association, Bethesda, MD.