Click here to access the corrected complete pdf version.

Part I: A Micro Perspective on Medical Decision Making

Lawrence L Weed, Lincoln Weed

. . . a whole calling may have unduly lagged in the adoption of new and available devices. It may never set its own tests. . . . there are precautions so imperative that even their universal disregard will not excuse their omission. � Judge Learned Hand⁽¹⁾

A. Introduction

A pervasive design flaw in advanced health care systems is their unnecessary dependence on fallible, idiosyncratic inputs from clinical workers. In medical decision making this flaw is pivotal. Medical decisions are still based largely on the recall and processing of complex information by highly trained physicians. Yet, their cognitive inputs fall short of what medicine requires, too often producing decisions that are deficient in quality and resistant to organized improvement.

Improving cognitive inputs can take place if the burden of recalling medical knowledge and integrating it with patient data is shifted from the unaided human mind to modern electronic tools. That design change is key to releasing medicine from hidden constraints on quality and productivity. Yet, a false ideal of physician decision making stifles genuine reform of this kind. Practitioners, academic medicine, government-sponsored systems and private sector systems alike have never faced up to medicine's unforgiving demands. Accurate, cost-effective decision making requires information retrieval and processing that the unaided mind in real-world conditions cannot achieve. Aiding the mind with practice guidelines and computers for knowledge retrieval alone is not sufficient, because the mind must still integrate that general knowledge with complex data on unique, individual patients. Critical patient data may not be comprehended or not gathered at all, because of the mind�s limited ability to integrate general knowledge with patient data. Much of those data are readily available from medicine�s enormous array of inexpensive, safe, simple tests, observations and procedures. Yet, because they are unable to exploit that data, doctors often resort unnecessarily to risky or expensive procedures and technologies.

In short, medicine demands scientific rigor that the unaided mind can achieve only in sheltered, academic environments. Yet, medicine is practiced in commercial and publicly-funded environments under severe time and cost constraints. Medicine must emulate other commercial and scientific endeavors that have developed tools and processes for escaping the limits of the individual mind. Medicine requires a new division of intellectual labor, a division between tools that retrieve and process information, and users who apply judgment and values to arrive at decisions.

In addition to misplaced faith in the efficacy of doctors' decisions, another false ideal stifles genuine reform. The ideal is that professionalism and altruism lead doctors to the same decisions that their patients, were they fully informed, would otherwise make for themselves. But two realities undermine this ideal. One reality is that financial considerations, cultural background and other personal factors affect doctors' decisions. The other reality is that in difficult cases only patients themselves are ultimately entitled to choose among risky, uncertain medical interventions. Giving doctors or health care organizations authority over medical decisions or the information needed to make those decisions disenfranchises patients from inherently personal choices concerning their own bodies and minds.

Because of its archaic intellectual infrastructure and its confusion about decision making authority, the health care system produces decisions that frequently do not conform to the individual needs of complex, unique patients. Such decisions cause substantial harm, in both medical and economic terms. Moreover, the prevailing confusion and the lack of patient autonomy block genuine reforms in many areas, including "consumer protection," credentialing, graduate medical education, analysis of clinical outcomes, and the framework for marketplace competition.

When we turn from medical decision making to skill in executing decisions, we again find that inputs from the health care workforce are of variable and uncertain quality. Faithful execution of medical decisions requires high quality inputs of manual skill and technique from caregivers and well-designed systems to support them. Many virtuoso performers in medicine achieve skill of a high order. Yet, their achievements are largely individual; health care institutions do not establish the standards and systems needed to assure that decisions are executed skillfully by all practitioners. Medical care thus too often fails even in cases where decisions are correctly made.

Without effective scrutiny and control of inputs, marketplace competition is distorted. Competition occurs in the dimension of price, not quality, because quality is unmanageable and because patients � who have the greatest stake in quality � are not adequately informed decision makers. Mere price competition, disconnected from quality and patient choice, is corrosive and demoralizing rather than productive and creative. For patients and caregivers alike, the health care system is thus increasingly inhumane. For society as a whole, productivity gains are elusive, and the goal of universal access to care seems utopian. A sense of futility prevails.^{(2) (3) (4) (5)}

These realities call into question a prevailing theory among policy makers. Seeing marketplace competition occur in price rather than quality, policy makers theorize that outcome comparisons and other quality measures can create economic incentives for providers to compete on the basis of quality as well as price.⁽⁶⁾ What this theory overlooks is that providers have always had strong incentives to improve quality for their patients � personal, professional, legal and (in part) economic incentives. Yet, providers are often unable to act on quality incentives effectively. When doctors, for example, miss diagnoses or overlook cost-effective treatments because they lack the tools needed to recall and process complex information, rearranging financial incentives accomplishes nothing. By default, providers act on countervailing incentives for financial gain, which tends to be more readily achievable than high quality. Therefore, policies built around outcome comparisons, capitated financing, quality "report cards" and other economic incentives are misconceived.

Reforms to address these problems require that we distinguish between two forms of expertise in the health care workforce � knowledge and skill � and the two corresponding functions � making medical decisions and executing those decisions. For medical decision making, reform must involve better guidance tools � properly designed software and medical records. Software tools, in contrast to the unaided mind, can effectively integrate general medical knowledge with patient data, thereby extracting the maximum useful information from the safest and most cost-effective medical procedures. Equally essential are properly structured medical records. Rigorously documenting the medical condition of patients and the actions providers in medical records is necessary for corrective feedback loops (within both the process of patient care and the body of medical knowledge). Moreover, such guidance tools mean that the role of physicians and other providers is not to be decision makers for their patients but rather to help patients decide among the options that the guidance tools reveal.

Once the patient�s decision is made, the provider�s role is to execute it skillfully. To improve providers� execution of medical decisions, the reforms needed are better management practices of two kinds. One is periodic credentialing of workers based on demonstrated competence in discrete, hands-on skills (rather than one-time or infrequent educational credentialing and apprenticeships purporting to certify a broad range of knowledge and skills). The other is improving the systems within which workers perform in order to enhance their performance and protect against individual failure.

