For four years running, the IBM Blue Gene/L has qualified as the biggest and fastest computer in the world. It clocks sustained computation speeds of 360 teraflops; in laymen’s terms, it’s blindingly fast. But this high-performing computing environment is not performing complex military simulations or modeling weather across the entire planet. It needs all of its capacity just to cope with the task of mapping the human genome.
And that gives a hint of the information technology challenges that molecular diagnostics presents for laboratory information systems that were designed for traditional clinical laboratory testing.
While immunoassays have been around for a long time, “current LISs are just not equipped to deal with the whole new area of biomarker panels, genomics, and proteomics, not to mention molecular imaging,” says Hal Weiner, president of Weiner Consulting Services in Florence, Ore.
“In chemistry or regular clinical laboratory testing,” explains Federico A. Monzon, MD, medical director of medical diagnostics at The Methodist Hospital, Houston, “the information management is at the beginning when the order comes in and at the end when a result is reported. Everything in between is managed by an analyzer or the instrumentation.”
“That’s not the case in molecular. We don’t have large automated analyzers, a lot of the testing is manual, and as the sample goes through the laboratory, information is generated to determine whether you can proceed to the next step or not. And that information needs to be handled and managed.”
That’s where there’s a lack of support from current lab information systems, Dr. Monzon says.
For example, the frequency of molecular testing often needs to be adapted according to the test. Many guidelines are in flux, such as those on how to use test results in oncology, or how often to follow up a patient with infectious disease testing.
The guidelines for HIV testing are fairly well established, but for many other tests they are still evolving, he notes. “You would think HCV would be fairly established, but there are a lot of new data and recommendations from certain groups to do the testing more often to detect response to therapy earlier rather than later and make adjustments as soon as possible.”
“How long after you change therapy do you measure viral load? What’s the most effective test ordering patterns for patient management? Those are some of the areas we have to struggle with.” The variations become especially complicated when hospitals implement ordering on a different system from the one used for the laboratory. Methodist Hospital, for example, orders infectious disease testing through the hospital electronic medical record system.
How to handle such complex information has become a more pressing issue as the molecular diagnostics market continues to grow at 15 to 25 percent annually—a rate that is three times to five times as fast as surgical pathology or the rest of clinical pathology, where the growth rate is more stable, Dr. Monzon says. In the molecular diagnostics laboratory at The Methodist Hospital, volume grew 21 percent from the year before.
Proper management of consent forms presents a separate set of challenges for molecular diagnostics on the ordering side. “With most of your current clinical tests, you don’t request consent. The patients sign a general consent when they enter the hospital, to everything necessary to treat you, and that includes lab testing. But when you do genetic testing, you need to do informed consent. You need to tell them exactly which tests you are going to do, what the possible results are, and the implications for them and their families.”
Different hospitals also enforce consent requirements differently. “For example, some laboratories require the physician to attest that they have obtained informed consent. Other labs indicate that by ordering the test, the physician is responsible for obtaining the informed consent. In some places, tests with extremely serious implications of a positive result, like Huntington’s disease, have specific consent forms that have been developed by the lab.”
In molecular diagnostics, “one of the big differences is that the sample you test is not the specimen you receive,” Dr. Monzon says. “You actually have to purify the DNA or the RNA that you’ll need; that’s your sample to analyze. And that’s very different from most other clinical tests where you measure analyte directly in the sample you have.”
The LIS has to monitor those testing workflow processes. “You need to know what methods are used, what instruments—there’s no universal method, and not all methods are suitable for all analyses,” Dr. Monzon says. “If you’re going to measure viruses in a sample, there are extraction methods that are better for that purpose. If it’s a tumor sample, then you have a different way of handling the sample. Unlike other areas of the laboratory, where you either have enough sample or not, you might need to ask, Did I obtain enough DNA to do three tests or only one?”
In addition, myriad analytical techniques are available, even for the same test, and all of this information needs to be recorded and tracked. “Beyond which samples are running, you want to record which reagents are being used, how the process was performed, and who did it, in case something fails and you need to go back.”
