Printer Friendly

New directions in interpretive test reporting.

Computerized expert systems are already solving diagnostic problems in the laboratory, and the near future promises an explosion of new applications.

The author is medical director of the Tri-Counties Blood Bank Santa Barbara, Calif , and patholagist at Santa Barbara Cottage Hospital. Medical Group Pathology Laboratory in Santa Barbara, and Santa Ynez (Cant) Valley Hospital He is also associate clinical professor of pathology at the University of Southern California in Los Angeles

Modern laboratories produce reams of results, with the risk that critically important abnormal findings will be overlooked. Indeed, Charles Altshuler, M.D., has observed that test results outside a specialist's areas of expertise are often overlooked or not acted upon.(1)

How can we in the laboratory insure that a clinician is made aware of the significance of a set of laboratory results? In the long run, the best answer will be to close the feedback loop-that is, make diagnosis and treatment information readily available to the laboratory so that we may verify that the appropriate response is being made. Until our automated information systems reach this stage, however, one of the most useful tools is interpretive reporting, the subject of this article.

Interpretive reporting has been defined in a variety of ways.(2) Some authors even consider listing of age-corrected reference ranges a form of interpretive reporting. For the purposes of this article, however, we will define it as provision of a textual comment, explaining the probable significance of a laboratory result or panel of results.

For what sort of tests is interpretive reporting appropriate'? Figure I suggests some possibilities.(3) Clinicians will generally welcome interpretive comments for tests they do not commonly (or routinely) order, particularly those outside their area of specialization. Certain types of tests, because of their nature and complexity, should be accompanied by interpretive comments whenever the results are reported. For other tests, interpretive reporting may be provided only when requested by the attending physician.

Interpretations are most appropriate when the tests yield relatively specific results, and the significance is mostly independent of factors in the patient's clinical history. For example, it is easier to say something intelligent about a strongly positive serological test than about an elevated lactate dehydrogenase. Given sufficient clinical data, however, it is quite possible to generate useful "interpretive comments for as common (and nonspecific) a package as the 20-test chemistry panel.

Even in the absence of clinical history, single abnormal results for certain tests may be automatically appended with an interpretive comment, describing the probable significance of the finding. For example, an elevated thyroid-stimulating hormone (TSH) could indicate "probable hypothyroidism," or a depressed TSH, "probable hyperthyroidism, or overtreatment with thyroid medication." The pathologist need not review every such result, but should periodically verify that proper reporting rules are being used, especially when methodology and reference range are changed.

In most cases, however, the significance of a given laboratory result depends on the clinical context and history. The most extensive and useful applications of interpretive reporting have been developed with formats calling for the ordering clinician to supply clinical data to the laboratory.(4) In one instance,(5) the requiredclini cal data can be reduced to a limited, standardized set of questions (e.g., is the patient taking a diuretic, is the patient a known diabetic?), allowing detailed interpretation of results of an executive physical program. As medical care moves progressively toward the automated medical record, more clinical data become available to the pathologist, and we can do a progressively more effective job of interpretive reporting. 6

In general, useful interpretive reports require the exercise of medical judgment, and should be generated by a pathologist. This is particularly true where a patient's clinical history must be integrated with the laboratory results, or where electrophoretic or other morphological patterns must be evaluated.

Even when an automated interpretive algorithm, or an expert system, can be applied to a set of laboratory data, the results should be reviewed by a pathologist before reporting; laboratory results are not totally precise, and interpretive algorithms are often structured to make binary decisions, which may be inappropriate when a value is close to a decision level. As an example, an automated algorithm to interpret protein electrophoreses may yield an interpretation based on numerical scan results, but a pathologist's visual examination of the plate indicates that the "abnormality" is actually an artifact, or statistical variability in the densitometry process.

These important qualifications aside, it is worth noting the few exceptions in which laboratory data are sufficiently standardized, and pathognomonic, that the majority of panel results can be interpreted and reported by an automated algorithm without human intervention.

In our laboratories, we had been providing a pathologist interpretation with the majority of hepatitis profiles--consisting of hepatitis B surface antigen (HBsAg), hepatitis B core antibody (HBcAb), hepatitis B surface antibody (HBsAb), and hepatitis A IgM antibody. It became apparent that the vast majority of panels fell within a few patterns, with clearly defined interpretations. We are fortunate to have a laboratory computer system with extensive user-defined calculation capabilities, and were readily able to program interpretations for commonly occurring patterns of hepatitis test results into this system (Figure II).