Table I summarizes traditional approaches to these issues and the new approaches we discuss below. Under the old approaches, an enormous voltage drop occurs between what medical science has made possible and what patients actually receive. The enormity of this failing is not well understood, because medicine�s remarkable scientific achievements overshadow it. Yet, those achievements for many patients go to waste. As concluded in one review of quality studies from 1993 to 1997: "For most care that has been studied, there are large gaps between the care that people should receive and the care they do receive. This is true for all three types of care [acute, chronic and preventive]. It is true whether one looks at overuse or under use. It is true in different types of health care facilities and for different types of health insurance. It is true for all age groups, from children to the elderly. And it is true whether one is looking at the whole country or only one city."⁽⁷⁾

TABLE I � Concepts of Reform in Medical Practice

Function Problem area Old approaches New approaches

Medical decision making Basis for decision making The mind of the educated physician Software tools and patient choices

Structuring and recording the decision process Broad discretion for physicians� "clinical judgment"; manual medical records varying in structure and completeness Defined, explicit structure for physicians and others; complete electronic medical records with a problem-oriented structure

Execution of decisions Individual performance Educational credentialing and apprenticeships, certifying general knowledge and skill Credentialing based on periodic demonstrations of actual competence in discrete skills

System performance Individual responsibility for errors System improvements to support individual performance

Studies show not only that failures of quality occur but that they inflict harm and waste resources on a large scale. And such studies have been appearing for decades. Consider this example from 1948.⁽⁸⁾ "In reviewing the records of more than 6,000 hysterectomies in thirty-five hospitals, physician James Doyle found hundreds of women who had received either no preoperative diagnosis or a simple diagnosis such as �pain.� Postoperatively it turned out that 30 percent of all the patients ages twenty to twenty-nine who were subjected to hysterectomy had no disease whatsoever, a number that Doyle rightly called �appalling.�"⁽⁹⁾ Equally troubling, that study cannot be dismissed as outdated, because substantial evidence exists that in managed care organizations inappropriate hysterectomies still occur.^(10)(11)

Just as such failures in quality are widespread, so successes in quality improvement are isolated and often hard to disseminate. This stagnation is especially discouraging in a technologically advanced, information-intensive field like medicine. Many other industries have found ways to achieve remarkable gains in quality and productivity. But reform efforts in medicine "resemble a team of engineers trying to break the sound barrier by tinkering with a Model T Ford."⁽⁵⁾ Medicine's Model T is the unaided human mind. It is unproductive in harvesting medicine's most abundant resource � information. Yet, false ideals about physician functioning block awareness of this barrier to change. Outcomes are improved where the relevant inputs are identified and changed. Doing so on a systematic, continual basis requires new tools and approaches for routine patient care. This paper focuses on one element of such reform � using specialized software tools to couple medical knowledge with patient data. That element of reform in decision making is of the most immediate interest, because it is a necessary foundation for other reforms. Skillful execution of medical decisions is useless if the wrong decisions are made. In turn, making the right decisions in real-world conditions depends on eliciting individually relevant medical knowledge and patient data, while filtering out clearly irrelevant information, in order to identify relevant options and the pros and cons of each option for the individual patient. That process is the foundation for everything else that happens in medicine, and flaws in that foundation undermine the entire health care system.

B. A case study of failure in diagnostic decision making

Upon this gifted age, in its dark hour,

Rains from the sky a meteoric shower

Of facts . . . they lie unquestioned, uncombined.

Wisdom enough to leech us of our ill

Is daily spun, but there exists no loom

To weave it into fabric . . .

- Edna St. Vincent Millay⁽¹²⁾

As an example of flawed decision making, consider the following case, described in a 1996 New England Journal of Medicine article on "Clinical Problem Solving."⁽¹³⁾ A 15 year old girl was unable to attend school for three months because of excessive fatigue. For seven months she had experienced shortness of breath, weight loss and amenorrhea. On the initial physical examination the girl's doctors noticed mild hypotension and multiple nevi (moles), but the examination seemed generally unremarkable, as were her medical history and that of her family. During the next month, the girl was admitted to the hospital three times with additional symptoms of distress, including epigastric pain, nausea, bilious emesis, diffuse abdominal pain, diarrhea, dehydration and further weight loss. Other findings were normal except for some borderline test results (e.g. a serum sodium of 134 mmol per liter, observed in the second hospitalization) and some gastrointestinal abnormalities that would not account for her symptoms.

During this process multiple consulting specialists at a teaching hospital subjected this girl to what the article described as "dozens of blood tests, immunologic studies, endoscopies, scans, other radiographic tests, and biopsies." No diagnosis was established, but several medications were started. Nasogastric feedings were tried and then abandoned when persistent vomiting developed. Parenteral nutrition was begun. "The endocrinology consultant thought that the patient's amenorrhea was consistent with that seen in eating disorders." During the third hospitalization, a urinalysis was positive for emetine (an ipecac derivative). "Both the girl and her family adamantly insisted that she had not used ipecac," but the test result suggested bulimia (the girl perhaps was using ipecac to induce vomiting) or "poisoning by a family member." The doctors considered involving child protective services and obtaining a complete psychiatric evaluation. A preliminary evaluation, however, did not support the diagnosis of an eating disorder, and follow-up tests were negative for emetine. The earlier test result appeared to have been an error.

Then blood tests revealed abnormal serum electrolyte levels (e.g. a serum sodium of 128 mmol per liter), which suggested new diagnostic possibilities. Further investigation narrowed these down to Addison�s disease, an insufficiency of adrenalcortical hormones. Left untreated, Addison�s disease is fatal. The girl�s doctors began hormone replacement therapy, and her condition rapidly improved. "Fortunately, the patient survived not only her illness but the myriad tests and treatments administered before the telltale electrolyte levels revealed the correct diagnosis. Fortunately as well, this happened just in time."

Whether the girl suffered with the psychiatric symptoms that usually accompany Addison's disease is not stated. Nor does the article describe the emotional impact of her health care on the girl and her family. The article does not indicate the duration of her care, the costs incurred, or whether all those costs were deemed "medically necessary" by a third party payer. The article does, however, summarize the girl�s physical state; it also candidly describes the perplexity of her doctors and an expert clinician who discusses the case:

At one time or another, the unfortunate child described here suffered from weakness, breathlessness, abdominal pain, nausea, vomiting, diarrhea, weight loss, and severe malnutrition. Examination never disclosed any findings other than tachycardia, mild hypotension and dehydration. Neither the patient�s physician and several consultants nor the discussant considered the correct diagnosis � Addison�s disease � on the basis of these clinical findings. Yet the diagnosis became obvious as soon as hyponatremia, hyperkalemia and hypobicarbonatemia developed months after her initial presentation. . . .