In the area of reporting, molecular diagnostics creates challenges in providing physicians with the information they need. “With a blood glucose level, the actual value obtained from the instrument is reported and the clinician knows what to do with this information. Some tests in the molecular diagnostics laboratory are also relatively straightforward. For example, you can report a viral load value or the presence of certain well-known mutations, and most physicians will be able to make decisions based on that information.
“But with more complex texts, like gene sequencing, it gets more complicated because they need to do interpretation of the relevance of a sequence variation.” Is it something that has been seen before in patients with this disease (that is, a mutation)? Is it actually causing the disease or not? Could it be a polymorphism? “Sometimes we’re not sure,” Dr. Monzon says. “This is akin to the interpretation of morphologic data done by surgical pathologists. Both of these interpretive results usually get reported as a long text document.”
“So molecular is kind of a mixture of clinical and surgical pathology, where some tests might be better handled in an automated fashion,” he says, “as many other clinical tests are done, while others require more text-style reports where you want to explain the significance of a finding and perhaps link to different types of information.”
The LISs now in use are not designed to handle the complex needs of molecular laboratories, and that’s why most laboratories record that information on paper and file it in folders or have developed custom databases. However, software solutions are appearing on the market.
“The first completely functional module to tackle molecular diagnostics was the Cerner Millennium Helix module, and several laboratories are actually up and running with it,” Dr. Monzon says. Methodist Hospital is going to move in the near future to another solution for molecular diagnostics laboratories called SoftGene, in development by SCC Soft Computer.
Another major player is Misys Healthcare, which is using the approach of enabling communication between its AP and CP products to address the needs of molecular, Dr. Monzon says. “So instead of developing a new module, they’re enabling you to use the tools you need from CP and tools you need from AP in your molecular laboratory and enabling these systems to communicate and exchange information.”
Despite the action on the part of LIS vendors, IT in molecular is not keeping pace with the needs of molecular diagnostics laboratories, Dr. Monzon says. “There are new technologies coming out, and clearly none of the systems are completely prepared to deal with them.”
“For example, we’ve seen an explosion in testing with array platforms. And this goes from very focused arrays—testing for a limited number of mutations in certain genes—like the ones used to test for warfarin sensitivity, up to arrays for comparative genomic hybridization [CGH] or SNP genotyping, with thousands to millions of probes in one array.” CGH and SNP arrays, he says, look at thousands of measurements to determine whether a patient has abnormal chromosome complement or has alterations or abnormalities in their chromosomes that are detectable with these technologies.
“These array technologies place a significant information burden on the LIS—on the laboratory, actually—because none of the LISs are able to handle this type of information. Most of us have to deal with it offline, either in homegrown databases or data repositories that are not real LISs as we know of them in the clinical sense,” Dr. Monzon says.
The offline solutions can be something as simple as Access, which comes with every Microsoft Office package, or a more sophisticated application with more robust databases such as Oracle or MS-SQL. “There are a lot of flavors of how people deal with these at this point.”
In many ways, Weiner notes, medicine has traditionally been practiced through a process of trial and error. “You try this, you try that, you rule things out, you run all these tests and hope something will come out.” But in the next five years, he says, molecular diagnostics is likely to change that.
“Right now, people go and get a laboratory test to find out if they have a disease. Very soon, people will go to have a laboratory test to find out if they will have a disease in the future.”
But that’s not the most exciting piece of molecular diagnostics, in Weiner’s view. Backed by the science of proteomics, there are already drugs in clinical trials that could potentially prevent an individual with the target gene, or genes, from contracting a disease. To probe these drug metabolisms or drug interactions for individuals requires understanding the genetics of individuals and the genetics of disease.
“It’s the mapping of those two together that’s going to make the changes,” Weiner says. It will take only another 10 or 15 years to identify the millions and millions of genetic variations involved, and from the LIS vendor’s standpoint, this will soon mean an exponentially larger set of data.
Cerner Millennium Helix and SoftGene are the two live options that Dr. Monzon says he is aware of now. But there are other specialty products that cover specific areas such as stem cell laboratories, and in many places those suffice. “If a laboratory only focuses on infectious disease, it could probably just run with its current clinical LIS that handles numeric data. Or other laboratories that only do genetic testing and can use one of the very focused products.”
