%% Text written by Zak Kohane, modified by Jon Doyle, for AMRMC BAA


The implementation and deployment of Electronic Medical Record Systems
(EMRS) has lagged significantly behind available technological
solutions.  One important reason has been the lack of standardization
of both the structure and content vocabulary of EMRS.  We propose to
address the lack of standardization by employing public-domain
technologies to implement viewer/editors of multimedia compound
documents and their transmission over the Internet without
compromising confidentiality.  To avoid unnecessary duplication of
effort we will use open-architectured commercial tools, particularly
to provide interfaces to heterogeneous data-bases.  The overall
objectives entail the following specific aims:

1.  Develop a multi-media interface to the medical record that supports
customization for specific clinical environments and needs.  The
HyperText Markup Language (HTML) will provide the foundation for this
work.

2.  Establish a canonical electronic medical record structure with
supporting data abstraction processes to provide consistent views of
medical information independent of underlying database
structures.  Emerging standards of medical record structures will be
applied, and commercial products for database processing will be
utilized to build the abstraction procedures.

3.  Employ standardized medical vocabularies in the EMRS, and provide
processes for translating/ mapping of local or non-standardized
medical vocabularies.  Using UMLS as a foundation, various methods of
mapping local vocabularies to standardized medical vocabularies will
be explored.

4.  Provide communications support to allow access to the EMRS from
multiple sites.  Since support of several de facto Internet
communications standards is integral to HTML, we plan to investigate
the use of the Internet as the backbone communications structure for
the EMRS.  The application of encryption/security measures will be
investigated.

Emerging Standards

Improvements in the management of health information anticipated with
the growing use of computer information systems rely on the pervasive
adoption of standard representations of patient specific clinical data
and general medical knowledge.  The standardization of computerized
medical record systems has been the primary focus of the Message
Standard Developers Subcommittee (MSDS) of the Health Informatics
Planning Panel (HISPP), whose efforts are coordinated at the national
level by the American National Standards Institute (ANSI).
International approaches to standards have been directed by the
International Standards Organization (ISO), of which ANSI is a member.
Broad requirements for standardization in medical information
technology have been summarized in a white paper presented by the
American Medical Informatics Association [4] .  In addition, the final
report of the ANSI HISPP MSDS: Common Data Types Subcommittee is
awaited.

Common standards, which one leader in the field describes as Òthe
lubricants of technical and economic progress,Ó have arisen slowly
from the competing challenges of promoting uniformity and universality
of the whole while maintaining the diversity and granularity of the
components.  Accomplishments and work in progress in this area can be
divided into standardization in data interchange or syntax, and
standardization of terminology or concepts.  Since the early efforts
were aimed to achieve the business goals (provider - payor
communication), progress in and implementation of data interchange
standards has surpassed the development and use of standardized
medical vocabularies in the documentation (provider to provider
communication) of medical information.

State of the Art EMRS

Comprehensive electronic medical record systems have been discussed
for decades [5] , but comprehensively implemented in relatively few
sites (e.g. Regenstrief [6] . HELP, [7] , TMR [8] , Veterans Affairs
[9] , Columbia- Presbyterian [10] ). Where they have been broadly
implemented, there is some evidence to suggest that these systems
contribute to improved clinical care and reduced costs [11-13]
. Subsequently there have been multiple efforts to develop
workstation-based, graphics-intensive on-line patient charts (e.g. PWS
at Hewlett Packard/Stanford, [14] DeSyGNER [15] , CWS at Children's
Hospital [1,3] . Although some of these systems have been implemented
at several sites, many of them work most productively and smoothly at
the site at which they were developed. Typically, the task of "gluing"
an externally-developed EMRS architecture to even a rudimentary
"legacy" clinical data-base is substantial. One important reason for
this difficulty is the lack of standardization noted by the National
Library of Medicine's RFA (LM-94-002).

Several standardization efforts are currently under way to define
specifications for structure and content of EMRS (e.g. ASTM
E1384-1991, ASTM E1238-1991, HL-7 [16] , Medix from IEEE, and ASTM
Draft E31.12-4 (Revised 7/93)). Nonetheless, widespread agreement on
these standards or their implementation is lagging.

Structured Medical Vocabularies

The simplest method of storing patient information would be to allow
users to enter description that they feel are most accurate. While
this use of uncontrolled text simplifies the entry of information, it
complicates retrieval. Algorithms have been developed to allow
retrieval of uncontrolled text with accuracy that rivals material
indexed with a controlled vocabulary [17,18] . However, in the patient
care domain, the highest accuracy possible is needed. For example,
gynecologist may feel he or she is being accurate by describing a
patient as having a Òcervical massÓ. A neurologist or
orthopedist viewing the same record may draw a different and erroneous
conclusion about the patientÕs problem. This kind of problem can
only be avoided by controlling what terms are used in the content of
the record.

Numerous controlled vocabularies have been developed in the health
care domain (MeSH, ICD9-CM, SNOMED III, Read, Gabrielli, NANDA). Each
of these vocabularies has been developed for a specific purpose. None
of these vocabularies has been developed to encompass all the purposes
of an EMRS. For example, MeSH is one of the broader and better
structured of the controlled vocabularies mentioned. It was developed
specifically for indexing medical literature. Because there are few
papers that discuss patient history and physical exam findings as the
focus of the paper, MeSH is relatively weak in this area as compared
to the Read vocabulary. At the same time, the Read vocabulary, which
was developed to describe physical exam findings, is relatively weak
in the area of procedure terms.

Combining the strengths of each of the existing vocabularies might
seem like a solution except that there are areas where the terminology
in different vocabularies conflicts. For example, where
ÔAbortionÕ in MeSH is meant to mean spontaneous abortion, the
same term in CPT is assumed to mean induced or surgical abortion.
Similar examples can be found by looking at terms for various
substances that are found in patient blood samples. In one vocabulary
they refer to the substance while in another vocabulary they refer to
the act of measuring the substance.

The NLM's Unified Medical Language System (UMLS) project is developing
tools to address problems with linking multiple medical resources [19]
. In the area of multiple medical vocabularies, they have developed a
ÒMetathesaurusÓÊá(Meta) [20] . Meta is not a controlled
vocabulary. It is a database of information about existing controlled
vocabularies. The current version includes many of the vocabularies
mentioned above. For medical resources that use the included
vocabularies, Meta can provide an independent application enough
information so that it can navigate through the vocabularies of those
medical resources. In some instances, Meta can also allow a
ÒtranslationÓ between the vocabularies of two applications. For
medical resources that are not included in Meta, Meta includes a broad
enough sample of terms in many domains so that a Ògood guessÓ
can be made as to a term that will be included in the medical resource
[21] .

Another important tool developed by the UMLS project is the UMLS
Semantic Network [20,22,23] . A generic semantic network is a
knowledge structure which defines concepts primarily on the basis of
their relation to each other. The UMLS Semantic Network is a small
controlled vocabulary which provides information on how to categorize
vocabulary terms and how those terms might be related to each
other. As mentioned above, the context or purpose for which a
particular term is being used can change our interpretation of the
meaning of that term. By categorizing terms with Òsemantic
typesÓ we can reduce the ambiguity of a term. For example, Meta
contains two terms ÒAbortionÓ one of which is typed as a
ÒDiseaseÓ while the other is typed as a ÒProcedureÓ. Using
the Semantic Network and Meta together, a controlled vocabulary that
is not included in Meta can be typed and linked to the existing
Meta. These local extensions to Meta improve it's ability to make
Ògood guessesÓ about the term a user wishes to use to interact
with local medical resources.