Some might view the diagnostic struggle that took place here is an inevitable part of medicine. After all, highly qualified physicians were involved. They were faced with a rare disease whose classic manifestations are each non-specific symptoms like fatigue or hypotension each suggest innumerable diagnostic possibilities. In turn, each possible diagnosis suggests many findings to check and various treatments to try. This girl�s doctors thought of numerous possibilities and explored them aggressively, generating much more data to ponder. Burdened with other patients to care for, they had limited time to analyze the data on this one patient. They were thus unable to separate the wheat from the chaff in a vast field of information. Accordingly, the article observes:

Who is at fault here? In retrospect, the diagnosis seems obvious. Fatigue, weakness, dehydration and hypotension are classic manifestations of Addison�s disease. . . . A rare diagnosis that is obvious in retrospect, however, is often not so obvious prospectively. . . . Only the toughest critic could fault any of the physicians for not making the diagnosis earlier.

Yet, what is striking about this case is that the stated "classic manifestations" of Addison�s disease were observed at or near the outset of care, without invasive or costly tests. Moreover, as we discuss below, other clues to the disease were also readily observed in the early stages of care. Had the significance of even some of these clues been recognized, the need for testing to confirm or rule out Addison�s disease would have been "obvious," especially given the paucity of strong evidence for alternative possibilities. Yet, Addison�s disease was not even considered "until it was nearly too late to save the child�s life."

The authors of the article acknowledged that diagnostic failures of this magnitude are all too common:

Disaster lurks when a patient has a life-threatening disease that not only is rare but also presents with either atypical or nonspecific symptoms and signs. In patients with diseases that fit this description, vastly excessive testing and numerous attempts to treat putative diagnoses are the rule. We can be certain that in such instances some patients die because the correct diagnosis is never entertained and that even after an autopsy the mystery often persists.

The doctors� perplexity in this case is especially alarming when one considers that the case was quite simple compared to many � the abnormal findings to analyze were relatively few, critical findings were available early, and together many of them pointed to a single, clearcut, treatable diagnosis. Yet, the article�s authors, and some letters to the editor, suggest that the case appears simple only in retrospect.

Properly approached, however, this case was not complex or difficult. Contrary to the authors� view that the findings on the girl were "atypical" and "nonspecific," the early findings were quite typical and highly specific, when considered in combination. At the first examination of the girl�s fatigue problem, the findings mentioned by the article included mild hypotension, weight loss, and "multiple nevi, none with abormal characteristics." In noting this latter finding, the article was incomplete; numerous caregivers had "commented on the patient's large number of deeply pigmented nevi," according to the authors� subsequent response to letters to the editor.⁽¹⁴⁾ A careful review of the medical literature would have revealed that fatigue, hypotension, weight loss and unusual pigmentation are among the signs and symptoms that characteristically appear with Addison�s disease.⁽¹⁵⁾ ⁽¹⁶⁾ Indeed, with respect to pigmentation, the article�s authors subsequently uncovered a 1990 report in the Archives of Dermatology describing an Addison�s disease patient in Denmark with eruptive nevi similar to this girl�s.⁽¹⁷⁾ Moreover, within two weeks of the initial visit, other clues to the disease were observed, including abdominal pain, nausea, dehydration, vomiting, and a serum sodium level that on its face was borderline and in the context of dehydration should have been interpreted as below normal (hyponatremia).

In short, a simple linkage between the initial findings and the correct diagnosis could have been identified quickly, without costly or elaborate tests and without resort to sophisticated clinical analysis. To accomplish this, the girl�s doctors needed to focus on their patient�s individual combination of signs and symptoms, and ascertain the possibilities suggested by that combination. Instead, their approach was backwards. They focused not on detailed patient data but on general knowledge about large populations. As the article explains: "the clinician usually begins a diagnostic investigation by considering (and excluding) the most common diagnoses. As these most common diagnoses become less likely, many less common diagnoses are considered" (emphasis added).

Yet, what diagnoses are common or uncommon depends on what population one examines. In the general population, Addison�s disease is indeed uncommon. But that fact is wholly irrelevant to the population of people with excessive fatigue, low blood pressure, unusual pigmentation, gastrointestinal symptoms, and hyponatremia, because for them Addison�s disease is far more common than in the general population.

Had the girl�s doctors recognized even some of those clues to Addison's disease, they immediately could have watched for other clues or ordered the testing sequence needed to confirm or rule out the disease. The only barrier to an earlier diagnosis, therefore, was the physicians� cognitive inability to recognize the linkages between the diagnosis and the findings on their patient. The way around this barrier is apparent � routine use of computer software to recognize linkages between patient data and medical knowledge. Without the aid of software, expert clinicians viewed the clinical findings on this girl as "atypical and nonspecific"; with software, even laypersons could recognize that the girl's combination of findings was quite typical and highly specific.

But recognizing such linkages depends upon what patient data are selected as inputs to the software. This selection of inputs is critical. Medicine imposes rigorous demands concerning the nature, scope and sequence of inputs to the decision making process. In complex cases those demands are rarely satisfied or even understood by practitioners. Software must therefore guide the selection of inputs. Otherwise, use of software in medical decision making may perpetuate some of the very cognitive constraints that need to be remedied. This issue, we believe, accounts for much of the dissatisfaction that clinicians have expressed concerning decision support software in medicine.⁽¹⁸⁾ ⁽¹⁹⁾ In medicine as in other fields, computerization alone is no solution to disorder in underlying work processes.