The interpretive comments are intentionally terse, to fit within the constraints of our clinical laboratory reporting format; others (7) have described more extensive interpretations, which could be printed using similar rules. These interpretations are automatically produced as each batch of testing is completed and are then printed directly onto the patient reports; the pathologist reviews the interpretations after reporting.

We must emphasize that only certain patterns result in an automatic interpretation. Other patterns-isolated HBsAg positivity, for example-or borderline results are brought to the pathologist for review and interpretation. These reviews, with the pathologist's attention focused on the significance of an unusual pattern or result, rather than on the large mass of expected patterns, have allowed us to identify and question unusual patterns of results. Upon examination, we have found and corrected assay problems undetected by our usual QC protocols.

Production of interpretive reports is greatly facilitated by a computerized laboratory information system. Even where the report requires extensive medical judgment, the computer serves as a very efficient secretary. The computer can be used to assist the reporting process in a variety of ways, from pure secretarial to high-level artificial intelligence and expert systems :(8)

*Clerical aid. The pathologist may choose one of several dozen "canned" interpretive comments. For rare cases, the option of a free text interpretation should also be available.

*Deterministic and categorical approaches. The computer may be programmed with an algorithm, decision tree/table, or flow chart. Based on a series of binary (true/ false) decisions, the program selects a coded comment (as in the hepatitis panel interpretations, Figure II). For all but the simplest of circumstances, these approaches are difficult to maintain-they must be made very complex to represent reality. Although they cannot tolerate missing data, and one wrong turn may result in an invalid answer, decision-tree approaches have been among the more popular tools for development of interpretive reports.

* Statistical/probabilistic approaches. Bayesian and multivariate statistical systems, very popular in the laboratory and medical research literature a decade ago, have fallen out of favor because 1) they require collection of large batteries of test results, and "gold standard" diagnoses, on hundreds of patients; 2) they assume the patient being tested is a member of exactly the same population as that used to develop the statistical database; and 3) they cannot explain to the user how or why a particular pattern of test results indicates a given diagnosis.

* Knowledge-based expert systems. These combine the knowledge of persons who are expert in a subject with a separate "inference engine" computer program that can use this knowledge to solve the question at hand. Knowledge may be represented and stored in a variety of ways. The most common paradigm in laboratory medicine systems has been as rules: "If A is true and B is false and C is true, then there is an 80 per cent probability that F is true."

A major advantage of such systems is that they are usually designed to "explain" their reasoning, by recapitulating the rules connecting the original observations and laboratory values with the final conclusion.

The major bottleneck in development of these systems is usually the conversion of the seemingly intuitive knowledge of the human expert into the carefully defined rules of the knowledge base. For example, there are no good programming tools for dealing with the time course of disease. Another caveat is that cases presented to an expert system must be selected to resemble those used to build the system; expert systems, unlike humans, have a great deal of difficulty recognizing "I've never heard of something like that before."

Knowledge-based systems have been developed for a variety of applications, including protein electrophoresis interpretation.(9) The incorporation of an expert system into a densitometer made many of us in laboratory medicine sit up and take notice that useful applications were forthcoming.

Other laboratory application areas have included hepatic panels, leukemia classification, (10) lymph node diagnosis,(11) blood utilization monitoring,(8) endocrine diagnosis, and coagulation disorders. (12) Experiments in general internal medicine applications, including Internist/Caduceus, Mycin, Oncocin, and AI/Rheum, have advanced the scientific understanding of AI/expert systems and have shed light on the application of these systems to medical problems.

The major advances in interpretive reporting over the next few years will rely upon expert systems. Some of these system (5) have been programmed to incorporate properly encoded clinical history. The more powerful tools even enable the system to learn from new cases.

As recently pointed out by Hugo C. Pribor, M.D., Ph. D., information overload affects pathologists as much as it does other physicians.(13) Therefore, an algorithmic or expert system can greatly assist pathologists in providing consultative services-particularly for infrequently performed profiles.

After several years of glowing reports on the "promise of artificial intelligence," the current generation of expert systems software tools, supported by extremely powerful microcomputers, is finally making good on these promises. We expect to soon see a virtual explosion of articles on laboratory-based expert systems and commercial applications.

What options are available right now to the laboratory wishing to use computer-assisted interpretive reporting? Several commercially supported laboratory information systems allow the laboratory to add decision rules and interpretive algorithms to the standard reporting format. Partners in Medical Computing (500 Peterson Road, Libertyville, III. 60048, 312/680-9488) markets an interpretive reporting module (Figure III), running on IBM PC-compatible microcomputers, that may be interfaced with a variety of laboratory information systems or integrated with MUMPS-based systems.

