The Permanente Journal: Medical Informatics' Questions

The Permanente Journal: Medical Informatics' Questions

Dr. Krall poses two sets of questions about knowledge coupling software. First, he asks about overcoming practical difficulties and potential errors that he sees in use of the software. Second, he asks about the relationship between knowledge coupling software and evidence-based medicine.

1. The source of potential error that Dr. Krall identifies is "abstraction/interpretation of responses and of physical exam findings by an observer." That risk of error of always present, but knowledge coupling software reduces it. In other words, the guidance given by Couplers protects against the "GIGO" problem. That protection is absent when providers are left to their own devices. For example, knowledge coupling software:
• precisely defines needed data for the specific problem situation,
• includes explanations for questions that might not be understood,
• provides graphics that may be helpful users performing portions of the physical examination,
• couples the responses with the best medical literature,
• provides citations to the literature relied upon,
• documents the encounter for later use and reexamination.
Moreover, knowledge coupling software accommodates uncertainty and flexibility. For example, data entry may include "not sure" as well as positive and negative responses, and these responses may be supplemented with free text entry of observations other than the choices presented by the software.

The practical difficulty that Dr. Krall identifies is the possible unwillingness or inability of the patient or family to engage in detailed data collection. In reality, when data are put to good use within an effective system of care, patients and families are usually more willing and more able than physicians to engage in detailed data collection. Unlike physicians, patients and families do not feel pressure to end the encounter and get to the next person in the waiting room. Compared to physicians, patients and families have the greatest stake in getting all the right data at the outset. Physicians may perceive patients as reluctant to pursue detailed data collection, but that perception reflects patients' negative experiences with the present non-system of care. Too often patients find themselves asked the same questions over and over by multiple physicians, or find that those physicians reach different conclusions from the same data. Patients have no assurance that their data will be accurately assessed or even taken into account.

In contrast, when patients use knowledge coupling software (and experience careful follow-up with problem-oriented medical records), they find themselves navigating a trustworthy and transparent system of care for their total medical needs. Patients who experience such a system tend to embrace it.* For further discussion and examples of patient reactions to knowledge coupling software in primary care practices, see part IV.G.4 of Medicine in Denial.

2. Dr. Krall's second set of questions concerns evidence-based medicine. He asks: "How is accurate and reliable medical knowledge developed and encoded into knowledge couplers, absent evidence based medicine?" But evidence-based medicine is not absent from knowledge coupling software. On the contrary, the software incorporates evidence-based medicine, including randomized clinical trial results. Most important of all, the software helps the user navigate quickly to the specific evidence most applicable to the individual patient's needs.

Evidence-based medicine is a body of well-documented scientific evidence about disease entities plus their diagnosis and management. Without external tools, that evidence gets transmitted and applied to individual patient needs at the time of care through the minds of physicians. Their limited memory and other cognitive vulnerabilities cause an enormous voltage drop when they try to couple detailed general knowledge with the complex needs of unique patients.

Practitioners and their patients need protection from this voltage drop. That is what knowledge coupling software provides. Loaded with the best evidence-based knowledge available, the software couples that knowledge with detailed patient data. Software accomplishes this simple matching process with speed and accuracy unachievable by the human mind alone. Moreover, because the knowledge coupling software matches very detailed data with comprehensive medical knowledge, the software's output reflects the patient's uniqueness. In this way, the software can be used to individualize care rather than standardize it—thereby addressing a basic criticism often leveled against evidence-based medicine.

This knowledge coupling process sometimes yields uncertainty. For example, coupling available data with the medical literature might suggest an array of possible diagnoses none of which can be either ruled out or confirmed based on initial data. In these situations, Dr. Krall asks, "Do not PKCs also have to use some type of relevancy ranking (statistical or otherwise) to prioritize diagnostic possibilities … ? If so, how is this fundamentally different from 'an evidence based ranking'"? The difference is that an evidence-based ranking of diagnostic possibilities is based on statistical generalizations about large populations, while the ranking provided by Couplers is based on evidence specific to the individual patient. Suppose, for example, a patient is worked up for a problem of severe fatigue, and suppose further the Coupler results show that the patient manifests 5 of the 6 findings that typically indicate adrenocortical insufficiency but only 1 of the 5 findings that typically indicate some form of sleep disorder. For that particular patient, adrenocortical insufficiency is a higher priority for diagnostic investigation than sleep disorders. This is the case even though adrenocortical insufficiency is a rare condition that would not rank high in an evidence-based ranking of the possibilities. Alternatively, suppose the Coupler results show that the patient manifests 3 of 6 findings for adrenocortical insufficiency and 3 of 5 findings for sleep disorder problems. In that equivocal situation, it might be useful to prioritize between the two possibilities by considering the statistical fact that sleep disorders are much more common than adrenocortical insufficiency. That illustrates how Couplers can incorporate evidence-based medical knowledge drawn from population data while protecting against misguided use of statistical generalizations.

