Merging terminologies for a new mother tongue
April 2002 Karen Southwick
Three years of effort by the CAP and the U.K. National Health
Service have culminated in the first release of SNOMED CT, the long-awaited
merging of two highly regarded clinical reference terminologies.
SNOMED CT First Release encompasses more than 325,000 fully specified
concepts complemented by 800,000 descriptions, compared with 121,000
concepts and 190,000 descriptions for its predecessor, SNOMED RT.
By incorporating the National Health Service’s acclaimed Read Codes,
SNOMED CT has expanded into such areas as primary care, nursing,
and specialty medicine.
"It’s a milestone," declares Kent A. Spackman, MD, PhD, chair
of the SNOMED International Editorial Board and associate professor
of pathology at Oregon Health Sciences University, Portland. When
the CAP and the NHS first reached agreement in the spring of 1999,
"our target date was to have a version of CT available within three
years," he says. "We’re very pleased with making the deadline."
The initial goal, says Dr. Spackman, was to have a preliminary
version of SNOMED CT available for modifications and revisions in
December 2001. "We sent the CD (with that version) back to the U.K.
on January 31, 2002," he adds. For a project that started three
years ago and involved meshing two terminologies with hundreds of
thousands of concepts, that’s an impressive on-time performance.
Beyond SNOMED’s traditional purview of pathology coding in laboratory
information systems, SNOMED CT has a variety of other uses, including
supporting the electronic medical record, standardizing surgical
records, coding patient problem and diagnoses lists, supporting
order entry, tracking infectious diseases and cancer, and facilitating
bioterrorism surveillance.
SNOMED CT First Release has thus far been shipped to only a select
number of alpha users, primarily companies in six countries that
develop systems for their customers. "They were able to provide
us with feedback on the structure and concepts" that will then be
incorporated into future releases, Dr. Spackman says.
The next release is scheduled for July, and Dr. Spackman expects
"substantial changes" related to fixing problems. "Most of what
we do for the July release will be essentially repairs," he says.
"There will be a few new concept additions." SNOMED International
intends to issue releases of SNOMED CT every six months for an indefinite
period. Over time, as repairs are completed, "a larger proportion
of the changes will be new additions, concepts, and terms," he adds.
User response
Early adopters of SNOMED CT say they are pleased with the technology,
which is fairly clean for a first release. They add that the enriched
hierarchical structure is more accessible than previous versions
and can handle more complex relationships. SNOMED CT will not, however,
go into widespread use until vendors of laboratory and hospital
information systems begin to incorporate the technology into their
offerings.
James R. Campbell, MD, professor of medicine at the University
of Nebraska Medical Center, Omaha, is a longtime user of SNOMED
and a member of the SNOMED International Editorial Board. "We use
SNOMED [currently RT] throughout our system," says Dr. Campbell,
including as the basis of an electronic medical record. "We will
be upgrading [to CT] when our internal analysis shows us that the
release is ready for clinical ’prime time.’"
The infusion of "a large amount of language and concepts from
the Read Codes" means SNOMED CT has wider applicability in health
care settings, Dr. Campbell says. For example, the University of
Nebraska compared the use of ICD-9 versus RT versus CT in the capture
rate for symptom codes of patients entering the emergency room and
found that CT outperformed the other two.
The study, authored by James McClay, MD, assistant professor in
the Department of Surgery/ER, looked at 310 encounters in the emergency
room during the past year. ICD-9, the national standard for diagnostic
coding systems, captured 40 percent of the symptom codes. SNOMED
RT captured 87 percent, and CT was at 93 percent.
"That 93 percent is leagues ahead of what you can get with anything
else," says Dr. Campbell, who handled the SNOMED coding for the
study. The figure means that in 93 percent of those ER patient encounters,
SNOMED CT captured the content of the patient’s symptoms to the
satisfaction of the reviewer. "These were 93 percent exact matches,"
he says.
Drs. Campbell and McClay are pushing to make SNOMED the coding
system of choice in the ER. The only obstacle is billing mandates,
which require ICD-9 coding. Dr. Campbell, however, has already developed
a customized interface that links RT with ICD-9 coding using SNOMED
mapping tables to integrate the electronic medical record with billing,
and the same could be done with CT. "The point is that SNOMED, in
general, and CT, in particular, are clinically much more accurate
than ICD-9," he says.
With the Centers for Disease Control and Prevention pressing emergency
rooms and clinics for more detailed information to detect the possibility
of bioterrorism, SNOMED is a better choice than alternative codes,
Dr. Campbell says. For example, a person with anthrax might come
in with respiratory problems and skin rashes. There’s no way ICD-9
would distinguish that from many other ailments, he says. "SNOMED
is much more specific." The study at the University of Nebraska
"demonstrates how much it could offer."
