Helping clinicians cope with CPOE systems

November 2002
Eric Skjei

Laboratory information system vendors know well the patient safety benefits of computerization, and they’re laboring to bring the insights of the LIS world to the creation of CPOE, or computerized physician order entry.

"Electronic ordering allows fewer errors to occur, not only in the patient data—for example, getting the name wrong, but also in the lab data—for example, choosing the wrong test for a patient," says Curt Johnson, vice president of sales at Orchard Software.

Mention patient safety and physician ordering to most LIS vendors and you will almost certainly hear the word "alerts" in their replies.

"The givens in CPOE are an electronic ordering system, alerts and rules, and a clinical database to fire them against," says Bernard Kasten, MD, vice president for medical affairs at MedPlus, a Quest Diagnostics company.

When it comes to clinical alerts, the rules or algorithms can become "as complex as the organization would like," says Floyd Eisenberg, MD, MPH, physician executive for clinical systems at Siemens Medical Solutions Health Services. "Our rules engine can extract data from various locations, perform the appropriate calculations, and notify the physician synchronously during the ordering process or asynchronously at the occurrence of new events outside an existing order session," Dr. Eisenberg says.

Not long after the Institute of Medicine report "To err is human" set off a firestorm of concern in 1999, more than 100 organizations joined under the rubric of the Leap Frog Group for Patient Safety to help organize an industrywide system of defining and certifying solutions to the problem of avoidable medical errors. Like the IOM, the Leap Frog Group named CPOE as a critical component of such solutions and began specifying and certifying the components of a CPOE system. The coalition determined that CPOE systems have two main features: they encourage physicians to order electronically rather than through handwritten or oral instructions, and they provide a high level of information about patients by integrating clinical information from a variety of sources, most notably the pharmacy, laboratory, and radiology. CPOE, according to the Leap Frog Group, falls within a larger function known as clinical decision support and should be distinguished from the more general function of simple orders communication.

As of August 2001, reports the Leap Frog Group, about 13 CPOE products were on the market or slated for imminent release. The group named several vendors supplying CPOE systems at that time: Cerner, Clinicomp, Eclipsys, IDX, Meditech, Per Se, Siemens, VisualMed, 3M, Autros/ Epic, McKesson, and GE Medical Systems.

The art of the alert

Neither alerts nor the technology underlying them are new. The ability to generate alerts, especially those based on the operation of a rules engine and a clinical database, is a proven technology in the LIS world. Nonetheless, the technology, like most, is still a work in progress and is in many ways as much art as science.

In its fullest expression, the application of rules technology seeks to completely replicate clinical decision-making—an ambitious and often impractical undertaking. Rob Bush, president of Orchard Software, recalls a project in which he worked to implement rules to alert technologists about results requiring a second look before being released. "We created a system that would electronically consider everything a technologist would consider in making this decision," he says. That meant considering a lot of variables—which physician ordered the test, where the patient was located in the hospital, medical history, patient age and gender, and lab-oriented data, such as previous results, instrument flags, and delta checks.

"Our contention was that if it was something the technologist used to screen results, and we couldn’t replicate it in the rules system, we weren’t offering a real alternative solution," Bush says. "If you can’t do that, you’re not really offering an alternative; you are, in effect, asking your client to be ready to do the same thing twice, if need be, once in an automated way and, if that isn’t completely reliable, then once again by human beings."

Extend this experience to CPOE, he continues, and you may find that, beyond the simplest level of alerts, you face a problem of rapidly diminishing returns—of pouring more and more time and energy into work that pays you fewer and fewer real gains—and that the challenge of creating an electronic duplicate of the human decision-making process is nearly insurmountable.

Fortunately, the alerts required to help physicians make safer orders do not require so comprehensive an application of rules technology. In fact, many flags, notifications, and warnings do not require a rules system in the purest sense at all. Moreover, for alerts of any kind, the more likely question is not whether they can duplicate human judgment, but whether they will be so numerous and self-evident as to annoy rather than assist.

"If every little trivial thing rings a four-star alarm, the doctor is soon going to lose interest in any of the rules," Dr. Kasten says. In fact, he argues, this is exactly what has happened in many rules-based alert systems: Physicians have chosen to opt out of them, to run them off or de-subscribe to them, because they have been seen as cumbersome, impractical, and "riddled with false-positives," Dr. Kasten says. Adds Dr. Eisenberg of Siemens, "My concern is that the system can provide many alerts, but if the number and/or quality is more an annoyance than a benefit, clinicians of any type will quickly learn to breeze right by them." Alerts, he says, must either provide a clinician with significant, timely information that poses a real, immediate risk to a patient or request information the clinician perceives to be important in the clinical care process.

The "nuisance potential" of alerts has been acknowledged as a serious issue by the Leap Frog Group. In a December 2001 report, the group said that "a hospital must achieve the right balance between useful alerting and ‘over-alerting’ or intercepting orders that have a very low risk of ADEs [adverse drug events]. Such ‘nuisance’ alerts can seriously impair physician acceptance of CPOE and, more important, lead users to ignore all alerts, thereby decreasing the value of clinical decision support." (Kilbridge, et al. Overview of the Leap Frog Group evaluation tool for computerized physician order entry. Leap Frog Group [in conjunction with First Consulting], December 2001, page 4.)