What, then, does orderly decision making require? Diagnostic decision making for a non-specific symptom such as fatigue should begin with a predefined workup based on the best medical knowledge � a comprehensive set of safe, simple, inexpensive findings that, in combination, best identify the diagnostic possibilities and discriminate among them. Similarly, treatment decisions must always begin with a predefined set of comprehensive, safe, simple, inexpensive findings that, in combination, best identify what treatments are available for the patient�s condition and the factors relevant to selecting among them (e.g. their side effects and interactions with the patient�s other conditions and treatments). Absent an emergency, extracting as much guidance as possible from numerous, readily available findings in combination creates a basis for intelligently deciding what further investigation is medically necessary and what diagnostic and treatment options are best. Such an approach we refer to as combinatorial analysis, to be distinguished from probabilistic reasoning and rigid, "if-then" algorithms.⁽²⁰⁾ Those other approaches are useful in limited contexts (such as certain emergency situations) but should not be the primary basis for coupling medical knowledge with the problems of unique patients.

The article does not address these distinctions, but it reveals that a probabilistic approach is accepted professional practice ("the clinician usually begins a diagnostic investigation by considering (and excluding) the most common diagnoses") (emphasis added). Indeed, some diagnostic software incorporates a probabilistic approach.⁽²¹⁾ Nothing in the article suggests that a combinatorial approach was understood or employed. On the contrary, the very structure of the article suggests idiosyncratic, discretionary data collection, erratic comprehension of the data, and premature judgments.

The article�s structure is to present limited patient data in stages to an expert clinician, who applies his clinical judgment at each stage. The first two paragraphs describe the girl's initial physical examination and history findings, including the fatigue, weight loss, hypotension, and multiple nevi. The article then, before presenting laboratory results, states the expert clinician�s initial analysis. He recognizes that the findings suggest an important organic illness, but he prematurely speculates on what the illness might be and does so in a fragmentary, non-systematic way (e.g., proposing illnesses inconsistent with the initial findings as one letter pointed out,⁽²²⁾ proposing hyperthyroidism rather than endocrine disorders in general and thereby overlooking Addison?s disease). The remainder of the article, and most of the letters to the editor, go on in a similar vein, with incomplete analysis and no recognition of how to investigate productively before launching into diagnostic speculation and serious medical intervention. What comes through especially clearly is the need for decision making to escape specialty boundaries. Addison's disease is an endocrine disorder, but it typically manifests itself with some combination of metabolic, gastrointestinal, dermatologic, cardiovascular and psychiatric symptoms and signs. Here, one of the consultants was an endocrinologist, but he missed the diagnosis in his own specialty. The authors were gastroenterologists, and their perceptions appeared to be influenced more by their specialty orientation than by their patient?s characteristics. The same appeared to be true of a letter to the editor criticizing the authors in this regard; the letter proposed an exceedingly indirect, complicated cardiovascular analysis, overlooking the initial findings that clearly suggested the correct diagnosis. As the article's authors observed in response, "clinicians work from short lists, and these lists vary from specialty to specialty."⁽¹⁴⁾

In short, medical decision making can become an intellectual Tower of Babel. Every doctor exercises his or her own clinical judgment about what data should be collected, every doctor is potentially confounded by limited comprehension of the data, and every doctor is subject to preconceptions and time constraints that further compromise care. These failings are inevitable without a systematic, efficient, combinatorial approach.

The reason for the absence of a combinatorial approach is the apparent difficulty of employing it. The difficulty is that combinatorial analysis requires collecting detailed patient data and integrating that data with the enormous medical knowledge base. The amount of detail to consider is so great that the most diligent, sophisticated doctors are unable to follow a combinatorial approach rigorously under the time constraints of real-world medical practice. When they attempt it, error and oversight are inevitable.

Some readers will question our generalizing in this way from the Addison�s disease case study. Because that disease is rare, the case study may seem unrepresentative and unimportant as an example of medical decision making. Skeptical readers might thus ask � does such a case provide any basis for concluding that software based on combinatorial analysis should be used in medical decision making routinely? Would doing so be cost-effective? This case involved diagnosis, not treatment � would solutions for diagnostic decision making be useful for difficult treatment decisions?

In fact, the case study is highly representative of everyday patient care in fundamental respects. First, although the diagnosis in this case turned out to be rare, the presenting symptoms were not. And a few routine symptoms and signs can present an enormous array of diagnostic possibilities. Faced with the task of Investigating those possibilities, the unaided human mind can easily fail.

Second, such cognitive failures frequently cause unnecessary suffering and expense. As the article acknowledges with admirable candor, when cases are difficult to comprehend, "vastly excessive testing and numerous attempts to treat putative diagnoses are the rule. We can be certain that in such instances some patients die . . . ."

Third, isolated pieces of statistical information (e.g. that Addison�s disease is rare in the general population) are normally misleading as a basis for individual patient care decisions. In contrast to public health and resource allocation decisions, such information is useful in patient care decisions only to the extent that we are ignorant of the patient�s unique combination of relevant individual characteristics.⁽²³⁾

Fourth, when we consider that an ordinary patient complaint may have hundreds of possible diagnoses and that relatively rare diagnoses such as Addison�s disease collectively account for a large proportion of the total, it becomes even clearer that rare diagnoses are no less worthy of organized diagnostic investigation than more common and obvious possibilities. Prioritizing diagnostic investigation based on what diagnoses are most common can, as the Addison�s disease case illustrates, cause critical delays, harm the patient and waste enormous amounts of money.

Fifth, these considerations apply to treatment as well as diagnostic decisions. Just as diagnostic decisions may require considering many possible causes for a symptom in light of the patient�s specific characteristics, so treatment decisions may require considering many possible therapeutic options in light of the patient�s characteristics and therapeutic needs. To illustrate, more than 100 therapeutic options exist for diabetes. Doctors can easily overlook some of these options and or information relevant to choosing among them.

For these reasons, the Addison�s disease case study is highly representative of normal patient care. Skeptics would be quite right to point out, however, that the case study is less representative in one respect. Medical decision making is sometimes far more difficult and uncertain than in the case study. Often, the correct diagnosis, or the correct treatment, or both, are inherently difficult to determine, no matter what information tools are employed or how much time is taken.

Does software based on a combinatorial approach have anything to offer in these more difficult cases? The answer is yes, because such cases are especially likely to require taking into account a large volume of detail. Chronic conditions such as diabetes and hypertension, for example, evolve over time in ways that cause enormous individual variation. Known diseases are sometimes unrecognizable in their early stages, because findings normally associated with the diseases may not appear initially. Sometimes, even after months of observation and tests, no known diagnosis matches the findings on a patient, or no standard treatment protocol turns out to be appropriate, because the patient�s condition departs from what is known to medicine. This can happen even if the patient has only one significant medical problem. More typical are patients with multiple problems and multiple treatments that interact in poorly understood ways.