As is typical of a first release, SNOMED CT is still undergoing
editorial review and expansion, Dr. Campbell says. But in his review,
he has found CT to be improved in the way it expresses relationships.
Dr. Campbell does guideline support for the University of Nebraska
and is particularly interested in such areas as clinical course.
The CDC, for instance, recommends offering a flu shot to all patients
with chronic cardiovascular or pulmonary diseases. "The question
is, what constitutes that?" he says. With ICD-9, it’s more difficult
to separate chronic from acute conditions. "With CT, there’s a very
simple relationship which clearly defines itself" and makes obvious
who should receive flu shots. "You can ask, ’Is this a chronic cardiovascular
disease?’ and get back an immediate answer." SNOMED RT would define
a condition as cardiovascular but not as chronic or acute.
Another example of CT’s richness relative to relationships involves
diabetes. If someone comes in with a foot ulcer, "most coding systems
don’t allow for the inference that it was caused by complications
of diabetes. CT does," Dr. Campbell says. "It’s fair to say there’s
a substantial improvement in the quality and number of semantic
relationships with CT."
The SNOMED CT development team also worked to improve the user
interface. New editorial features, such as menus of choice, allow
for a narrow or broad focus on certain areas of medicine. "It makes
it much easier for a vendor to implement" for a particular use,
Dr. Campbell says. In addition, SNOMED CT enables users who don’t
need all of the complexity "to navigate in their own area without
being aware of others." With CT, he adds, "the vast majority of
users should not be aware there is such a thing as a code."
Health Language, a Denver-based developer of terminology engines,
would like to make SNOMED CT even easier to use. "SNOMED is a very
thorough offering," says Brian Levy, MD, vice president and medical
director of Health Language. "Our industry-leading vendor partners
are able to maximize SNOMED’s benefits in their solutions through
the use of our language engine technology."
Health Language’s clients are not clinicians but those who sell
to clinicians, including electronic medical record companies and
medical online content providers who want to standardize on a terminology.
"We’re the ’Intel inside’ the terminology," Dr. Levy explains. "Doctors
and nurses will not be browsing SNOMED hierarchies at the point
of care," he says. "They need tools like ours to pull out information.
Our goal is to make SNOMED transparent."
Health Language used SNOMED 3.5, then RT, and is now an alpha
user of CT, having started testing the latter in early fall 2001.
CT Alpha "was not designed to look at breadth of content," Dr. Levy
says, but rather to allow users "to look at the model and test the
improvements, like table structures." Health Language didn’t encounter
any unusual problems in testing CT, he adds. "We did give feedback
about the new way they modeled the body site hierarchy."
In the way it functions, "CT is not fundamentally different from
RT. It uses the same model and relationships," Dr. Levy says. But
he has noticed improvements in the content, such as the added material
from the Read Codes. "There’s a more complete set of signs and symptoms"
and a richer hierarchical structure. "SNOMED was always good at
relating a disease of the heart to the heart itself. CT has added
new qualifiers," he says, such as relating a disease to its possible
severities. For example, asthma might be related to the lung but
also be described by such qualifiers as severe, moderate, or mild.
Dr. Levy appreciates SNOMED because "it’s the terminology that
most closely resembles the actual words used by clinicians at the
point of care." And adding the primary care material from the Read
Codes moves SNOMED CT closer "to the everyday practice of medicine,"
he says. "It’s also more useful in outcomes research."
TheraDoc Inc., a private company in Salt Lake City that builds
clinical decision support tools, uses SNOMED within its engines.
"Our premier product is a new version of the Antibiotic Assistant
aimed at clinicians," says Robert Hausam, MD, director of systems
architecture. Given clinical data on a patient’s condition, the
software makes therapeutic recommendations on antibiotic treatment.
The technology was spun out of work done at LDS Hospital, part of
Intermountain Healthcare.
Dr. Hausam says SNOMED is especially useful in the area of organism
hierarchy. "We have our own vocabulary, but it’s based on SNOMED
as the core. We ship SNOMED with our product," he says. The technology
will also be used in future TheraDoc products, encompassing such
areas as pain management, depression, and diabetes.
TheraDoc started with SNOMED RT and was an alpha user of CT. "We’re
waiting to get our hands on the real release," says Dr. Hausam.
"What I’ve seen in the alpha version and what I expect to see in
the [real] release suggests a much richer set of concepts and relationships.
I’m a longtime fan of the NHS Clinical Terms [Read Codes] and I’m
very excited to see the two best systems merged."
One improvement Dr. Hausam singled out is anatomic modeling. SNOMED
RT used "part of" to handle certain hierarchical relationships,
as in "the lung is a part of . . . the respiratory system." Separately,
there was the "is a" relationship hierarchy, as in "E. coli is a
. . . gram negative rod." SNOMED CT combines these hierarchies to
allow anatomic "part of" relationships to also be modeled as "is
a" relationships. "It’s much easier to use for what we do," he says.