In short, knowledge coupling software promotes the central purpose of evidence-based medicine—to base medical decisions on objective medical evidence, rather than extraneous factors. An external tool like knowledge coupling software protects unique patients and their caregivers from such extraneous factors as personal opinion, local custom, statistical generalizations, vendor marketing, the tunnel vision of specialists, or the financial interests of third party payers and providers. Moreover, using knowledge coupling software in conjunction with electronic, problem-oriented medical records generates feedback on medicine's evidence base. It becomes possible to trace detailed connections among patient problems, medical interventions, the patient's individual characteristics and ultimate outcomes. This can lead to a highly individualized form of medical knowledge, quite unlike the "evidence-based" generalities that physicians and patients are too often expected to follow.

We hope this discussion is helpful for Dr. Krall and other readers.

Lawrence L. Weed
Lincoln Weed

* In particular, consider the following from Dr. Ken Bartholomew. He describes how knowledge coupling software improves the doctor-patient relationship:

"Not only do patients see the thoroughness involved in the use of couplers, but they sense that we care enough to give them the kind of thoroughness that they feel entitled to. With the coupler’s systematic review of details in the patient’s life that could be relevant to the current problem, the patient feels that his or her individual situation has been thoroughly examined and all possible conclusions have been taken into account. In management couplers, they further see the many different combinations of therapy and understand that the care of a complex, long term problem requires a detailed understanding of the patient’s unique situation, followed by a careful monitoring of the options chosen. Even when a diagnosis is still in question, they have, in my experience, been completely satisfied with the outcome of the encounter. In addition, by receiving a printout of the findings and possible causes, they feel empowered to review the situation at home and to watch for signs and symptoms that may aid the diagnostic process in the days or weeks to come. The use of couplers teaches them that there is a time course to disease and that not all signs and symptoms necessarily occur “by the book” or simultaneously. By thus empowering our patients with information, as opposed to leaving them in a void, we reinforce their collaborative role as part of a team working toward an understood goal."

Bartholomew K., "The Perspective of a Practitioner," in L.L. Weed et al., Knowledge Coupling: New Premises and New Tools for Medical Care and Education (Springer-Verlag, 1991), pp. 238-39.

Medical Informatics' Questions

It was a pleasure to read this interview with the visionary and inspiring Dr Weed. I was given his 1969 book to read, by a professor of medicine and important mentor of mine, while I was still a college student on an internship at a medical school research lab. I have had the pleasure of hearing Dr Weed speak on 2 or 3 occassions and his thinking has had an important impact on me.

As a description of the POMR and PKC journey, I think this interview is of interest and is good to publish. That said, I had some questions that I would love to have been able to ask the author- some of which are below.

Knowledge Couplers require "detailed data collection", which implies time, effort on the part of the patient or family (who may be unable or unwilling to comply), and abstraction/interpretation of responses and of physical exam findings by an observer. How do you overcome these barriers and sources of potential error? As a computer program, the PKC is only as good as the data it has to work with ("GIGO").

How is accurate and reliable medical knowledge developed and encoded into knowledge couplers, absent evidence based medicine?

Are randomized controlled trials and similar means of developing reliable knowledge about what works and what does not work to diagnose, treat and manage disease not still fundamental in the paradigm you describe?

Do not PKCs also have to use some type of relevancy ranking (statistical or otherwise) to prioritize diagnostic possibilities when there is not a single diagnosis that is certain based on the available data? If so, how is this fundamentally different from "an evidence based ranking"?


Michael Krall, MD
Primary Care Physician,
Medical Informatics