Dr. Hugo Pribor has developed a number of interpretive reporting tools. His recent expert system, IBM PC-based version of the Pathology Consultation System(12-14) (Figure IV) is being marketed by Lea & Febiger (John F. Spahr Jr., Lea & Febiger, 600 Washington Square, Philadelphia, Pa. 19106).

One of the most commercially successful AI/expert-based systems for laboratory medicine is in hematopathology: Dr. Bharat Nathwani's Intellipath lymph node consultation system, developed in collaboration with a group of Stanford-trained computer scientists, includes both a textual and a visual (videodisc) database, making this a most impressive tool for the pathologist. Additional organ systems will be added in the near future. Intellipath is being marketed by the American Society of Clinical Pathologists (2100 W. Harrison St., Chicago, Ill. 60612, 800/621-4142).

A number of LIS vendors are devoting considerable resources to the development and implementation of sophisticated, artificial intelligence/expert system workstations for interpretive purposes. Some of these have been demonstrated at ASCP/College of American Pathologists national meetings, and some are in use at test sites. We expect to see them soon in widespread use.

If your department has the necessary resources, including access to computer scientists well versed in artificial intelligence techniques, you may wish to tackle the job of "teaching" a generalpurpose, expert system development tool about laboratory tests. A recent survey showed that 25 per cent of respondents from academic laboratory medicine programs are now pursuing such projects.(8) We hope these efforts will soon evolve into transportable, commercially supported tools that all of us can use in our laboratories.

You are limited only by your imagination and your creativity in providing interpretive information for the clinicians you serve. The ideas presented here represent only a sampling of ways to make sure your message is getting through!(15)

1. Altshuler, C H Building a database for monitoring and facilitating heath care. Clin. Lab. Med. 3:179 204,1983

2. Speicher, C.E., and Smith, J W Interpretive reporting in clinical patho ogy. JAMA 243: 1556 1560,1980.

3. Aller, R.D. Interpretive report ng, Clin Lab. Med. 3(1): 205 -217, 1983.

4. Hosty, T.A. The consultative laboratory assessment system, the CLAS profile Arch Path. Lab. Med, 112: 1203-1206, 1988

5. Van Lente, F.; Castellani, W. 'Chou, D.; Matzen, R.N.; and Galen R S. Application of the Expert consultation system to accelerated laboratory testing and interpretation Clin. Chem. 32: 1719-1725,1986.

6. Clayton, P D . Evans, R S.. Pryor, T.; el al. Bring ng HELP to the clinical laboratory--Use of an expert system to provide automatic interpretation of laboratory data Ann Ciin. Biochem. 24(Suppl.1):5 11, 1987.

7. Soloway, H.B Interpreting hepatitis profiles. Diag Med. 2(1): 29 34, January/February 1980.

8. Spackman, K.A., and Connolly, D.P Knowledge based systems in aboratory medicine and pathology Arch. Pathol. Lab Med 111: 116-119, 1987.

9. Weiss, S M Kulikowski, C.A.: and Galen, R.S. Representing expertise in a computer program The serum protein diagnostic program. J. Clin. Lab. Automation 3: 383 387 1983.

10 Fox, J . Myers, C.D.. Greaves, M F.; et al. Knowledge acquisition for expert systems: Experience in leukemia diagnosis Meth Inform Med. 24: 65- 72 1985

11. Nathwani, B. Lymph node pathology systems. CAP Today 2(8): 39, 1988.

12. Hurlbut, T.A., and Pribor H C . Interpreting coagulation profiles A comparison of expert systems Lab Management 27(1): 32-36, January/February 1989

13. Pribor, H C Expert systems and laboratory medicine. Lab. Management 26(11): 36 41 , November 1988.

14. Pribor, H.C. The pathology consultation system Lab. Management 26(12): 30 35, December 1988

15. Krieg, A.F. "Laboratory Communication Getting Your Message Through." Oradell, N.J., Med ca Economics Books, 1978
COPYRIGHT 1989 Nelson Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1989 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Aller, Raymond D.
Publication:Medical Laboratory Observer
Date:Apr 1, 1989
Previous Article:Two-pronged attack seeks to end physician self-referral.
Next Article:The cost impact of decentralized testing.

Related Articles
Despite ongoing challenges, pathologist reimbursement rules take effect.
How 1995's instrumentation will impact on the lab.
CAP offers voluntary standards for nontraditional test sites.
Quality control in the new environment: microbiology.
Automating laboratory send-out tasks.
Doctors speak their minds on lab profiles and problems.
Patient self-referral.
Teach staff to walk in surveyors' shoes: all staff need to be aware of regulatory guidelines regarding abuse. (Feature Article).

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters