Applications: LAS in Canadian Medical Informatics

Canadian Medical Informatics, May/June 1995, Volume 2, No. 3.

An interactive expert system for the ordering and interpretation of laboratory tests to enhance diagnosis and control utilization.

Introduction

Expert Systems have a potentially powerful, but little recognized role in changing the manner in which clinical laboratory services can be provided. The introduction of such systems will require a paradigm shift in the way laboratory tests are ordered and interpreted. Such a shift is in keeping with the rapid changes which are occurring in health care and offers a way to address the two main concerns which have been identified with laboratory investigations. These concerns are:

utilization - controlling costs, and
ordering and interpretation - developing a means to ensure the appropriate tests are ordered and interpreted properly.

Utilization

Over the past decade the cost of laboratory testing has increased at a faster rate than most other medical services. Although the unit cost of performing each test has declined relative to inflation, the number of tests ordered has increased significantly. Various initiatives have been introduced to control the increase in "utilization" and these include: feedback, participation, education, cost awareness, financial incentives, penalties or risk sharing, administrative change, and rationing [1, 2, 3].

As demonstrated in several provinces, ordering protocols - i.e., generally accepted and officially sanctioned test ordering sequences - have proven to be a very successful approach. For example, in British Columbia a thyroid testing protocol, introduced in 1984, and modified in 1990 has reduced the expenditure on thyroid function investigations by two million dollars per year on an ongoing basis [4]. Unfortunately, to have an effect on costs and to be workable, such protocols:

must address high volume ordering areas,
must be amenable to a few, simple rules which can be easily remembered by clinicians, conveniently expressed in the test order, and easily carried out by the laboratory, and
require general agreement amongst clinicians, laboratories, and payment agencies. This is a bureaucratic process which often takes longer than the introduction of new techniques and diagnostic approaches. It goes without saying that such protocols must also ensure that medical care is not compromised through their use.

Thus, while ordering protocols offer a powerful, clinically appropriate approach to utilization control, the number of areas where they can be implemented successfully has been limited because of practical constraints.

Ordering and Interpretation

A number of studies [5, 6] have demonstrated that there is great variability amongst physicians in laboratory test ordering for similar clinical problems. It has also been demonstrated [5, 6] that clinicians often do not have all the information necessary for optimal clinical interpretation of the results they receive.

Clinical protocols and educational programmes have provided some help in this area but are limited in their ability to provide a general solution. In particular, the potential of clinical protocols (practise parameters) is hampered by the requirement for them to be both easily remembered and medically acceptable.

Use of Expert Systems

An Expert System implemented as an interface between the clinician and the laboratory offers the possibility of addressing these areas of concern. The significant benefit of expert systems technology is the ability to represent more sophisticated and widely applicable protocols than can currently be implemented with traditional approaches; and the ability to make that information available to clinicians at the time of ordering.

Advances in office computerization and in expert systems integration have made the development of an expert system interface timely. Today, many physicians employ computers as an integral part of their office practices. Some physicians work interactively with a computer during their office activities to review their patients' medical records. It is expected that this trend will continue.

Laboratory tests are generally ordered by writing the required test on a chart or a laboratory requisition. There is little doubt that, in the future, such orders will be made directly into a computer. With properly designed computer systems, ordering in this fashion will be more efficient than writing and will offer the benefit of adding the order directly to the patient's record while simultaneously conveying the request electronically to the laboratory.

On the return limb of this process, the result from the laboratory will be delivered electronically to the physician's office. This will speed up the result turn around time and will allow direct entry of the results into the patient's chart. There is already a great deal of activity in this area and up to 15% of all laboratory reports are now transmitted in this fashion.

The Laboratory Advisory System is designed to reside on a physician's office computer with communication links to the laboratory. The system is being developed in a modular fashion with each module or knowledge base addressing a particular diagnostic dilemma. The initial knowledge base reported here advises clinicians on test ordering strategies and on the interpretation of results in the investigation of thyroid function. An earlier system which provides interpretive reporting of a wide variety of laboratory test values has been in use for 17 years [7].

