History of the HL7 RIM and HL7 version 3
The contents of this whitepaper are published under the Creative Commons Attribution-Share Alike license.
The history of the RIM goes back to the IEEE P1157 (MEDIX) organization (1986) who introduced the concept of a common healthcare data model. This paper covers the major steps in the development of the HL7 RIM and HL7 version 3 as a development framework.
1. IntroductionThe HL7 Reference Information Model (RIM) as we know it today has been created over a period of 25 years, and its early history actually lies outside of the HL7 organization. This paper aims to describe the history of the RIM, or rather the history of the concept of 'having a common data model for healthcare' which ultimately lead to the HL7 RIM. It also covers the HL7 Development Framework, which is the basis for the creation of HL7 version messages.
The content is based on research, interviews with George W. "Woody" Beeler, Abdul-Malik Shakir and Mead Walker, and on comments by readers. Any related work that is not related to healthcare has been declared out of scope for this paper, as is any work without a traceable link to the development of the HL7 RIM or its direct predecessors.
There are known gaps when it comes to the European/CEN history and its intertwined relationship with the origins of the RIM and the HL7 v3 methodology. We're open for suggestions as to how this paper can be improved.
2. IEEE P1157 MEDIX (1986)During the 1986 MEDINFO conference an international group, representing both end users as well as systems vendors, determined that the development of standards for the communication of medical data between heterogeneous healthcare information systems is a necessary condition for accelerating the diffusion of automated information system technologies in healthcare. [HAR90, RUT89]
In 1987 the IEEE board approved the creation of the IEEE P1157 Medical Data Interchange (MEDIX) committee. [HAR90] Meetings were scheduled four times a year, with at least one meeting held each year outside North America. Around 1990 MEDIX had North American as well as a European branches (the latter was chaired by Sigurd From, the former by John J. Harrington); participants mainly resided in North America (75%) and Europe [HAR90]. It was recognized that standards for healthcare data interchange must be international, and that they must support multiple network technologies, communications protocols, and data types. It was of interest to determine where there are regional differences in medical data interchange needs, and to allow sufficient flexibility in the standard to support them.
2.1 The MEDIX FrameworkThe framework adopted by MEDIX states that in order for applications to exchange data they have to have a 'shared universe of discourse'. Within MEDIX the description of the shared universe of discourse consisted of a conceptual schema, referred to as the MEDIX Information Model. This model was defined in terms of objects in the shared universe of discourse for healthcare interchange and the operations that may be carried out upon these objects. Initially the model was represented using ER diagrams, and based on a hospital data model provided by Tom Rutt (AT&T Bell Labs) and a comprehensive healthcare data model created by the UK National Health Service. Around 1990 MEDIX started to use an object-oriented information model.
The MEDIX Framework Model for Healthcare IT
Messages that are exchanged in a particular instance of communication are the result of a mapping of the relevant portions of the common information model onto a particular interchange format [HAR90]. Although there is a desire to minimize the number of interchange formats that must be supported in the healthcare sector, no one format has been identified which meets the user needs in terms of simplicity and expressiveness.
2.2 Standards CoordinationIn 1988 MEDIX observed that various groups were actively creating healthcare interoperability standards (e.g. HL7, founded in 1987). To prevent overlapping and conflicting standards representatives from these groups met in January 1988.
As a result of that meeting the Healthcare Information Standards Committee (HISCC) was created. Comprised of various standards committees including HL7, DICOM (a workgroup of ACR/NEMA), ASTM, MEDIX and others, HISCC provided a forum where representatives of various standards efforts can meet, exchange information and develop recommendations for promoting compatibility. ANSI formed the Health Informatics Standards Planning Panel (HISPP) in 1991, as a result of a request by European standardization agencies with whom they could coordinate health informatics standards. HISPP was replaced with an ANSI designated Health Informatics Standards Board (HISB) in 1995. [CDM96, BLO97]
2.3 ContributionThe really important bits of MEDIX were right at the very beginning (1986). It was chaired and established by Tom Rutt (AT&T Bell Labs), who was/is a real standards guru, having played a major role in the development of various OSI standards. The main contribution of MEDIX towards the eventual development of the HL7 RIM are probably that is has put the topic of a 'object-oriented reference model for healthcare' on the map, and its framework which relates messages to an information model. "Their objectives .. now read as the underpinnings of HL7 version 3." according to Woody Beeler.
The MEDIX work was also influential on the work of CEN TC251 WG3. Sigurd From and Tim Benson were two of the first people from Europe to regularly attend MEDIX meetings where they came across Coad/Yourdon Object Oriented Analysis methodology, which was used later in both CEN and HL7 until replaced by UML. Sigurd chaired the CEN project team CR1350:1993 which came up with the idea of syntax-independent models and led directly on to CEN CR 12587:1996 (Methodology for the development of healthcare messages). This approach was subsequently (around 1996) used as the basis for the HL7 v3 methodology.
