Feature Story

Digital wonders dazzle in Pittsburgh

January 2000
Mark Uehling

Video conferencing to allow pathologists to talk to pediatric surgeons in a distant operating room? Ho-hum in Ann Arbor. Internet-based telepathology consultations on slides from a hospital in Palermo, Italy, in real-time? No big deal in Pittsburgh. Mt. Everest-sized data warehouses, originally developed for satellites and climate forecasting, being adapted for pathology? Old news in Baltimore.

So it went at the fourth annual Advancing Pathology Informatics, Imaging and the Internet (APIII) conference, held in Pittsburgh last October. Participants discussed eye-popping technology as if it were the norm. A vast buffet of the latest digital wonders for pathology—filmless cameras, image databases, Web servers, and more—was the topic of discussion for three days. But the spice in every dish was "when?" "When will my group be able to afford this?" "When will this technology be ready for prime time?" "When will my competitors adopt it?"

To hear the panelists in Pittsburgh tell it, the answers could be "right now" or "maybe never." But all agreed it is not too late to consider the digital revolution and what it may allow and how it may alter the practice of pathology. As Eric Schubert, MD, of Memorial Hospital, Chattanooga, Tenn., said during his chronicle of a still-incomplete attempt to handle thousands of images electronically: "Are you behind by not having it? No, not yet. But you will be. It’s catching on."

Like water in a hurricane, informatics and the Internet could seep into every nook and cranny of pathology. Which is not to say the attendees in Pittsburgh resigned themselves to changes beyond their control. To the contrary, there was a sense of anticipation that the seemingly unstoppable increases in the power of the integrated circuit could assist pathology—potentially more than any other medical specialty. The meeting’s keynote speaker, Steven McGeady, vice president and director, Internet Health Initiative, Intel Corp., explained that easily enough. McGeady outlined what he said was all too often the effect of the Internet on any industry—a diminution of power for established, centralized authorities and an accumulation of power by newly energized and informed individuals. The travel industry, financial industry, and book industry have weathered such changes, he said.

In medicine, McGeady commented, the days of putting a patient’s test results online are here. Pathologists, by controlling the majority of data central to a patient’s medical record, may have a historic chance to help patients manage the digital details of their medical destinies, he stated.

In other words, the right approach to technology may allow patients to rely on pathologists to a much greater degree than is the case today. A principal behind the APIII meeting, course director Michael J. Becich, MD, PhD, agreed. "Pathology, by its very nature, is disease-specific information, and that is the primary driver for patients to seek out information," he said.

"There’s only one product that I’m really concerned with in pathology," Dr. Becich added. "That’s our reports. Any stra-t---egy that focuses on improving value-added services has got to include the pathology report. Information systems are essential for this. If we don’t learn how to externalize our data in our information systems, we will be lost."

The computer systems of even the most sophisticated academic departments fall short of where they should be, warned Dr. Becich. In the future, he suggested, anatomic pathology reports will have to be united with clinical pathology reports; a hospital’s laboratory information system will have to be fully integrated with computers used by pathologists; and images will have to be incorporated into all of the above at the touch of a few keys.

Exhorted Dr. Becich: "We need to make the pathology report portable to the patient, easy to move around, and flexible and [able to interact] with all of these health care repository channels. Pathology should lead the charge. We can’t follow. We need to lead the charge." As an example, Dr. Becich noted that his wife, a neurologist, is being inundated with paperwork, much of it from pathologists.

"She’s got charts in the garage—charts on the kitchen table. And pathology, she says, is the major part of the problem," he related. "We send out the cumulative, the integrative, the stat, and yesterday’s report. It’s all the same data, but it’s on five pieces of paper. Thirty percent of the time, she’s making decisions based on tests that she ordered and that the patient paid for but that are not there when she needs the information."

Better reports, Dr. Becich conceded, have eluded even him, an associate professor and director of pathology informatics at the University of Pittsburgh Medical Center. "I’m perceived as a leader, and I have been awarded as an Internet Health Hero by Intel, but, quite frankly, I can’t do this on a production basis, and I’ve been talking about this for three years," he said.

More upbeat is the following statistic: A separate Web page for the university’s transplantation pathology division helped spark a 300 percent increase in consultations for UPMC physicians.

Even better, Dr. Becich forecasted, officials at the Health Care Financing Administration are beginning to see the wisdom of using digital technologies to extend the reach of the most experienced pathologists. "We are very close to having billing codes for telepathology consultation," he said. "There will be revenue opportunities here. You need to be a player. You need to have a digital strategy. If you have consulting opportunities you can leverage, there are also opportunities for growing your practice. If we build it appropriately, they will come, and our value will increase accordingly in the health care environment."