These predicaments reflect a single underlying phenomenon � patient uniqueness. Different patients with the "same diagnosis," for example, can have radically different therapeutic needs, because the diagnosis explains only a few elements that they have in common (e.g. elevated blood sugar resulting from insulin deficiency) while the innumerable differences among them may be decisive for treatment decisions (e.g. the effects on blood sugar levels of exercise, emotion, infection, diet, obesity, other medical problems and various medications). The extent of physiological uniqueness will become all the more apparent as the human genome project advances and as we better understand the effects of developmental and environmental events, the body�s homeostatic mechanisms for self-regulation and repair, and the effects of complex medical interventions.

These features make it far more difficult to trace abnormalities to their origin and to decide which abnormalities should be regarded as primary in a particular patient. . . . we can expect that each illness will be a unique course of events, resulting from an evolving derangement of certain structures that will never be precisely reproduced. . . . This means that a patient's problems may be rather unlike the model problems that one finds treated in textbooks.⁽²⁴⁾

Patient uniqueness is familiar to experienced physicians, who regularly encounter the imperfect fit between accepted medical concepts and individual patients. "The reason is not that fundamentally new biological phenomena are being discovered every day in clinics and hospitals. It is that new combinations of known possibilities are appearing every day. The recombination of a finite number of preexisting pieces or possibilities can produce enormous diversity."⁽²⁴⁾ Yet, the decision making resources and analytical approaches that physicians depend on are insufficient for physicians to take this diversity into account.

Physiological uniqueness means that textbook descriptions of disease and treatment are radically incomplete, often taking the form of rough generalizations about large populations � statistical constructs to which few or no individuals in the population actually conform at any given point in the course of disease and treatment. "In truth," as Dr. Ken Bartholomew has observed, "the textbook case is so rare that everyone runs to look at it in the medical center when it is found."⁽²⁵⁾ And not only physiological uniqueness but also other aspects of individuality � the patient�s social and economic circumstances, emotional make-up, and personal values � may be critical to medical decision making.

Like textbook disease classifications, other conventional decision making resources cannot capture patient uniqueness. Those resources include simplistic practice guidelines, outcome studies of large populations (for "evidence-based" medicine), and judgments of consulting specialists. By focusing on only some of the potentially relevant variables, and requiring the human mind to integrate general knowledge with patient data, these resources tend to incorporate, rather than escape, the mind�s limitations. The result is "cookbook medicine," to use a term commonly applied by practitioners critical of managed care. What those practitioners overlook is the extent to which the same term applies to the care resulting from their unaided minds. Although the mind may be able to function adequately without external aids if given enough time, in real world conditions there is never enough time. We can no more function without information tools to aid the mind than we can function without microscopes to aid the eye.

When these dilemmas are squarely faced, their solution becomes quite obvious. New information tools can serve as an intellectual loom for weaving together patient data and medical knowledge from every specialty into the fabric of care. This perspective on medical decision making, however, is absent from the Addison�s disease case study. The authors simply assumed that we must rely on what the doctor can remember or figure out:

. . . we would be irresponsible if we failed to learn a lesson from this patient. . . . Addison�s disease, though rare, does occur and can be present for long periods without its classic manifestations [note this contradiction from the article�s prior recognition of "classic manifestations" early in the case]. Perhaps the only way to have made this diagnosis earlier would have been to appreciate that none of the diagnoses entertained by any of the physicians involved in the patient�s care explained all of the clinical findings. At that point a resourceful physician might have explored exhaustive lists of conditions that � no matter how rare and atypical � might be responsible. . . .⁽¹³⁾

Such conclusions demonstrate the hold over us that false ideals of physician decision making still have. A willingness to depart from those ideals and make use of properly designed software can empower the mind for complex decision making. Before describing the use of such software, however, we further analyze the decision making process.

C. Two stages of medical decision making

There are and can exist but two ways of investigating and discovering truth. The one hurries on rapidly from the senses and particulars to the most general axioms, and from them, as principles and their supposed indisputable truth, derives and discovers the intermediate axioms. This is the way now in use. The other constructs its axioms from the senses and particulars, by ascending continually and gradually, till it finally arrives at the most general axioms, which is the true but unattempted way. � Francis Bacon⁽²⁶⁾

In medicine as in any field, orderly decision making is usefully conceived in two stages: (1) recalling and processing information in order to identify options, and the pros and cons of each, for the problem at hand; and (2) choosing among the options by applying reasoned judgment and the values of the parties involved. The first stage of decision making is a process that we call knowledge coupling: coupling or matching patient-specific data (e.g. fatigue, hypotension, unusual pigmentation, gastrointestinal symptoms, and hyponatremia in combination) with general medical knowledge (that combination of findings is typically associated with Addison�s disease). Three requirements must be satisfied consistently in that coupling process for every patient problem :

(1) In retrieving general medical knowledge, all options relevant to a diagnostic or management problem, and all knowledge bearing on the suitability of the options, must be taken into account. Unusual options (rare conditions, non-standard treatments) and unusual details potentially relevant to determining the suitability of these options for unique patients must not be excluded from consideration merely on the basis of large population studies or a practitioner's own experience with other patients.

(2) In retrieving patient-specific data, the maximum feasible amount of readily available data (that which is safe, easy and inexpensive to acquire) known to be relevant to discriminating among the relevant options should be collected at the outset of care, in order to capture patient uniqueness and facilitate accurate, timely, cost-effective decision making.

(3) All linkages between (1) and (2), without error and omission, must be identified and presented by the coupling process, so that the known implications of the patient�s unique combination of relevant characteristics will be readily apparent to the caregiver and patient when they choose among the diagnostic or management options in the second stage of decision making.