The Laboratory Advisory System acts as an intelligent interface between the clinician's office management system and the laboratory information system (LIS). It is built in the Acquire^® expert system development environment. Acquire^® has a knowledge acquisition system and inference engine. The knowledge acquisition system provides a structured approach to developing the expert system application.

In addition to the standard production rule representation of knowledge available in other development systems, Acquire^® offers a knowledge representation scheme based on a pattern recognition approach. The latter representation scheme provides a greater level of specificity in the knowledge base and is particularly useful for representing the exceptional cases not handled well by production rules. The Laboratory Advisory system can be fully integrated with other software (i.e., patient management systems, laboratory information systems) using Acquire^®SDK (Software Development Kit).

A "walk through" the system using thyroid testing as an example is helpful in describing its features.

To order tests for a given patient, a clinician selects "Requisition" on the main menu and a facsimile of a laboratory requisition form appears on the screen.

"Clicking" on the desired test (in this case "Thyroid Function"), immediately causes a dialogue box to appear which asks whether the testing is for making a diagnosis of thyroid function abnormality or for monitoring thyroid replacement therapy. An example of the requisition with the dialogue box overlaid is shown in Figure 1. This requisition is a facsimile of a BC Med Plan approved requisition currently in use with the modification of a "thyroid function" check box (indicating the diagnostic dilemma) replacing the specific TSH and T4 test check boxes.
If "diagnose" is the choice, indicated by a "click" on the check box, a new dialogue box appears with the recommendation of test(s) to order. In this example: "A TSH test is recommended. If the result is outside the reference range (> 3.5 or &< 0.04) a T4 test will be conducted on the same sample." See Figure 2. If "monitor" is the choice, additional information regarding the stability of medication is requested before a test ordering strategy is recommended.
It should be noted that this recommendation is based upon the current (but somewhat outdated) BC thyroid protocol which has been jointly accepted by the Medical Plan and the BC Medical Association. The dated nature of this approach is a reflection of the difficulty in maintaining simple, rigid protocols. Expert systems on the other hand readily accommodate new knowledge. Other versions of the thyroid diagnostic program have been developed to follow the current recommendations of the American Thyroid Association [8]).
At this stage the clinician can click on the appropriate button to "Accept recommendation", seek "Further Information", or "Cancel", by clicking on the appropriate button. Clicking on "Accept Recommendation" will cause the screen to revert to the requisition and the "Tests to be done" field is automatically filled in. In this example with "TSH followed by T4 if necessary".
Clicking on "Further Information" will provide a new window which explains succinctly (and with further references) the scientific and mandated basis underlying the proposed approach.
Clicking on "Cancel" allows the clinician to override the protocol and fill in the "Tests to be done" field with whatever tests he or she wants. This flexibility is critical for user acceptance.
When all desired tests have been ordered, the clinician can e-mail the requisition to the laboratory by clicking on the "Send-to-Lab" button. It should be noted that since not all tests are subject to ordering protocols, tests can be ordered directly. An option to display background information on the test is available.
When the laboratory testing is completed the results are transmitted electronically back to the ordering physician's office computer where it is captured by the Laboratory Advisory System.
To review the results, the clinician selects "View results" from the main menu and a screen appears, listing the results sent from the laboratory. The results are in chronological order and identified by the patient name. Also displayed for each patient is the diagnostic dilemma under investigation. A square symbol (in red on colour screens) appears in this column next to results that are outside the normal reference range, thereby drawing the clinician's attention to these files. In the "Status" column a square symbol (in green) indicates a new result. It will disappears after the clinician has viewed the file.
To view a specific patient file the clinician "double clicks" on the patient record and the numerical value of the test is displayed along with the appropriate reference range and an interpretation. An example is shown in Figure 3. The report outlines: the information given by the ordering clinician; the result/interpretation of the tests ordered; and further investigations recommended given these results. The ability to consult a laboratory physician regarding these results is always available to the clinician.

