Seeing the light on LIS data conversion
One alternative strategy:
cold or hot?
January 2002 Eric Skjei
The once-complex process of converting data is yielding slowly
to expertise and experience.
As labs move from one vendor’s system to another’s, upgrade their
LISs, and interface with other systems within their organizations,
they need to make the data in those systems usable across widely
varying database structures and formats. New tools help, but data
conversion still demands time, attention, and expertise. And the
demand for conversion is growing.
"It’s almost a quarterly need for us," says Michael Becich, MD,
PhD, chairman of pathology at the University of Pittsburgh Medical
Center Shadyside Hospital. Over the last five years, Dr. Becich
has led a process in which 18 hospitals in the medical center system
work to merge their anatomic and clinical pathology systems into
a common platform. "Every time we retire an old LIS," he says, "we
aggressively look at the value of old data and convert anywhere
from five to 15 years of that data into the new system."
The need for data-conversion services, some vendors say, has become
so commonplace that it has in effect become a basic business service.
"At this point, almost every sale we do is a replacement," says
Steve Tablak, CEO of Tamtron Corp. "We find that almost every contract
we sign now includes a data-conversion service."
Widespread as the need for data conversion has become, there is
general agreement that the need is stronger in some areas of the
lab—namely anatomic pathology and blood bank services—than
others. "There is the expectation in anatomic pathology that clinicians
will have the history of AP results indefinitely," says Dov Kaufman,
manager of engineering services for Sunquest Information Systems.
"If a patient had cancer five years ago, the treatment may be different
than if she or he did not, and that’s one reason why physicians
want to see all of the data."
Having as much historical data available as possible, says Dr.
Becich, means that a physician can factor in information obtained
from previous specimens when new tests are ordered. "When a patient
comes in with a breast cancer, or we do a repeat biopsy on a breast
or a lump on the skin, knowing that 10 years ago this same patient
had a breast cancer removed at hospital A means that even if she
or he is now at hospital B, we can still quickly and efficiently
work our way through the process of deciding what we need to do
with that current sample to give the patient and the physician what
they need," he says.
While it is generally true that anatomic pathology and blood bank
systems evince the clearest need for access to legacy or historical
data, and thus require a high level of data conversion as systems
are integrated, upgraded, or replaced, some areas of clinical pathology
also require this intensive level of data conversion. "In general,
clinical pathology testing tends to be oriented toward obtaining
a snapshot of where you are today," Dr. Becich says, "and it’s probably
not as useful in most respects to have such extensive, immediate
access to years and years of past data."
There are, however, exceptions. "In some areas—for example,
microbiology, virology, hematology smear interpretations, serial
results on liver function tests, some endocrine tests—there
is an equally high value in having that depth of data quickly available,"
Dr. Becich says.
Heterogeneous conversions
Data conversions between different vendors traditionally have
been the most challenging. "Understanding the way the fields are
defined, at what level they are defined—all that is a very
definite challenge," says Sunquest’s Kaufman. "When clients go from
our system to another vendor’s, we typically spend many, many hours
on the phone with those other vendors trying to map our fields to
theirs and vice versa." These types of conversions are time-consuming
and people-intensive, likened by some to building interfaces between
heterogeneous systems in their degree of complexity.
The reasons such interfaces stir fear in the hearts of even the
most seasoned IT pros range from small to large. A relatively minor
example: Older LIS databases often stored data in a coded format.
"Instead of saying ’This was called to the doctor,’ the field may
just have a code in it, like ’12QR47,’" says Hal Weiner, president
of Weiner Consulting, Eugene, Ore. Granted, there are almost always
ways to decipher these codes and to automate the decoding process
in the new system, but doing so represents a speed bump in the conversion.
Another relatively minor example, this one from Gregory Francis,
president of the information technology consulting firm Korchek
Technologies, about a conversion his firm recently undertook: "The
old database had a comment field in which users had been allowed
to become very verbose, but the new system had space in its comments
field for only 60 bytes of comment." Francis adopted this solution:
He analyzed the data being imported from the old system, and when
the field contained more than 60 bytes of comment, he included a
simple notice that the user should check the old system—still
kept online for just this type of need—for the additional
content.
