College of American Pathologists
Printable Version

  Credit clever middleware for spike in
  labs’ productivity






June 2007
Feature Story

Anne Paxton

Okay, it’s true: Laboratories can’t make a complete jump to light speed by installing middleware. But mid-size hospital laboratories and reference laboratories are reporting that middleware—a catch-all term for rules engines or expert systems residing between the diagnostic instruments and the laboratory information system—has let them shift into a much higher gear, making impressive productivity gains in the process.

At Meriter Laboratories, a mid-size laboratory on the campus of a 448-bed hospital in Madison, Wis., middleware helps compensate for the shortcomings of a “very aged” laboratory information system, says Becky Collar, MT(ASCP), section head of the core laboratory. The aged LIS is due to be replaced soon by a new clinical laboratory system (part of the Epic system the entire hospital will be using). But in the meantime, middleware has given the old LIS a shot in the arm.

In fact, starting in 2002 all of the analyzers in the core laboratory became hooked up through Data Innovations middleware, beginning with the Roche standalone general chemistry analyzer, the Cobas Integra 800, and including the Ortho Diagnostics ECIs, Radiometer 735 gas analyzers, Beckman Coulter hematology analyzers, Stago STA coagulation analyzers, and Iris Diagnostics urinalysis system. “The middleware does autovalidating for all of them,” Collar says.

When the new LIS is installed, she says, “we are planning on enhancing the rules and comments we can use with our Data Innovations. Right now we can put rules in our middleware‚ but we’re still held to what our aging LIS can do, and that’s limited. It won’t accept add-on orders, reflex testing, or comments from DI right now.”

“We can tell the techs what to do with a specimen, but because of the age of the system, they have to go into the LIS and put their own comments in. It also will not transmit cancels to DI, or results with more than nine digits past the decimal point. We’re hoping the new LIS will allow more of that to cross over from the middleware.”

When the middleware came in, the laboratory tried to ease the impact on staff. “It’s hard to get used to not looking at results that are going through automatically. We began with really tight controls/rules, so for anything high or low, we checked gaps, we checked different calculations, and we held a lot of results to begin with. And two or three weeks after we went live, the techs were coming to us and wanted the rules to be a little looser. So now, if the interfaces go down, they keep wondering where they would be if they had to go back to re-releasing all those results.” The laboratory’s chemistries are now 90 to 95 percent autovalidated.

Test volume increased 12 percent at Meriter Laboratories from 2002 to 2005, but staff size decreased 13.5 percent through natural attrition. “We would not be able to do the number of tests we do with the number of staff we have and still be able to offer the high quality our customers expect without middleware,” Collar says.

She appreciates, too, that the DI middleware standardizes the use of the analyzers. “It doesn’t vary depending on who’s working that day, whether they’re experienced or new techs, or off shift. With DI we’re applying the same rules to everything; it’s not the individual technologist making the judgment.”

Rule writing is somewhat of a complicated art, however. “Some of the analyzers have capabilities of having their own rules—for example, if something is out of range or high and low. But with middleware rules you can set the instrument analyzers to stop those from going across. On top of that we have some rules built into the old LIS, like forced releasing if there’s a critical value.” So it adds up to a three-stage rule-building process for autoverification.

The rules available with the middleware are flexible and can be tailored to specific clients or specific floors, Collar explains. For example: “If you wanted to apply different BUN and creatinine rules to dialysis by source facility, you can do that, and maybe you would choose not to hold ones that, if performed on the normal population, would be of concern.” If the creatinine and BUN don’t match—for example, creatinine is high and BUN more normal—then the middleware will hold the result on the screen and tell the technologist to check, because normally if one is high then the other one is high.

If a test like protein or albumin is critically low, “we’d hold the whole sample and check it for integrity to make sure it’s not diluted out or there’s some interfering substance floating in it.” So some rules will pick up any one of numerous results and, in certain cases, will hold all the results.

Another rule links results with the patient order. For the result to proceed, “the order has to be received and waiting for results—if not, then everything holds on the screen. If the order is received and waiting in the LIS, the test will go right through.