That already appears to be happening in Massachusetts. Conference speaker Michael Laposata, MD, PhD, director of clinical laboratories at Massachusetts General Hospital, Boston, walked the audience through a deceptively low-key but extremely versatile solution to his own curbside consult problem.

As Dr. Laposata related, he often found himself unable to leave the hospital without giving colleagues informal advice about a patient with a clotting problem.

To minimize the need for curbside consults, which he perceived as providing only a few "passing bullets" to the physician being consulted, he instituted two programs. One was an option to use a series of reflex test algorithms in coagulation to increase the efficiency of test selection and eliminate the need for the patient to return for blood draws on multiple occasions until a diagnosis is reached. The other was a daily interpretation rounds, as found in anatomic pathology, to provide physicians ordering complex batteries of coagulation tests with a narrative paragraph on the diagnosis, often after obtaining clinical information about the patient.

A survey of physicians at Massachusetts General Hospital receiving coagulation interpretations indicated that 71 percent ordered fewer laboratory tests as a result of an interpretation. In addition, 58 percent said the interpretation reduced the time to diagnosis, and 71 percent reported that it helped prevent a misdiagnosis. The overall approval rating for the narrative interpretations in coagulation was 98 percent.

Use of the thyroid-stimulating hormone test in one algorithm to limit further thyroid testing has cut the cost per test from $16 to $6.25 in Massachusetts General’s endocrinology laboratory, stated Dr. Laposata.

The algorithms and interpretations enabled Dr. Laposata and his colleagues to secure considerable amounts of new business from the surrounding area. The message for laboratories elsewhere, Dr. Laposata proposed, may be that decisions about whether to move hospital laboratories off-site will be made partly on the basis of whether a hospital laboratory and its pathologists can provide interpretive and consultative services.

APIII speaker Michael Gagnon offered a similar heartening story. Gagnon, director of business development and informatics at the University of Vermont, Department of Pathology, offered an account that is unique to his corner of the world and that may be of interest to anyone who uses or is considering using an ordinary Internet browser program, such as Netscape Navigator. The latter is Gagnon’s main information retrieval and delivery vehicle. (For physicians, the browser can be a powerful, inexpensive software tool that can rapidly present information, including images and tabular data or text, via hospital or desktop-based computers.)

As a veteran database expert, Gag-non appeared to think grandly (envisioning statewide disease-forecasting capabilities in future years) and conservatively (guarding against mission-creep or the addition of nice but non-es-sent-ial elements). The origins of his system in Vermont were humble, but they worked dependably. Some of the early stages of development proved the overall concept was not a boondoggle. "We built an online test catalog, a telephone directory online, the ability to record cumulative reports for one of our big physician offices," he explained.

Gagnon noted that one traditional difficulty with database design has involved making a wide variety of hardware and software work together. This systemic issue forced him to implement a robust, simple design. "We put more weight on system interoperability than features and functions," he noted. "System interoperability may be far more important in the long term than the actual features and functions you get. I’ve preached this for a long time."

Gagnon used garden-variety software, eliminating licensing fees. He also used a simple Internet computer code, HTML. As Gagnon explained, "We wanted low-bandwidth solutions; our state is pretty rural. No intricate things going down that browser, just pretty much plain-vanilla HTML. Most customers do have PCs. If they don’t, we provide them. It’s that important that we have good data coming in to serve that customer base."

In other words, Gagnon could not assume that his users had new or fancy modems or access to more futuristic devices for cruising the information superhighway. Partly because of that limitation, his programmers could not get too detailed. They restricted themselves to bare-bones HTML, eschewing more powerful but less universal alternatives.

But the Spartan system in Vermont includes several features that should save the hospital considerable amounts of money. For example, as Gagnon explained, samples are bar coded in physicians’ offices, reducing the potential for data entry errors later. Next, the system automatically detects which billing codes go with which types of tests. The system can spot inappropriate combinations of tests and codes or those that were not reimbursed.

Noted Gagnon: "You can find all the tests where the letters ’G-L-U-C’ appear in the test. Obviously this is a nice standalone, but it would be a whole lot nicer if you embedded it into the order entry, and you did the order right then. I can add tests and diagnoses, and I can ask the system to check the compliance. And it will tell me, ’Oh, glucose is not covered under this diagnosis.’ And if I wanted to know what does cover glucose, I can press a button and it shows me." Such assistance from his system, Gagnon predicted, will save his facility $250,000 annually.

Gagnon also described how he has made his system up-to-the-minute and easy for busy physicians to master. The Web site contains a page that shows users how to change the paper on their printers. As he explained, "It reduces the number of phone calls I have to take."