The transition from SNOMED RT to CT should go smoothly, Dr. Hausam
says. "The new table structures were pretty easily incorporated.
You could essentially put it right into the existing structure."
SNOMED CT also adds new layers of functionality, such as formal
mechanisms for recognizing subsets. "Certain types of doctors want
a particular subset of terms," says Dr. Hausam. With RT, "we devised
our own," but CT should make it easier to tailor a set of concepts
or relationships for specialty users. SNOMED CT also allows users
to customize how they display terms.
Upgrade timetable
SNOMED users who upgraded to RT, including the University of Nebraska,
did so with the knowledge that it was a bridge to CT, but many users
remained on earlier versions because "they were waiting for the
other shoe to drop," Dr. Campbell says. With the release of SNOMED
CT, "at least that will no longer be a problem." Dr. Levy says Health
Language will provide SNOMED CT and RT in its applications and "help
clients make the shift. People can’t stay on RT [which will no longer
have new releases]. CT is the wise choice."
Dr. Campbell expects CT to be widely adopted, as anybody using
Read or SNOMED will eventually upgrade. However, "the time frame
is somewhat undefined," he says. "The product has to stabilize and
everybody feel comfortable that it has done so."
The U.K. National Health Service has said it will spend the next
12 months conducting an intensive evaluation and gathering feedback
for CT "with the notion that there will be an endorsement in mid-2003."
Likewise, it takes eight to 12 months for LIS/HIS vendors to handle
a coding change and another 12 months for users to implement it.
"It will be a while before you actually see CT appearing in PC systems,"
Dr. Campbell says.
Pat Button, Ed, RN, senior product manager for clinical knowledge
and integrated documentation at Cerner Corp., agrees with Dr. Campbell.
Cerner is using SNOMED RT in its PathNet LIS and is also "actively
pursuing broader uses of RT in our clinical documentation applications,"
such as order entry, care planning, and documentation. As SNOMED
matures, "that approach will extend to CT," she says.
Button adds that SNOMED RT has already proved to be very effective,
especially for laboratory uses. "We’re very pleased with what SNOMED
has done in incorporating wider terminology," she says. Adding CT’s
enriched terms in primary care and nursing "will make SNOMED even
broader and deeper. It’s what the industry has been needing for
a long time."
As a nursing informaticist, Button says she appreciates that SNOMED
CT incorporates all the major nursing codes, including the North
American Nursing Diagnosis, Nursing Outcomes Classification, and
Home Health Classification. "This will provide real value for uses
across the care continuum," she says.
Working groups
Dr. Spackman says a team of more than 100 was closely involved
in developing SNOMED CT, including staff from the NHS, CAP, and
Kaiser Permanente System, which was the primary development site
for CT’s predecessor, SNOMED RT, and also contributed terminology.
Kaiser, one of the nation’s largest managed care organizations,
and the NHS helped bolster SNOMED CT’s primary care content. Other
"significant contributors," says Dr. Spackman, were the American
Academy of Ophthalmology and the American Veterinary Medical Association.
Dr. Spackman cites three advantages of SNOMED CT. The first is
increased content coverage. Research by the National Library of
Medicine found that the two most effective codes for representing
a medical record were SNOMED and Read, with more than 90 percent
content coverage. Based on frequency of use, "you’re talking well
over 99 percent coverage" for the two of them combined, he says.
The second advantage is ease of retrieval. With previous versions,
"people used to complain that you could put things in but it’s uncertain
whether you can always retrieve them. That problem has been solved
with CT," says Dr. Spackman. "It’s reliable for retrieving information."
Third, CT is designed to accommodate multiple levels of technical
sophistication. "If all you want to do is put in one code, you can
do that. CT also has the resources to support more advanced capabilities,
all the way up to a fully electronic medical record," Dr. Spackman
says.
All of this was user- and vendor-driven, he adds. The alpha testing
was conducted in six countries among such specialties as surgery,
obstetrics/gynecology, pediatrics, allergy, internal medicine, nursing,
podiatry, applied health professions, geriatrics, and primary care.
"We have met the requirements that users and vendors expressed,"
he says. Among them, the ability to analyze old data, use many different
terms to find the same code, index concepts, have a common mechanism
for doing look-ups, and create subsets and extensions.
SNOMED originated more than 35 years ago with the development
of SNOP in 1965. Then came SNOMED v. 1 in 1976, v. 2 in 1979, and
v. 3 in 1993, followed by RT in November 2000, and now CT. "CT totally
subsumes RT," says Dr. Spackman, and is the product for the foreseeable
future. "We have nothing on the horizon that will fundamentally
change it."
Karen Southwick is a writer in San Francisco.
|