Bilirubin is another test that can prompt varying responses. “We test total and direct, and if there’s a sample with very little of both, sometimes on rare occasions you’ll get a higher direct than total, so DI is going to hold and say to check the direct bilirubin. Then the technologist can repeat it.”

One client requested that every time a high analyte was reported, it would be repeated and sent to another source, and that a lipase enzyme test would be added. “We built a rule that said if the analyte is high, attach a test comment. Then the comment didn’t go anywhere but it showed up on our screen and said ‘Add lipase.’ It was specific for just that one client, and it only fired if the analyte was high.”

Even when the new LIS is implemented, Meriter will still use middleware, Collar says. In fact, Collar is hoping it helps lessen the impact of moving to the new system. “The new LIS will change test building and results fields, which the implementation team will deal with,” she says. She expects the technologists who primarily work with the middleware won’t have as large of an adjustment to make.

When the laboratory at Bon Secours Health System, Richmond, Va., first installed its Data Innovations Instrument Manager middleware last November, the major draw was not the rules engine but the quality control module, says Bon Secours systems analyst Jan Dreelin, BSMT(ASCP).

“When you program the QC to run on your instruments, the middleware captures the results based on the name of the QC you assign, and drops them directly into the QC program, so there’s no manual intervention. The technologists don’t have to plot it on a chart or remember to send it anywhere, and they don’t erroneously delete any values. It gives a really clear picture of all your QC and it does it automatically.”

At the moment, the four-hospital system has the DI middleware hooked up to all nine chemistry analyzers, which are Dade instruments. But the middleware will work with any analyzer. “DI has interfaces or drivers to every instrument out there, and a hookup can be made for about a $1,500 one-time fee per instrument.”

The laboratory is happy with the rules included in its LIS, a Misys system that already conducts autoverification and other rules, and doesn’t feel the need to use middleware for those purposes. But that is largely because of one expert the laboratory has on staff, “a really good rule writer,” Dreelin says. “We don’t feel any pressure to change right now, but we have that functionality with our middleware; we have all the modules DI makes and could do it in the future if we wanted to.”

The middleware’s specimen storage and retrieval model works well at Bon Secours, which uses a robotics line to process specimens, she says. “We’ve had a few other systems, and this one has some interesting features in that when you scan a test with a bar code, it assigns it a spot and then you can recall it. If the tube has any tests still pending, then it flags them; you can still sort the specimen but it lets you know tests are pending.”

The downside is if a test is pending on a Bayer instrument, the system always holds the specimen as “pending” because the Bayer instruments aren’t yet hooked up to the middleware. “But that’s a problem within our control, not a problem with the system,” Dreelin says. Overall, for a relatively inexpensive price—a one-time cost plus a maintenance agreement—she says the middleware has given her laboratory valuable help.

The Southern Arizona Veterans Affairs Health Care System, Tucson, made the move to middleware in 2002 with a Dawning Technologies system. The initial version, Results Net Plus, was not much more than passive data transfer, says Evelyn D. Harrison, MPH, MT(ASCP)SM, laboratory information systems manager, but when the next generation, JResultNet, came out, the laboratory was able to start implementing middleware rules.

The rules were fairly simple at first—just providing that analytical values outside of defined ranges would not be reported. “We had that capability already in Vista” (the open source medical record application funded by the federal government). “But Vista required programmer access, which is delegated only to a few individuals. With the middleware, we didn’t have to make a special call to the informatics department to use it.”

The addition of comments is one feature employed frequently at the Southern Arizona VA. “We used to have all these little acronyms that the technologists would enter and they would expand out, but the technologists would have to remember to actually enter those comments and sometimes they would forget,” Harrison says. “The interface now handles these automatically.”

For example, if a troponin test is below a certain value, an interpretation of that value will automatically be used in the comments. For drugs-of-abuse testing, where the result is only used as a screen and not as a quantitative value, the middleware system adds a comment that the results are not to be used for legal purposes.

Reflex testing capability has also proved helpful. For example, on an LDL test, “if the patient is fasting, and the triglycerides are less than 400, then the system knows to do a calculated LDL. If the patient is not fasting or the triglyceride result is greater than 400, then we actually have to perform a direct LDL test; the middleware rules will automatically order the direct LDL test and send the request over to the instrument.”