Video interview (recorded in 2010) with George W. &qout;Woody&qout; Beeler, Abdul-Malik Shakir and Mead Walker
3. ANSI HISPP MSDS Joint Working Group for a Common Data Model (JWG-CDM) (1992)In 1992, HISPP together with other standards developing bodies formed the Message Standards Developers Subcommittee (MSDS). In this environment, IEEE has provided the secretariat for the Joint Working Group for Common Data Model (JWG-CDM), including ASTM, HL7, DICOM (a Working Group of the ACR/NEMA), NCPDP (National Council of Prescription Drug and Pharmacies), and X12N (Insurance Subcommittee of ANSI Standards Committee X12) [BLO97].
The JWG-CDM, headed by George W. �Woody� Beeler, began meeting with HL7 in 1992 up to about 1996 when participants effectively decided to work with HL7, it being the only organization at the time with a real interest in such a common data model. The JWG-CDM still was a very active group whereas the interest in the remainder of the MEDIX activities had waned.
3.1 Meta-model standardJWG-CDM was an open standards effort with the objective to support the development of a common data model (CDM) that can be shared by developers of healthcare informatics standards. "The idea was that by having the various SDO's create a common data model", to quote Abdul-Malik Shakir, "we can achieve harmony across these standards". In order to harmonize the data content of standards developed by independent SDOs the data content must be expressed in a manner that facilitates comparison. Data modeling was the method chosen for specification of the data content of health informatics standards for the purpose of harmonization across the SDOs.
The JWG-CDM "Data Model Framework" [CDM96] defines graphical and textual notations and recommends procedures for development of standardized components of an overall object-oriented information model to be used in healthcare data interchange. The challenge of the JWG-CDM was to find a method of modeling the data content of the healthcare domain without imposing model components that relate to the standard-specific tasks of defining implementation details such as data structures, communication trigger events, messages, message containers, etc. The art of the JWG-CDM's task was to provide a framework for the Common Data Model that can be readily extended by all of the standards bodies to meet their own standards specific requirements.
The JWG-CDM also had the aim to define a high-level object model of the healthcare domain, but this never came to fruition.
3.2 ContributionThe main contributions of the JWG-CDM towards the eventual development of the HL7 RIM are the meta-model (which forms the basis of the RIM meta model) and the process it developed to create one common data model by means of harmonization. The meetings of JWG-CDM were co-hosted with the HL7 Working Group Meetings (WGMs) - the process of harmonization and the modeling approaches used by the JQG-CDM were well familiar to the HL7 organization.
The JWG-CDM activities also attracted (across all existing standards organizations) all those persons interested in creating a common reference model - a group of persons that would not only play a major role not just in the development of the HL7 RIM and HL7 version 3, but would also play a major role in the makeup of the HL7 organization.
4. HL7 QADM (1996)The HL7 Quality Assurance/Data Modeling committee (QADM, headed by Mead Walker) was created around 1990, with the idea of creating models prior to the creation of message specifications. The CEN organization (the European standards body) also used this as part of their methodology which seemed to produce better quality specifications than HL7 v2. The various groups in HL7 maintained their own segments without checking for inconsistencies with segments managed by other HL7 groups. A common data model (much like the approach used by JWG-CDM) would be helpful to ensure consistency across all segment definitions. The development of such a common data model (based on existing segment definitions) started around 1994.
4.1 The initial RIMHL7 was already working on HL7 version 3.0 in 1990, which had convergence with MEDIX as one of the stated goals [HAR90]. The process by which messages are developed from the Information Model, as well as many characteristics of the methodology as a whole was drawn from the work done by CEN TC251 WG 3 (headed by Georges deMoor), notably on CEN CR 12587:1996 Methodology for the development of healthcare messages. [MDF99]. Once the outline of the HL7 v3 methodology was in place, and the RIM was envisioned to be at the heart of that methodology, the decision was made to develop the RIM in April 1996. [RIM04]
Data Model sources for the initial RIM [SIL02]. UK and Australian national healthcare models are missing from the table.
The initial RIM was based on a set of data models from HL7 members and other sources (e.g. MEDIX experimental models, vendor data models, provider data models, and data models created by other SDOs). The harmonization process as created by JWG-CDM was adopted and used to harmonize those models into a single construct.
Abdul-Malik Shakir created the initial model, which went through about 8 iterations before there was a model worthy of bringing to the HL7 TSC with the suggestion to use it as the common data model for all HL7 modeling efforts. In 1997 RIM 0.8 was published, followed by the first ever HL7 harmonization meeting. By 1999 the RIM had reached its grandest proportions, 130 classes, 987 attributes, and countless associations. The layout of the classes was such that the "left-hand-side" was oriented towards classes related to clinical use cases, and the "right-hand-side" to claims, invoices and other financial and administrative stuff.
The HL7 RIM (0.92), 1999. See high resolution image for details.