Other complications run deeper, pose more of a challenge to the
data-conversion process, and pertain to inevitable differences in
how LIS vendors organize their data. One example, Weiner says, arises
in conversions from databases that are test-oriented to those that
are results-oriented.
Say, for example, a laboratory needs to convert data from a test-oriented
system that stores sodium results only or primarily as part of a
chemistry profile to a foreign database that, being results-oriented,
creates the same profile by assembling individual results—including
sodium—as needed. Mapping the sodium result data from one
to the other, in this case, could be unusually tricky.
In-house conversions
Data conversions sometimes occur within a vendor’s own user community.
A client may want to reorganize its data; may have changed hospital
information systems and may, therefore, need to change the format
of, for example, the patient ID number; or may simply be upgrading
to a newer system from the same vendor.
In some respects, says Kaufman, in-house or internal conversions
are easier, as one might expect. "In general the database structures
will be quite similar," he says. Not exactly the same—some
conversion inevitably is required—but close enough to make
the conversion more straightforward. Information stored in the old
system at the patient level will, in all likelihood, still be stored
at the patient level, rather than, say, at the encounter level.
But the process still takes time and attention. "For example, we
still need to validate the data as if it were coming from a foreign
system," Kaufman says.
Even an otherwise routine installation of a new LIS is increasingly
likely to require data-conversion services. "Rather than having
to manually enter the names of all your physicians, nursing units,
and so on, it can be very beneficial to be able to import that data
from an existing system that’s already online," Weiner says. And
that process will more likely than not require some level of data
conversion.
Effective and affordable new tools are now available to simplify
the process. "These new tools basically allow you to define a database
schema for just about any kind of file structure out there and then
say, ’I want you to convert it from this method to that method,’"
Weiner says. "Once you have it all mapped out, the tool does much
of the work for you."
Much but not all. Helpful as they are, these tools don’t fully
automate the process and probably never will. "I’m running most
of those same tools," says Korchek’s Francis, "and the process is
still a headache." Tools alone won’t confirm, for example, that
the John Smith born in 1932 and seen as an outpatient with no medical
record number is the same John Smith born on the same date in 1932
but seen as an inpatient and assigned a medical record number.
Making sure it’s right
Even when the conversion is over, it’s not over. Any translation
of clinical data requires intensive error-checking. "Data integrity
is more than 50 percent of all the work involved in a conversion,"
Dr. Becich says. "There will be errors, and that fact alone means
that there will also be a need for hours and hours of time spent
going over the data to make sure those errors are found and fixed."
Asks Francis, "What happens if you’re off by one byte in a glucose
result that was 67 but is now appearing in the chloride field in
the new system?" What if that small dislocation means that now your
chloride is 67 and your glucose is, say, 6? "That’s one big reason
why the data-conversion process is definitely not plug-and-play,"
he adds.
It is in the related area of what he calls "patient mapping" that
Tamtron’s Tablak sees the real challenge of converting data. "Most
modern systems have a method to decide if two patients are the same
patient and, therefore, if all of their historical records are the
same," he says. "But a lot of the older systems didn’t necessarily
have a sophisticated means of doing that."
So in the conversion from a less sophisticated to a more sophisticated
system, Tablak continues, much time must be spent with the customer
to determine the best way to map patient cases reliably and accurately.
"Sometimes you have records for a John Smith in which a Social Security
number is present in one case but isn’t in another, or a John Smith
who was an inpatient in one case and an outpatient in another,"
he says. "How do you decide when these John Smiths are one and the
same?" This requires sitting down with the customer and deciding
on a policy for resolving these ambiguous and inexact matches. It
may even require manual inspection—no small task when the
volume of cases needing to be imported is large and stretches back
for years.
Lessons learned
How can laboratories minimize the stress of the data-conversion
process? "Right up front, right in the beginning, identify the data
elements that you want to get," Francis says. "Base this decision
more on what the new system can accept as opposed to what’s in the
existing system." Francis recently worked on a conversion in which
the importing system was able to import data on the patient reaction—that
is, positive or negative—to antigen typing. Although the data
existed in the old system, they were extremely difficult to access.