While the motivating factor for the project at the University of Vermont has been mostly economic—Gagnon conceded he hopes to discourage competitors—the project has a philosophical foundation. "It’s really important that we get ourselves out in front of patients to differentiate ourselves," Gagnon said, referring to pathologists. He cited statistics about the use of alternative health care and the tens of billions of dollars paid out-of-pocket by patients dissatisfied with traditional medicine. "It’s clear people are choosing alternatives and are willing to pay for it," he stated. "We have this ’Intel inside’ problem in the labs. We need to get ourselves out there. We need to establish ourselves."

Although much of the Pitts- burgh meeting was practical in orientation, there was a sense that digital technologies could have clinical importance beyond the next software release from Microsoft.

The presentation by Stephen S. Raab, MD, PhD, of Allegheny University Hospitals, Pittsburgh, was a case in point. Dr. Raab’s background in cytology has not deterred him from becoming knowledgeable about outcomes research—and that informatics topic, in turn, bears directly on how pathologists use technology.

"If you come out with a diagnosis of cancer," Dr. Raab said, "it doesn’t mean as much to a clinician as using the diagnosis of cancer, calculating the likelihood ratio, predicting the probability of disease in the patient given that diagnosis. Likelihood ratios—even though we don’t do them in pathology—have been advocated for a long time in clinical medicine."

But Dr. Raab added that the field of outcomes research has produced an unequivocal but not especially surprising finding, which is that pathologists often disagree about the interpretation of a particular slide or section. Understanding how to minimize interobserver variability, he said, could help pathologists anticipate the needs of clinicians and patients, not be blindsided by them.

As an example, Dr. Raab cited the independent evaluation of how pathologists assess bronchial brush specimens—with and without the benefit of knowing the patient’s history. Data clearly show more accurate diagnoses result when pathologists know a patient’s history. Added Dr. Raab, "Even though we don’t interpret diagnoses like this, there have been advocates in the clinical literature that we should be interpreting results like this, because this is what affects patient care."

While the relevance of outcomes analysis is just now being determined, it is evident that the pathologist’s report could stand to be re--examined. "If you look at the literature," said Dr. Raab, "you can see that there has been poor agreement among pathologists on the information product of what we’re sending out. But there hasn’t been much [research] done on what does this mean for patient care, or what does this mean for pathology practices?"

Equally far-reaching was the lecture by James H. Harrison Jr., MD, PhD, associate director of pathology informatics at the University of Pittsburgh Medical Center. Dr. Harrison began by delving into the terminology of data warehousing and data mining. He defined the latter as the extraction of knowledge from data. Said Dr. Harrison, "It may be more useful to define it operationally as the detection and reporting of meaningful data patterns in large integrated data sets."

He also introduced a less conventional term that he hoped would gain wider acceptance, the "data-opsy," modeled on "biopsy." Dr. Harrison defined data-opsy as "the extraction and analysis of a sample of data from the medical record to identify clinically important data patterns."

In Dr. Harrison’s view, a data set could be: text, images, or statistics. And he is not imagining merely asking the computer to find rudimentary information, such as all patients with a particular condition. Rather, he is actively exploring the use of computers in ways that begin to approximate human problem-solving. Neural networks, he pointed out, allow physicians to tell a computer in general terms what they are interested in and let the computer use its methodical, tirelessly mechanical processes to find more of the same.

In Pittsburgh, Dr. Harrison related, such techniques are already helping the educational aspect of his program. "We were able to find rules that screened out patients with drug-level profiles that suggested the need for followup," he explained. "We let the students go out and evaluate the cases. It was a great way to match up students with cases that were appropriate for what they were studying at the time. Few of those cases would have come to the students in training if we hadn’t been able to look through large volumes of data to find them."

Dr. Harrison also noted what is on the cutting edge, what might work down the road. Today, he said, the computer might need hours or even overnight to return an answer or a set of patients. But in the future, if the data sets are configured and assembled properly, it may be able to provide virtually instantaneous responses. Such tools are not yet widely available in clinical settings, he added. But they will be. "You can explore the data by submitting queries that are designed to return populations of patients, then, based on analysis of that data, you can immediately submit additional queries to focus on the populations of greatest interest," he explained.

So what might these data warehouses do? Predict clinically significant events. Shed light on the effects of actual clinical practice over time. Even forecast which financial and medical resources will be required for a particular patient. That all sounds promising, but Dr. Harrison closed his remarks on a more subdued note. "If pathologists do not take ownership of this area," he said, "the management of this data and the resources required to produce it will pass out of the control of pathology."

Mark Uehling is a freelance writer in Chicago.