The results screen provides the clinician with the facility to order the additional tests which have been recommended by clicking the relevant "Yes" boxes under "Further investigations". These tests might include Free Thyroxine, Thyroid Antibodies and TRH Stimulation. If ordered, the request is automatically conveyed to the requisition screen and the clinician sends it electronically to the laboratory where the test(s) can be carried out on the sample which has been held there. When assayed, the results are returned to the clinician along with a composite interpretation which employs all of the information available. Figure 4 shows the composite interpretation for this patient. As will be discussed later, this facility provides faster turn around time and significantly lower cost than the traditional reordering process which involves a second patient visit to the laboratory.

Discussion

The Laboratory Advisory System described above has the potential for significantly enhancing the way in which laboratory tests are ordered and interpreted. This is because:

A wider range of applications is possible with Expert Systems than with non-computer models since the Expert System handles those situations where complexity prevents creating tidy, easily remembered protocols.
An Expert System can be easily modified to accommodate a variety of diagnostic approaches, this reduces the need for universal agreement about a single protocol. Even when this flexibility is not sufficient the system will allow overrides and record justifications.
As an added feature the Expert System provides an educational function for teaching protocols.
The test results are interpreted ensuring that they are put to optimum use.
The facility for allowing ordering of additional tests to be performed on blood samples held in the laboratory will allow the results of such secondary tests to be returned in a few hours, without the great expense of recalling the patient for a second venepuncture. This reordering feature also reduces the need to order all possible tests at the first patient visit. These facts are important since it can be shown that the cost of collecting and processing a specimen is now over 50% of the cost of the test.
In its full implementation such a system can provide tracking of test orders with the reason for ordering and the results of the analyses. Moreover, the system allows simulations to be carried out using alternate strategies to investigate more cost effective or more diagnostic approaches.

The early promise of the thyroid function module has encouraged us to expand the Laboratory Advisory system to include other diagnostic dilemmas (e.g., hepatitis, auto-immune, etc.).

References

Peters, M., and Broughton, P.M.G., "The role of expert systems in improving the test requesting patterns of clinicians." Ann Clin Biochem, 1993; 30: pp. 52-59.
Eisenberg, J.M. "Physician utilization: The state of research about physicians' practise patterns." Med Care 1985; 23: pp. 461-183.
Grossman, R.M., "A review of physician cost-containment strategies for laboratory testing." Med Care 1983; 21: pp. 461-483.
McNeely, M.D., "The Canadian experience." Seminar on Laboratory Utilization. WASP XVI World Congress. Vancouver, B.C., 1991.
Schmitt, B.P., and Elstein, A.S., "Laboratory use: attitudes implicit in a national examination." Med Decis Making 1988; 8: pp. 81-86.
Hilborne, L.H., Wenger, N.S., and Oye, R.K., "Physician performance of laboratory tests in self-service facilities. Residents' perceptions and performance." JAMA 1990; 264: pp. 382-386.
McNeely, M.D., "Computerized interpretation of laboratory tests: An overview of systems, basic principles and techniques." Clin Biochem 1983; 16: pp. 141-146.
Surks, M.I., Chopra, I.J., Mariash, C.N., Nicoloft, J.T., and Solomon, D.H., "American Thyroid Association guidelines for use of laboratory tests in thyroid disorders." JAMA 1990; 263: pp. 1529-1532.

Author information:

Dr. Michael McNeely is a clinical pathologist, past president of the Canadian Association of Pathologists and Head of Chemistry and Immunology at Metro McNair Clinical Laboratories. He has been involved in the application of computers to medical reasoning for many years. In relation to expert system technology, Dr. McNeely has been involved in the development of expert systems for diagnosing liver function; in supervising graduate students in the area and in investigating the application of technology to provide medical information support to remote locations.

Dr. Beverly Smith is a principal of Acquired Intelligence Inc, a Victoria-based firm specializing in expert system technology since 1987. Dr. Smith has a doctorate degree from Oxford University in Experimental Psychology. She has been involved in the development and evaluation of expert systems in a variety of application areas. In the medical field Dr. Smith has been involved in the development of a haematology screening expert system and in the evaluation of a clinical neuropsychology system.