This first stage of decision making is critical because it determines the information that the decision maker (usually a physician in the current system) takes into account. In the first stage of decision making, physicians do not readily acknowledge their own fallibility nor the havoc that can result from error. Indeed, physicians may not distinguish between the first and second stages. They tend to regard the two stages as inseparable, and the physician's knowledge, analytic capacities, and "clinical judgment" as central to both. And these cognitive inputs, they believe, are secrets of medicine�s successes, not of its failures. Physicians have faith in their capacity to recognize critical facts and relationships within the flood of data that complex cases involve. They view this capacity as highly sophisticated, requiring years of study and experience by the best minds in expensive institutions to master a large body of knowledge and acquire "clinical judgment." Indeed, the entire health care marketplace shares this view that the first stage of medical decision making requires the physician�s trained intellect.

Yet, the first stage of decision making need not require human recall, reasoning, values or clinical judgment. It requires simply information retrieval and processing. This is a mechanical process of linking one piece of data � an unexplained combination of symptoms, for example � with associated information � potential diagnoses suggested by that combination. The mind is not well suited to such information retrieval and processing when numerous variables are involved.

Why have we persisted in relying on the unaided mind for simple data processing (or "knowledge coupling," as we term the particular kind of data processing that medicine requires) when computers are so clearly superior? To explain this stagnation, it is illuminating to describe the origins of the knowledge coupling software described in the next section, and to examine a related issue in the history of science, both of which suggest that actual use of new tools and approaches is necessary to change understanding.

In the 1950s, work began on developing new standards for medical records. Adhering to scientific standards of data collection and clinical analysis is impossible in medicine with poorly maintained and illogically structured medical records, as discussed further below. Efforts to improve record keeping and other standards of scientific behavior in medical practice highlighted the need for more comprehensive data collection and more reliable retrieval of information (both general medical knowledge and patient data). For information retrieval, computers are clearly superior to human memory. Accordingly, during the 1970's work proceeded on development of a minicomputer-based, electronic medical record system designed to improve information retrieval. Solving the retrieval problem, however, revealed an even greater processing problem, that of integrating detailed patient data with general medical knowledge. Analyzing both problems revealed that the time constraints of real-world medical practice make it impossible for the unaided human mind to function with rigor. Physicians inevitably resort to dangerous shortcuts in retrieving and processing medical information. This recognition led to the development of knowledge coupling software designed to compensate for cognitive limitations in these functions. Development and use of knowledge coupling software in turn led to a deeper understanding of the extent of patient uniqueness and the fallibility of medical "knowledge."

The difficulty remaining is that the health care system as a whole has yet to go through this evolution in tools and perceptions. In contrast, a similar evolution has occurred in science, and enormous advances have resulted. At the birth of modern science Francis Bacon saw the issue clearly: "The sole cause and root of almost every defect in the sciences is this, that while we falsely admire and extol the powers of the human mind, we do not search for its real helps." He further explains: "The unassisted hand and the understanding left to itself possess little power. Effects are produced by means of instruments and helps, which the understanding requires no less than the hand . . . those that are applied to the mind prompt or protect the understanding."⁽²⁶⁾ In recent decades, cognitive psychologists such as Robyn Dawes, Paul Meehl and others have confirmed and refined Bacon's insights. As stated in a 1996 meta-analysis:

The human brain is a relatively inefficient device for noticing, selecting, categorizing, recording, retaining, retrieving and manipulating information for inferential purposes. Why should we be surprised at this? . . . The dazzling achievements of Western post-Galilean science are attributable not to our having any better brains than Aristotle or Aquinas, but to the scientific method of accumulating objective knowledge. A very few strict rules (e.g. don�t fake data, avoid parallax in reading a dial) but mostly rough guidelines about observing, sampling, recording, calculating and so forth sufficed to create this amazing social machine for producing valid knowledge. Scientists record observations at the time rather than rely on unaided memory. Precise instruments are substituted for the human eye, ear, nose and fingertips whenever these latter are unreliable. Powerful formalisms (trigonometry, calculus, probability theory, matrix algebra) are used to move from one set of numerical values to another.⁽²⁷⁾

It is important to understand the context in which these points are made and their relation to the two stages of decision making. The quoted statement (by professors William Grove and Paul Meehl) appears in a meta-analysis of studies comparing the effectiveness of subjective, impressionistic judgments with mechanical, algorithmic procedures in combining items of data (e.g. findings on a patient) to arrive at predictive conclusions (e.g. diagnoses). Forty years of studies in numerous fields, including medicine, lead Grove and Meehl to argue unequivocally that mechanical, reproducible methods for combining and assessing data (such as multiple regression, weighted sums of predictive factors, actuarial tables) are as good or better than the subjective, impressionistic judgments of expensive expert professionals. As one explanation for this phenomenon, Grove and Meehl point to the mind�s limitations as a device for retrieving and processing information.

Neither Grove and Meehl, nor those who view judgmental decisions more favorably,⁽²⁸⁾ ⁽²⁹⁾ clearly distinguish between the first and second stages of decision making. Our concern here is with the first stage. In the second stage, the difference between human judgment and mechanical alternatives to it are only marginal, and the effectiveness of both is frequently disappointing, based on the studies described by Grove and Meehl. We argue that both human judgment and mechanical decision procedures are unreliable in the second stage of decision making because relevant information is frequently overlooked in the first stage.

To improve the mind�s efficiency as a device for retrieving and processing complex information, scientists have embraced modern information tools. "The dominant trend in biomedical science and medical practice, as in every realm of science, is the increasing value and usage of computers. . . . The data so painstakingly extracted in past years are now, though progress in biomedicine, produced in such volumes as to require computers just to record them. The scientist spends more and more time using the computer to record, analyze, compare and display their data to extract knowledge," as recently observed by the National Institute of Health's Working Group on Biomedical Computing.⁽³⁰⁾ Unfortunately, medical practice lags far behind biomedical science in exploiting the power of computers. Although computer software is employed in clinical imaging and other devices for medical practice, software is not widely used to aid practitioners in retrieving and processing medical information.