“But if you’re in a more complex environment like a large academic center, or even I would say a large community hospital, where you’re looking at handling not only infectious disease but also genetics and oncology molecular testing, then you really want one of these larger information systems that will help you manage most aspects of the complex testing protocols.”
Weiner says there has to be a collaboration between the IT and genomics fields, but that it can’t just be the clinical or pathology components of genomics. “It has to include the molecular imaging component. All that information right now is either just a blob of text or it’s an image that has to be interpreted somehow, or it’s a massive genetic database that current LISs have no way of handling.”
This is where radiology and pathology have started to converge, he adds. Molecular imaging combines molecular biology and in vivo imaging to enable the visualization of the cellular, he says. “You can inject a biomarker and look at how the patient reacts to it—you can actually image down to very fine molecular changes and look at the cancer cell at the cellular level.” This capability may erase the traditional lines between pathology and radiology, he predicts.
Some vendors are trying to develop a commercial off-the-shelf product to unify all these pieces, while a lot of universities are promoting an open-source solution, Weiner says. But what’s stopping that from happening is that there are too many standards—so, in effect, there is no standard. At bottom the issue is the difference between “data” and “information,” he points out. “If I have a sodium value of X, that may have a different meaning in my system than somebody else’s because my reference ranges are different.”
Vendors are now deliberately taking a proprietary approach in developing their software modules, and the federal government has been cautious and not intervened with a mandatory standard, Weiner says. “There’s been no incentive to force the vendor community to adopt a consistent set of both content and transport standards. And until there’s a financial impact to the industry for not complying with a standard, there won’t be one—that’s just reality.”
The molecular diagnostics laboratory at MD Anderson Cancer Center in Houston has seen steady growth over the past five years, says medical director Dan Jones, MD, PhD. “Since we’re a cancer institute, about 75 percent of our testing volume is related to followup for patients with bone marrow transplants, leukemias, lymphomas, and to a lesser extent infectious disease testing.”
“We do some prognostic markers at time of diagnosis, mostly leukemias but also breast cancer and colon cancer, and other tumors have begun to emerge. Prognostic markers on solid tumors have received a lot of press attention, but they’re still a very small volume of molecular diagnostics, and most of the growth ahead will come in the sequential testing of blood samples.” He predicts a huge expansion soon of minimal residual disease testing.
Because of the high complexity of molecular diagnostics, the laboratory had previously used a partially manual workflow with a lot of paper superimposed on the underlying database, Dr. Jones says. The relatively small volume of molecular diagnostics testing until recently meant “it really didn’t pay to spend a lot of time and money developing an interface with your LIS.” That’s the chief reason it has taken specialty software so long to emerge—the combination of a complex workload and a low- to mid-range testing volume in molecular diagnostics.
When MD Anderson started working with Soft Computer, as co-developer of the molecular diagnostics module, “we recognized we weren’t going to be able to build something static. It had to be more flexible to build the tests in real time. So I think the influence of Web services for data transactions and Windows-style interfaces that are more user-friendly and customizable led to development of software with a more open architecture.”
This customizability is essential, Dr. Jones says, given the complex workflow in molecular diagnostics. “You not only have multiple products to deal with—RNA, DNA, proteins, and cells for FISH and cytogenetics,” says Dr. Jones, “but there’s also the fact that complex molecular tests need to be repeated more often, so you have to build in the ability to go back at any step in the process, to re-extract, retest, or re-analyze. That happens much more frequently for molecular testing than in high-volume chemistry testing.”
MD Anderson’s laboratory also spends a lot of time correcting and clarifying orders when they come in, “either because our previous results on that patient in the lab indicate that we should be doing a different test, or because the ordering clinician was not sure which test or tests to do.”
But the most complex part of molecular testing is on the reporting side, Dr. Jones explains, because there are many different platforms for DNA sequencing and PCR, making automated extraction of laboratory results difficult. “The approach Soft has taken, which we’ve also taken internally in our own software design, is to map the data that’s coming out of the instruments onto LIS-produced worksheets. This allows technicians to ‘drag and drop’ data elements from the instrument files into the proper location on the worksheets.” This feature of Soft’s modular architecture allows mapping of new data elements without having to reprogram, he says. “This is essential because there are so many different analysis instruments, and every time the software is changed, it changes the structure of the output file.”