In such a case, he says, the key question then becomes, How valuable
are these data? Are the data clinically significant? How often does
the lab need to access the data? Does the lab need historical data
on this topic, or will current results serve its purpose just as
well?
Second, labs should always consider the possibility that they
may be able to store some data offline, in a form that is not live,
not tightly integrated with the current database, in what is sometimes
referred to as a "cold" solution. Says Weiner, "A cold solution
typically archives data on a medium like laser disk or CD and indexes
it so the client can quickly find it, view it, and print it in a
report format if needed." This solution is good for data that do
not need to be available on demand in the lab’s main database. "By
contrast," Weiner explains, "hot solutions convert the data so it
is directly and immediately usable from within the new LIS, so that,
for example, prior results appear on the cumulative report with
the current results or so that older data can be used for performing
a delta check." (See "One alternative
strategy: cold or hot,".)
In either case, cold or hot, storage is rapidly disappearing as
a consideration in the process. Says UPMC’s Dr. Becich: "There still
exists a perspective that argues that there is so much data generated
by clinical pathology systems that it’s impossible to keep it all
online for 20 years. That’s hogwash today, with the dropping cost
of storage, and, quite frankly, I think we’re losing some very valuable
opportunities because of it."
Finally, say many data-conversion specialists, while data conversion
can be time-consuming and rigorous, it is still just a process,
one like many other IT endeavors, not a mystery or an arcane bit
of witchcraft. "You can’t just hook one machine up to another, press
a button, and watch the data flow magically from one system to the
other," Dr. Becich says. "The data-conversion process is still 90
percent personnel power and only 10 percent computer power." Adds
Tamtron’s Tablak, "There are enough of us now in the LIS industry
who have spent enough time doing this kind of work that we understand
the other vendors’ database structures very well, and that means
that the process is not only significantly easier than it was five
or 10 years ago but even somewhat routine."
Changing the rules of engagement
What isn’t routine yet is vendors who collaborate.
"We’ve actually been trying to help clients who are leaving Sunquest,"
says Ken Kark, Sunquest’s director of marketing communications.
It’s the professional thing to do, he notes, and it’s also good
business. Sunquest has had sufficient experience with clients who
move to another vendor’s system, stay with it for a year or two,
then decide to return to Sunquest, to see the long-term potential
of being helpful when customers want to leave. "We do it because
we want to maintain our reputation for excellent customer service
and satisfaction," Kark says, "and because we think they’re going
to come back."
Tablak takes a similar position. "You want to treat a customer
who is leaving as well as possible because you hope they’re going
to come back," he says. "We have the attitude that if they’re not
choosing us, they may not have fully considered their decision,
and that in a few years they may be back."
Enlightened as it is, this practice seems still to be the exception
rather than the rule. "The [outgoing] vendor rarely cooperates,"
reports Korchek’s Francis. "And if they do, they typically want
to charge an arm and a leg to do so."
"Vendors still do not want to work with other vendors," concurs
Tom Manning, vice president of sales and marketing for Mesa Corp.,
which specializes in archiving, imaging, and workflow management
systems. "That was true 20 years ago and it’s true today," he says.
"Vendor A is losing the business, vendor B wants the information
that is on vendor A’s system, and vendor A’s attitude is, ’OK, see
if you can get it, and don’t call me.’"
Even when the outgoing vendor does cooperate, the value of the
help may be minimal. "On one of our recent conversions," recalls
Francis, "the date of birth was stored in, packed way in the legacy
system, and we had to unpack it." Although the vendor being replaced
did supply an algorithm for accomplishing this, Francis quickly
discovered its limitations. "It didn’t account for leap years before
1900, which caused a problem that we didn’t notice until we started
examining the data and found that some older birth dates weren’t
matching up."
Ultimately, the incentive that overrides these differences and
drives cooperation is that everyone involved in the process has
a fundamental obligation to patient care. "Physicians and patients
depend on this data," Dr. Becich says. Agrees Tablak, "We’re in
this business because it’s a patient care business, and we certainly
don’t want to interfere with the patient care process."
Eric Skjei is a writer in Stinson Beach, Calif.
|
|
|