Common experience tells us (and cognitive psychologists have shown in detail) that the mind�s normal mode of operation includes simplifying approaches that compensate for our mental limitations at information retrieval and processing.^{(31) (32) (33) (34) (35)} In medical decision making, these simplifying approaches can be summarized as follows. Doctors tend to generate diagnostic and treatment hypotheses prematurely; they tend to look for data that confirms rather than disconfirms their hypotheses; they consider only a portion of the potentially relevant hypotheses and data; they tend to over-interpret the limited information that they do consider, and they misapply population-based, statistical information to individual cases.⁽³⁶⁾

These characteristics are unavoidable so long as we permit the unaided human mind to be the primary source of cognitive inputs in the first stage of decision making. By contrast, incomplete or biased data collection and unwarranted interpretations of data are less likely to occur, and more obvious to others when they do occur, if caregivers and patients routinely use software that identifies what data should be collected and what interpretations of the data find support in the medical literature. Physicians and policymakers have yet to face the full implications of this fact, in part because studies of the potential for software to improve medical decision making have involved separate, discrete areas and a multiplicity of software packages ^(37)(38) that are not readily usable by practitioners.⁽³⁹⁾ This has obscured the reality that the first stage of all medical decision making involves a mechanical linking process for which software with a single, simple design can be employed.

D. The case study revisited � how improved cognitive inputs could have prevented diagnostic failure

It is a profoundly erroneous truism . . . that we should cultivate the habit of thinking about what we are doing. The precise opposite is the case. Civilization advances by extending the number of important operations which we can perform without thinking about them. Operations of thought are like cavalry charges in battle ? they are strictly limited in number, they require fresh horses, and must only be made at decisive moments. � Alfred North Whitehead⁽⁴⁰⁾

In order to illustrate the potential for improving cognitive inputs in the first stage of medical decision making, we describe how software could have been used to arrive at a diagnosis in the Addison�s disease case study. Our discussion is based on "knowledge coupling" software developed by one of the authors and his colleagues at PKC Corporation.⁽²⁰⁾ ⁽²⁴⁾ ^{(25)(41)
(42) (43) (44) (45) (46) (47)} We believe that this discussion is of general interest, because it is concerned not with the details of a specific software package or a specific decision making problem but rather with generic issues presented by any attempt to couple general knowledge with individual patient needs. In particular, our discussion is concerned with the nature, scope and sequence of inputs to any software used for this coupling process.

In the Addison�s disease case, one of the girl�s primary initial symptoms was excessive fatigue. A PKC software module ("Coupler"), which has been in use for approximately ten years, is designed to aid diagnosis of fatigue and similar symptoms (the Coupler is titled "Depressed Feelings, Apathy and Fatigue"). If used when the girl first sought care, this Coupler would have presented questions on more than 500 potential findings relating to more than 100 known causes (or other diagnostic associations) of the fatigue problem. These question sequences for history, physical examination and laboratory tests provide the basis for complete and systematic evaluation, without dependence on the physician�s memory or clinical judgment. (See Table II for a listing of other diagnostic, management and screening Couplers.)

TABLE II � Knowledge Coupling Software Modules ("Couplers") Currently Developed by PKC Corporation

Coupler Sets for Specialty Services Use

Screening Set For complete health history, mental health screening, wellness, physical exam, and laboratory screening

Senior Assessment and Disease Management Set To support care planning, case management, disease prevention, OASIS reporting, and patient education for senior and disabled populations

Occupational Health Set To equip providers to identify and manage work-related injuries, exposures, and disease, as well as to navigate the complex rule sets surrounding specific work environments

Behavioral Care Set For assessing potential behavioral manifestations of medical problems, supporting the determination of DSM-IV diagnosis, and guiding patient/provider strategies for disease management

Chronic Disease Management Set For managing such conditions as diabetes, COPD, asthma, depression, and hypertension

Problem Knowledge Couplers Currently Developed:

Acne Vulgaris - Disease Management
Acute Abdomen - Diagnostic *
Anemia - Diagnostic
Anesthesia: Preoperative Screening - Disease Management *
Anxiety Disorders - Disease Management
Anxiety, Panic and Phobias - Diagnostic
Asthma - Disease Management
Asthma Attack: Assessment and Management - Diagnostic/Disease Management *
Atrial Fibrillation- New Onset - Disease Management *
Benign Prostatic Hyperplasia - Disease Management
Carpal Tunnel Syndrome - Disease Management
Chest Pain I - Diagnostic/Disease Management*
Chest Pain II - Diagnostic *
Contraception - Disease Management
COPD - Chronic Obstructive Pulmonary Disease - Disease Management
Dementia - Disease Management
Depressed Feelings, Fatigue, Apathy - Diagnostic
Depression: Management and Guidance - Disease Management
Diabetes Mellitus - Disease Management
Diarrhea - Diagnostic
ECG Interpretation- Diagnostic *
Gallstones- Disease Management *
GERD - Gastroesophageal Reflux Disease - Disease Management
Headache - Diagnostic
Heart Failure Due to LV Dysfunction - Disease Management
Heart Sounds Abnormal (Initial Evaluation) - Diagnostic*
Hematuria - Diagnostic *
History: Comprehensive Screening for Mental Health Problems
History: Screening to discover Patient�s Problems - complete Health History
Hypercalcemia - Diagnostic *
Hyperlipidemia or Dyslipidemia - Disease Management
Hypertension - Diagnostic
Hypertension - Disease Management
Hypertension, Malignant and Other Hypertensive Crises - Disease Management
Itching of Unknown Origin (Pruritis) - Diagnostic
Jaundice - Diagnostic *
Jaundice in the First Year of Life - Diagnostic *
Kidney Stone (known): Type of Stone & Its Management - Diagnostic/Disease Management *
Knee Problem - Diagnostic
Knee Problem Due to Acute Trauma - Diagnostic
Laboratory Screening - Guidance regarding interpretation of indicated lab tests
Low Back, Buttock and/or Leg Pain, Acute - Diagnostic
Lung Cancer Suspected: Abnormal Chest Film Work-up - Diagnostic *
Male Erectile Dysfunction - Diagnostic
Male Erectile Dysfunction - Disease Management
Memory Loss and Confusion - Diagnostic
Menopause - Disease Management
Obesity - Disease Management
Obesity or Unexplained Weight Gain - Diagnostic *
Otitis Media with Effusion in Children - Disease Management *
Parkinson�s Disease - Disease Management
Periodic Health Evaluation -Yearly update of Patient Problem List
Personality Problems - Diagnostic *
Physical Exam: Screening to Discover Patient�s Problems
Prenatal Visit: Anticipatory Pediatric Guidance - Disease Management *
Pressure Ulcers - Disease Management *
Psychotic-like or Bizarre Behavior, Thinking, Perception - Diagnostic *
Rhinitis Chronic - Diagnostic *
Senior Assessment - Functional Assessment and Improved Management for Seniors
Shoulder Problem - Diagnostic*
Sleep Problems - Diagnostic *
Smoking Cessation - Disease Management
Substance Use Disorders - Diagnostic
Substance Use Disorders - Disease Management
Suicidal Thinking and Behavior - Disease Management*
Swallowing Problems (Dysphagia) - Diagnostic
Syncope (Fainting) - Diagnostic *
Tinnitus ( Noise in the Ear or Head ) - Diagnostic/Disease Management *
Upper Respiratory Complaints, Acute - Diagnostic
Urinary Incontinence - Diagnostic
Urinary Problems, Female (Dysuria, Urgency, etc.) - Diagnostic
Urticaria/ Angiodema - Diagnostic
Vaginal Bleeding Abnormal - Diagnostic*
Vertigo and Dizziness - Diagnostic
Vomiting Not Easily Explained In an Adult or Older Child - Diagnostic *
Vulvar and Vaginal Problems (Itch, Discharge, Pain) - Diagnostic
Wellness and Health: Assessment and Guidance - Complete Health Risk Assessment