Alerts can also have subtle political and legal implications. Their tone must strike the right balance between obsequious and directive.

"It’s very important that rules be presented in a way that is not pedestrian or demeaning," says Dr. Kasten. "A cardiologist doesn’t want to be told something he or she already knows."

To some extent, this is probably a matter of simple professional courtesy, but it may also touch on the more serious issue of liability, Dr. Kasten says. One of the open-ended questions implicit in the deployment of a rules system, he suggests, is whether the providers of rules systems incur any medical liability. "I would hope they don’t, because they are always offered as suggestions to augment physician judgment," he says. "Their intention is to assist the physician and guide a possible course of action, not pre-empt physician judgment."

For LIS vendors and others, managing this problem is straightforward. It involves working at the outset of a new installation to make sure alerts are deployed to address problems that are neither trivial nor overambitious. "What we do is go in and sit down with the relevant people at the site and say, ’Let’s assess your problem, where do you want to focus?’" reports Scott Mattingly, product line director for enterprisewide solutions for Misys Healthcare Systems. Mattingly is describing the implementation of the Misys Insight clinical decision support system. The process begins with consultants who spend time on-site interviewing staff to determine the functions most in need of clinical decision support, those in which appropriate alerts can be quickly leveraged to return the greatest immediate benefit.

"We don’t try to draw on all of our 500-plus rules and roll this out to the entire enterprise up front," Mattingly says. "Instead, we try to select the top five problems, create the handful of alerts that will really help those wrestling with these problems, make sure the alerts work and that the site will get the reduction in errors it is looking for, then return later and begin extending the use of alerts beyond that initial target group."

By their very nature, many alerts must be presented at the time of ordering to have as much impact as possible. Fulfilling this function efficiently depends largely on the speed of the underlying hardware and software, which in turn touches on a longstanding topic in health care—that is, the relative performance merits of single-platform, homogeneous, highly integrated solutions versus the more highly interfaced, heterogeneous solutions (those supporting so-called best-of-breed systems). With respect to CPOE development, vendors with highly integrated solutions argue that they offer a significant performance advantage in ensuring that alerts are presented as immediately as possible to the ordering physician.

"If I’m on a single data model platform and the lab value is processed and available to me, as a physician, immediately, I’m getting real-time access to that value," says Bryan Christianson, product marketing manager of physician solutions marketing, Cerner Corp. To put it more metaphorically, says Cerner’s Julie Hull, director of product management, "the difference between alert speed of access in a system like ours versus others can be like the difference between driving a car with warning signals that sound off or light up before you hit a pedestrian and signals that warn you as you hit a pedestrian." In contrast, systems that present an alert only after the order has been sent and a resource such as a departmental database has been accessed may be at a significant performance disadvantage.

"I agree with that perspective," Dr. Kasten says. "If you have to crawl back into a legacy system with no real-time data access, your CPOE system won’t meet the needs of busy physicians." However, Dr. Kasten adds, the potential performance disadvantages inherent in heterogeneous, highly interfaced systems can be mitigated by creating a clinical data repository against which the ordering rules fire. Citing his own experience at MedPlus with ChartMaxx, designed for hospital clients, and eMaxx, designed for physician and clinic use, Dr. Kasten explains that since these solutions incorporate clinical data repositories, "users have instant access to their data—they’re not trying to go back to a legacy database on a real-time basis."

Of course, the use of rules, clinical databases, and alerts to shore up patient safety through CPOE systems implies a process for making sure the rules and the data are as current as possible, which is in itself no small undertaking. Well-trained, experienced, and highly informed though they may be, clinicians cannot stay abreast of all medical research and its clinical implications. "Some of what I learned in med school just isn’t true anymore; it’s been supplanted by more recent data," says Dr. Kasten. "But rules should be based on evidence, and that evidence must be kept current, because yesterday’s medical truth is not today’s medical truth."

Managing multiple workflows

The patient safety challenge presents another longstanding question: How can physicians best be encouraged to use CPOE systems? Understandably, physicians have responded to health care computerization in a decidedly mixed way.

It is still easier and faster to tick off boxes on a form, scrawl an order in a chart, or tell a staff member to take action, than to log on to a computer system—no matter how fast it is—call up the appropriate screen, examine the information presented, and respond.

"I saw a presentation many years ago on this topic in which the individual making it began by projecting an image of a No. 2 pencil and saying, ’Gentlemen, this is our competition. We may be able to beat it in our lifetime,’" recalls Bill Blair, vice president of sales and marketing for SIA. True, says Blair, the earliest electronic ordering systems were focused more on billing than on clinical data and were designed for use by clerical staff, not clinicians. That has changed, of course, and many ordering systems now in use are designed to present clinical, patient-centric information directly to physicians and other frontline caregivers.