“We’ve taken that approach to building tests and building worksheets as well, because we don’t want to be stuck with a system where two weeks after you go live and you discover a better way to do it, there is nothing you can do to redesign.”
MD Anderson has been somewhat in the vanguard of efforts to integrate specialty molecular diagnostics reporting into the LIS, he notes. “A lot of people ask, Why can’t we just skip the LIS and report from a middleware or homebrew laboratory database right into the hospital information system? I think there are lots of reasons why you don’t want to do that. Having a fully functional and flexible LIS allows you to integrate results from other laboratories and present more structured and coherent results to clinicians.”
Regardless of the LIS design, reporting across systems has increasingly relied on Web services for interfacing, he notes. “So as long as your LIS supports that approach, which most new-generation LISs do, it is a better way to approach reporting. The laboratory basically controls which data it provides and the receiving system then structures the results according to its own schema.” Cerner’s product uses some of these ideas as well, with a lot of emphasis on the structuring and coding of molecular results.
Dr. Jones would also like to see more movement toward open formats for instrument output files. “Every large-sized lab has developed homebrew solutions to exporting and transferring data essentially on their own. It’s a very scattered area, but we’re pushing all machine vendors to develop more flexible export functionality. This allows laboratories to quickly write macros or use Web services to pick up the data in an automated fashion and populate their LIS or databases.”
When instrument data are hidden or difficult to extract, it is a real problem, he says. “A lot of hacking of instrument data files has occurred in the past in flow cytometry and molecular diagnostics. Increasingly the manufacturers of these instruments are recognizing they have to make their output files more readable.”
Laboratory IT is behind the curve in another area: interfaces between the instrumentation and the LIS. “Most of the instrumentation that we use for molecular diagnostics is actually not able to communicate with the LIS,” Dr. Monzon notes. While this was probably true for chemistry analyzers, too, when they appeared on the scene, in molecular it’s because most of the instruments were designed with research in mind.
“As we move them forward to clinical use, some fall short of expectations. There is even some instrumentation we use for clinical molecular testing that manufacturers actually recommend you don’t even connect to a network” because of compatibility issues. “So sometimes you end up having your instrument running with a computer that is off the network,” Dr. Monzon says. “Then you have to print the data, or copy it to a CD, transfer it to another computer, load it into your homegrown database or the LIS if it’s capable of handling the information.”
But manufacturers may have awakened to the growth prospects for molecular. “For example, companies initially focused on reagents and instrumentation for research are now becoming major players in molecular assays,” Dr. Monzon says, “because they see the tremendous growth potential there. Companies like Roche keep developing analytical platforms that improve automation and are trying to keep up with new advances in molecular.”
However, information management is not a top priority, Dr. Monzon believes. “It’s reaching the point where it is going to have an impact on our ability to do testing, because if I have to handle high-dimensional data such as that based on array platforms, that will put a significant burden on us. While some of the instrument vendors have their own software under development, it doesn’t necessarily communicate with your LIS or with the other information systems out there.”
Based on Dr. Monzon’s experience at Methodist and at the University of Pittsburgh, where he was associate director of specialty laboratory informatics, he concludes: “We’re all struggling.” Unfortunately, while the molecular-oriented modules are becoming available, the solutions are not turnkey. “It’s not something where you open the box and put it in the computer and start using it. There’s a lot of development to be done because these systems need to be custom-tailored to the laboratory and the institution.”
“Many of us have had to deal with these information management issues with the tools we had, with the LISs we have, and we’ve gotten accustomed to fitting the current LIS into our workflow and just making it work,” Dr. Monzon says. But it would be better to develop more adequate systems that actually adapt to the molecular lab workflow.
In the process, the instrument and software manufacturers need to talk to each other, he says. “We really want to get everybody at the table to communicate what our needs are, what rules we need to define for the communication exchanges in molecular diagnostics, and how we are going to achieve that together.”
Anne Paxton is a writer in Seattle.