5. USAM (1999-2001)The Unified service Action Model (USAM) II proposal [USA00] came forward as a joint harmonization proposal by multiple committees to simplify the RIM, especially when it came to the large number of classes that dealt with clinical concepts. The main authors were Gunther Schadow, Dan Russler, Charlie Mead, Jim Case and Clem McDonald [Rus99]. The proposal aimed to generalize those classes. They have similar attributes and can be combined into one single class. The proposal introduced the Service (now: Act) and Material (now: Entity) classes as overarching concepts. The USAM II proposal was based on an earlier USAM proposal which was created in the spring of 1998.
The proposal was met with skepticism, until it became clear that the proposed RIM would be much more suitable to be used as the basis for the creation of refined models (e.g. R-MIMs).
5.2 Effect on the RIMThe proposal had the effect of collapsing a number of related classes into one class, partly through the introduction of the concept of mood attribute which allowed one single class to be used regardless of how it was used (e.g. to order an activity, to schedule an activity, or to perform an activity).
The resulting RIM abstracts what it can into terminology. The complexity of the previous RIM static model still exists, but it has been solved using terminology rather than by classes and attributes.
As a result of USAM proposal (to the left-hand-side of the RIM) and the application of USAM principles (elsewhere in the RIM) the number of classes dropped from 125 to less than 2 dozen. USAM also introduced the postulate that says that there can be only one path to get from class A to class B, in order to simplify the model and to avoid choices.
5.3 Follow-up workUSAM was initially only applied to the "left-hand-side" of the RIM. The "right-hand side" was initially left unchanged. Around 2001, Woody Beeler came to Canada for a week and the right-hand side was rebuilt to support the e-claims work that Canada was doing at the time.
In 2000, the modeling of Role was tweaked, introducing the idea of player, scoper and RoleLink. This was proposed during the "left-hand-side of the RIM project" one of a series of proposals addressing the remainder of the RIM after adopting USAM [RIM04]. The project team had Jane Curry, Charlie Mead, and Norman Daoust amongst its members. They were working on use cases for roles - and in particular Patient - when they realized that the role of an entity (player) was from the point of view of some other entity (scoper) so that one organization's patient was a different instance from another organization's patient, even if the playing entity was the same in both cases.
5.4 ContributionThe USAM proposal created the essential RIM structure as we know it today. The 1999-2001 timeframe was essential in the development of the HL7 RIM, from USAM up to the first non-draft RIM. The first non-draft RIM, version 1.0, was published in 2001. In 2003 RIM Release 1 was made normative, which became an ISO standard in the 2005-2007 timeframe. In 2010 RIM Release 2 became normative.
6. The HL7 Development Framework (HDF)The RIM and the finalized HL7 v3 methodology (the HDF) also attracted a large number of new volunteers to the HL7 organization to create refined models for use in message specifications.
Introductory tutorial for the HL7 RIM (Release 2)
The HL7 Development Framework [MDF99] details, amonst other things, a methodology to create RIM-derived HL7 version 3 message/document information models. This is essentially a two step process:
6.1 HL7 version 3 Tooling
HL7 v3 Tools: R-MIM Designer, RoseTree, v3 Generator
7.1 DeclineBy 2011 we had a substantial number of HL7 version 3 implementations. HL7 version 3 messages were used by nationwide data exchanges in (e.g.) the English NHS, the Netherlands, Finland, Estonia, and parts of Canada, and HL7 CDA documents were used by all sorts of (regional, national) exchange projects of electronic documents, mostly used in combination with IHE XDS. Apart from CDA and SPL (Structured Poruct Labeling, an FSA mandated standard) the adoption of HL7 v3 was very low in the us.
The uptake of HL7 v3 had been stalling for years, and software implementers were reaching the conclusion that there were significant issues around its implementation. For example, see Grahame’s blogposts about the concept of drive by interoperability, HL7 v3 has failed, or my blogpost How to lower the hurdle for HL7 v3 implementers.
HL7 as an organization, whilst committed to the further development HL7 version 3, was aware of these problems as well, and initiated the Fresh Look initiative, more or less a brainstorming exercise as to the future direction of its standards. This initiative ultimately lead to the creation of the FHIR specification, which was initially published in August 2011. Within a few years the HL7 community embraced FHIR, and the development and maintenance of HL7 v3 effectively dwindled to near zero. As of 2022, a few projects still use HL7 v3 messages in production (as a legacy standard), and a decreasing number of projects are still based on the exchange of CDA documents.
SummaryOn hindsight, the strong points .. Legacy of ..
8. AcknowledgementsThe author would like to thank George W. "Woody" Beeler, Abdul-Malik Shakir and Mead Walker for their willingness to be interviewed at length about their recollections. A special thanks to Alexander Henket for presenting an overview of the HL7 version 3 tools. A number of post-publication comments were received from Tim Benson, Lloyd McKenzie, W. Ed Hammond and others - those have mostly been incorporated into this whitepaper.
About Ringholm bvRingholm bv is a group of European experts in the field of messaging standards and systems integration in healthcare IT. We provide the industry's most advanced training courses and consulting on healthcare information exchange standards.