That decision point removes the human factor, she says. But if Harrison could make an improvement, she would like to see a library of rules developed so users of DI or Dawning middleware don’t need to devise their own.

Most recently, the laboratory added specimen integrity rules. In the case of hemolyzed or icteric specimens, for example, that are identified by the chemistry analyzer, the middleware system prevents reporting some results outright if a significant amount of interfering substance is detected. “On some, it keeps the result but inserts a comment indicating that the result may be falsely elevated or decreased due to certain factors. These interference cutoff points for each test are very different, and I don’t see how humans could keep it in their heads, so having these specimen integrity markers has once again removed a source for human error.”

Vista is not currently capable of handling autoverification, so that middleware functionality hasn’t been employed. “Improvement to Vista would need to be made in order to get maximum usefulness out of the middleware, and that’s in the works,” she says.

Now in beta testing is a database query capability. “Not only does it allow us to back test rules, but if the beta works as planned, we’ll be able to much more quickly query patient laboratory results in real time. For example, we could check the last time the patient had a hepatitis B antibody test done. If it was done more than once, and on the previous test was positive, it may suggest unnecessary testing.” They’re also intending to use it to be more cost-effective, to try to prevent needless ordering of laboratory tests such as PSAs mistakenly ordered on women. From the demonstrations Dawning has done, Harrison says, “It looks extremely powerful.”

The Southern Arizona VA was the first system in the western hemisphere to install an Ortho-Clinical Diagnostics robotics system. “There were no immediate personnel savings, but as we worked the bugs out and time passed, we are now down to needing only two people to run our chemistry area, whereas we used to require four or five,” Harrison says.

But middleware was essential to making it work. With all the different instruments by different manufacturers that are connected to the robotics system, she compares the middleware to a highly sophisticated traffic cop. “There are thousands of rules that could come into play in our middleware system each time an order comes in. Requested testing will be routed to the primary instrument first. If necessary, the specimen might be sent back for retest or a secondary backup instrument or duplicate testing. A specimen may need to be routed to two different instruments entirely or routed to a drawer for off-line testing. Then, when all testing is completed, send the specimen to storage. It’s a massive undertaking. And you could not run that robotics system without middleware. There is absolutely no way.” Clinical Reference Laboratory, Lenexa, Kan., is a multilab facility. Systems administrator Brian C. Sudis says CRL started with a Dawning Technology appliance-based version of JResultNet. “It’s a really slick device, but it didn’t handle our throughput well,” Sudis says. Once CRL switched to a server-based platform of the same middleware, the recent 2.32 release, “it has done quite a nice job for us”—particularly in adjusting to the increased volume.

A feature that Sudis finds reassuring is the Dawning middleware’s socket-based “fail-over” capability. “By taking either a terminal or some other appliance or IP-capable instrument, you can configure the JResultNet instance to a client-type server, and have the appliance doing the IP translation be the server. So if the instrument fails to connect for any reason, it can automatically ‘fail-over’ to another instance of JResultNet.”

Sudis’ team is trying now to envision what the next generation of systems infrastructure will look like. “Speaking as the guy who will probably end up having to support it, I think a modular, middle-tier approach to an LIS is the most attractive option. You would have a middle tier that does rules-based translation, a middle-tier reporting module, a data-repository module that may be an Oracle database or DB2 or Sequel. So all these various plug-and-play pieces you surround the laboratory with would make up an ‘umbrella’ of devices or applications.”

Sudis does not think the JResultNet middleware will necessarily make his laboratory management system or the LIS obsolete. “JResultNet is a value-added feature for us,” he says. But as the rules engines continue to develop and the umbrella of middleware evolves, it could take on an even more central role in the laboratory—as a component that is “in the middle” no longer.

“I do believe, from what I understand of communications flow from the instruments to the LIS to the database to reporting mechanisms, that the LISs of the future will predominantly be groups of middle-tier pieces.”

Anne Paxton is a writer in Seattle.