Virtual Medical Record (vMR)
- Kensaku Kawamoto - CDS Co-Chair
vMR Project calls
Thursdays (every week) from 4pm to 5pm US Eastern Time ET
Announcements are sent over the CDS list.
Live Meeting Dial-in: 770-657-9270, Passcode 687054
Strawman vMR Model
- see "Other Working/Reference Documents" section below for proposed refactoring of strawman model
- XMI file can be uploaded into any UML editor
- .eap file can be opened by free Enterprise Architect viewer, available at http://www.sparxsystems.com.au/bin/EALite.exe, which is within http://www.sparxsystems.com.au/products/ea/downloads.html
Strawman vMR Model Change Log
- (please note changes to model as well as rationale, along with date and author information)
|Date||Changes to model||Rationale||Author information|
|2009-11-12||Initial post of strawman model based existing Duke vMR.||Ken Kawamoto and Guilherme Del Fiol, Duke University|
|2009-11-12||Added Family History and Allergy classes. Added concept of population-based query.||Modifications based on group discussion.||vMR working group.|
Other Working/Reference Documents
ISO 21090 Datatypes References
These links reference the HL7 SVN server. If required, supply a username of "anonymous" and a blank password.
HL7 V3 Datatypes SVN
The HL7 V3 Datatypes SVN is browsable from here.
Snip from Abstract datatypes standard as to why we are using ISO datatypes
The data types defined in this specification are not intended to be implemented solely based on the details presented herein. For example, this specification does not differentiate between the information that should be represented explicitly in any technical representation such as XML and the information that should be derived from the explicit representation. This document corresponds broadly to the RM-ODP informational view ("concerned with the kinds of information handled by the system and constraints on the use and interpretation of that information"). The ISO data types specification is an ITS for these data types that corresponds to the RM-ODP computational view ("concerned with the functional decomposition of the system into a set of objects"). See  for further information.
What should be the primary objectives of this effort for the short term (next 3-6 months)?
Tentative: Generation of community consensus on
- what types of patient data may be used as inputs for CDS;
- what types of inferences may be returned as outputs by a CDS engine; and
- the simplest possible information models for these data
What should be our approach to fulfilling the objectives in a timely manner?
- Create straw-man UML diagrams for vMR - Discuss/modify UML diagrams as initial approach to arriving at common vMR consensus - Explore ballotable formats Potential initial ballot: UML Domain Analysis Model (DAM) Potential further ballot: HL7 v3 model based on DAM - Need to be aware of U.S. meaningful use criteria (e.g., for use of SNOMED for problem lists), other international requirements
What should be our decision making process?
- Use standard HL7 processes (attempt consensus; if needed, majority vote to decide on most issues at this stage in the process)
Other current work of note to this project
- NQF's Quality Data Set (QDS)
- IHE - PCC Content Modules
What is a vMR?
The vMR is based on the HL7 V3 model and datatypes. It is a "virtual" interface and is optimised for point in time clinical decision support. While it could be used as the basis for an EHR design it is intended to be simplified, to represent a snapshot and to omit important concepts that should exist in a real EHR.
In particular concepts such as audit trail, result history and display forms are omitted and only data relevant to making decisions based on the current patient state are represented. The hard bit is excluding as much complexity as possible while ensuring that any decisions made are safe and based on adequate data.
It is intended that each EHR system would create an "Adaptor" or "Facade" to present patient data in a consistent form so that the clinical decision logic can be common across many systems. Obviously some systems will be missing data and it is likely that the data will not be in a HL7 V3 form and this is where the word virtual comes in, there needs to be a translation layer to transform data into a form that complies with the VMR. There are many potential ways to do this ranging from creating a CDA document, a HL7 V3 message, a SOAP service or even creating HL7 V2 message(s) that contain the required structured data to allow the creation of a vMR interface. A direct database access layer is also possible. So documents, messages, services, archetypes/templates, CCD instances and database queries for example could all populate the vMR.
The aim is to allow a CDS language access to a standard model to reliably access patient data about the current patient's "Observations" (eg haemoglobin) or "Family History" or eg "Surgical History"/"Problem List". This is the single patient "Context" but other contexts are relevant. Access to population based data is very important for public health monitoring (perhaps context "Population"?) and in the case of templates access to data during the editing process (context "Template").
The project is currently under active development and these ideas are not fixed or standardized but give some idea as to what a vMR should be. An attempt to try and conceptualize the requirements using the HL7 "SAEAF" famework - the "Services Aware Enterprise Application Framework" is below. The SAEAF framework, when used in an agile manner shows great promise in structuring thoughts on the topic and allowing alignment when people come at the problems from different world views.
|Enterprise View||Information View||Computational View||Engineering View||Technology View|
|Conceptual||HL7/ISO Datatypes Classes in the vMR Model||vMR Examples|
|Platform Independent||vMR UML Model||ISO 21090 datatypes||GELLO Class definition|
|Platform Specific||LRA Models||CCD RMIM View|
More on the 'virtual' in vMR
Noone has an exact implementation of the vMR. It will always need to be mapped to local structures. It needs to handle missing data items, for example if the local structure doesn't have a smoking history for patient, the vMR needs to know it doesn't have it - it is null. Another example would be if the Allergies section of a populated instance of a patient's electronic medical record is not filled in - is this 'No - there are no allergies' or is this a null value? A test for the vMR is that is needs to be addressable by CDS languages such as GELLO. (The vMR is for clinical decision support). It exists for CDS, then it is gone.
Distinction from reference models
As an artefact of the HL7 standards community it might seem that this notion of a vMR is confused with that of a specialization of the RIM such as a RMIM. Remember it is a simpler, snapshot view and it's a slightly different view at that. So instead of top level Entities and Acts, we have explicit links, such as Patient (Entity) has Observation (Act). The vMR has a context which would often be a patient context, but it could be a population one. This context builds a specific patient class property rather than following the RIM links (ie Person playing the role patient..). So the link is clearer (perhaps) and stuff at the RMIM level can be exposed as top level constructs.
Additional classes can be included such as ProblemList, FamilyHistory and Procedures.
Potential vMR sources
So what's to be in and what's to be out? Ideally we'd be in the Goldilocks zone with a one pager that keeps most people happy. We should borrow the SNOMED-CT editorial policy of 'URU' - understandable, reproducible and useful. Here is a list of some potential vMR class sources:
- SAGE vMR model
- HL7 Care Record Model
- Logical Record Architecture (NHS)
- Different participating institutions’ vMR models
- Comment-only HL7 vMR ballot
- RMIMs such as Clinical Genomics