Nonetheless, some doubt this challenge can be accomplished. "Doctors will not tolerate the rigors that a really sophisticated order-entry system must impose on them," says Janet Chennault, vice president of Schuyler House Software. And this physician impatience is only exacerbated by a health care system that seeks to impose tight constraints on the time doctors spend with patients. "The sad state of medicine right now is that your typical physician encounter involves less and less encounter," says Bill Shipley, president of Schuyler House. "It’s probably not a wise idea to waste any of that limited time risking having the physician spend it cursing at the computer."

The problem may lie in the fact that ordering systems have not incorporated sufficient intelligence about the intricacy of ordering. Order entry often takes place within complex systems comprising many people and thus many overlapping, intersecting, and interdependent workflows. Next-generation CPOE systems are likely to include tools that reflect this fact more fully.

"Dealing with physician order entry requires managing many workflows," says Siemens’ Dr. Eisenberg. "It’s not just a matter of encouraging physicians to get online and put orders in an electronic system." A doctor writing an order is more often than not writing an order for other orders, taking an action that will affect many workflows. "For instance," Dr. Eisenberg says, "if I write an order for a lab study, I may just write that I want a certain drug level done in the morning, say half an hour before the next dose of the drug is administered." That brief note may translate into placing the order on a list of scheduled draws, timing the draw based on when the medication is to be given, performing quality checks to make sure the drug was not administered before the draw, conducting lab tests, and reviewing and transmitting results. Some of the tasks involved can be handled concurrently; others must be handled sequentially. Some of the people involved need to know only about their own steps; others need to have a comprehensive overview of the process.

If it lives up to its promise, an automated workflow manager should be able to help track these intersecting tasks, compiling information about their completion and moving the entire array of interdependencies through the patient care process with a minimum of errors. Such efforts are underway in many health care IT companies, including Siemens, which is incorporating a workflow engine into its next-generation clinical access system, Soarian, scheduled for release early next year.

Real benefits, slowly realized

CPOE implementation is still in its infancy. The Leap Frog Group estimates that only two percent of health care providers have deployed a CPOE system. CPOE prevalence will grow—but probably slowly. According to several researchers, fully realizing CPOE’s benefits is likely to take a decade or more.

Though CPOE is still new, its earliest implementations are already reducing medical error. Given the emphasis by the IOM, the Leap Frog Group, and others on medication errors, much of the data thus far focuses on avoiding adverse drug reactions and shortening medication delivery turnaround time. "A number of our CPOE customers are already reporting good results," says Dr. Eisenberg. "At the Ohio State University Medical Center, where 100 percent of physician orders are entered electronically, they have seen a 25 percent reduction in turnaround from the time a lab order is placed until the result is back from the lab, a 43 percent reduction in radiology turnaround time, and a 64 percent reduction in medication turnaround time."

Similarly, says Dr. Eisenberg, Siemens’ CPOE systems at Rush Presbyterian-St. Luke’s in Chicago have facilitated a 64 percent reduction in medication turnaround time (the time between when a medication is ordered and the patient receives it), and Meridian Health Systems in New Jersey has seen an 84 percent reduction in medication TAT. In addition, says Dr. Eisenberg, these sites report significant cost reductions derived from reduced lengths of stay and fewer unnecessary or duplicate laboratory, medication, and radiology orders.

Similarly, says Misys’ Mattingly, Palmetto Health Richland in Columbia, SC, among other benefits, has reaped $91,000 in reduced medication costs and an estimated $440,000 per year in total cost savings for all interventions based on alerts from its first year of Misys Insight clinical decision support system use.

While patient safety is the overriding rationale for developing and adopting CPOE systems, the systems provide a compelling opportunity to integrate additional information. For example, information on patient sensitivity to specific medications can easily be integrated with recommendations for alternative medication choices, including efficacy data and cost data.

The process might work something like this: "The physician could get a lab result back that says this patient has E. coli, along with a list of drugs that have proven to have varying degrees of effectiveness in dealing with this variant of this bug," says Will Campbell, senior product manager for Impac Medical Systems. Too often, Campbell says, the physician’s reaction to this information and his or her subsequent order is likely to be based on inadequate information. The physician will look at such a list and choose a medication based largely on past experience. "That’s not enough information," Campbell adds. "If all you present to the doctor is a list of four drugs of varying efficacy, you haven’t really given the physician as much information as you could and probably should."

The physician in this case should, Campbell argues, be presented with lab results, medication choices, the efficacy of those choices, and their cost. "He or she is then in a position to make an intelligent ordering choice," Campbell says. "He or she can say, well, Choice A looks like it’s the cheapest and it has good efficacy, but I know it will be incompatible with other things I am prescribing for this patient, so this next choice, which looks like it works just as well and is about the same cost, will do the job nicely."

Electronic ordering can, in short, leverage much more information during the ordering process, and in so doing contribute additional value while fulfilling its primary goal—patient safety.

Eric Skjei is a writer in Stinson Beach, Calif.