* Beta version

Source: PKC Corporation. PKC Corporation�s Medical Content Development team maintains and expands Coupler offerings, with updates and new topics added every six months. One user has reported that the Couplers currently available cover approximately 85% of all presentations in his primary care practice.

Among the findings that this Coupler would have directed the user to check for are blood pressure, the serum sodium level and darkening of the skin or mucous membranes. Based on the initial findings described in the article, the user would have made "yes" or "not sure" entries on each of these findings. Any one of these findings would have been enough for the Coupler to suggest Addison�s disease as a possible diagnosis. In presenting that possibility, the Coupler would also have described a variety of other typical findings, including vomiting, abdominal pain, nausea, diarrhea, psychiatric symptoms, and others (the Coupler provides citations to the medical literature for this information). The Coupler also describes the testing sequence (plasma ACTH and cortisol determinations) needed to confirm Addison�s disease. Moreover, the other initial findings described in the article, if entered in the Coupler, would not have pointed strongly to other diagnoses, and the need for testing to confirm or rule out Addison�s disease would have been obvious.

Some readers will object that the standardization and comprehensiveness imposed by a combinatorial approach are not practical or cost-effective. It may seem prohibitive, for example, for a doctor to make more than 500 findings and consider more than 100 possible diagnoses for every patient who complains of fatigue or depression. On this view, only the doctor�s expert clinical judgment can adequately balance between cost and efficacy.

Such arguments overlook the inefficiencies in the traditional system and the capacity for information technology to dramatically increase efficiency at multiple levels. Consider the operations involved in conducting a taking a history, physical examination, selecting laboratory tests and analyzing the data obtained. Traditionally, a doctor is heavily involved in each of these steps (whether or not he has the time and skill to carry them out reliably). The doctor must further analyze all the findings to determine their implications and decide on the next steps. In doing so, doctors make very limited use of the medical literature because of time constraints.⁽⁴⁸⁾ ⁽⁴⁹⁾ The doctor dictates or writes down both the patient data and his analysis, which are transcribed and entered into the patient's medical record. A far more efficient approach, however, is to rely primarily on non-physician personnel (for many organizations this is long been an economic necessity), equipping them with knowledge coupling software that shows what data to obtain and documents the completeness of the process. The software then instantly couples the data with information carefully selected from the medical literature, generating diagnostic and management options, evidence from the patient for and against each option, guidance on further data to collect, and literature references. This output can be exported to electronic medical records or other software, and printed out for the patient and any providers or administrative personnel who need it.

For the medical history, patients themselves complete a detailed questionnaire (on a printout, personal computer or, eventually, a handheld device), consulting with family members or providers when they are uncertain how to answer. The physician or other provider then can review abnormal and "not sure" findings with the patient, making changes if needed. This approach to taking a history is far more efficient and effective than a patient interview. That verbal process may be wandering and unfocused; the busy doctor may get interrupted or not give the patient time to ponder responses; communication may be inhibited by personal or cultural incompatibility or by questions that the patient is embarrassed to discuss; the doctor may run out of time and decide not to bother with some items. When items are omitted in this way, others may be unable to distinguish between omission and negative responses, which can lead to erroneous decisions or duplication of effort.

Some practitioners have found that use of knowledge coupling software and non-physician personnel make it possible to do comprehensive workups several times faster; moreover, follow-up visits and administrative overhead are reduced.⁽⁵⁰⁾ ⁽⁵¹⁾ As Dr. Ken Bartholomew has written of how he practiced with knowledge coupling software in his clinic:

let me dispel the notion that they [couplers] are extremely time consuming if used properly. Certainly, the first few times using a coupler will be time consuming, but no more so than reading a textbook chapter on a given problem. . . . When I enter the examining room, I have a coupler that is largely done. With a good nurse, a large portion of the common physical findings can be entered in the computer and a note left if something is in question. The physician then rechecks any physical findings that are positive or questionable. . . . The physician in this setting enters the equation at a higher level of expertise and, instead of spending the whole day gathering mundane data, spends much more time reviewing the complexities of the cases that need that extra caution to the patients' benefit. Furthermore, I must admit that the practice of medicine in this setting is simply more fun. . . . By having this extra time to spend on more complex cases, the physician can then begin to use the couplers to function at a higher intellectual level than a busy practice usually affords."⁽⁵⁰⁾

For these reasons, a rigorous combinatorial approach to decision making is both more practical and more appealing than most practitioners initially recognize. Moreover, the amount of information taken into account with a combinatorial approach is not necessarily large (not all symptoms require considering 500+ findings or 100+ diagnoses), and even when it is, complexity or delay need not result. With a nonspecific symptom like fatigue or depression, the effect of taking into account all readily available information is usually to simplify:

[a combinatorial approach uses] the power of simple observations in the aggregate to restrict the range of possible solutions to our problems. When we know many things, even many simple things, about a patient, we are in a better position to see the full range of possible causes for the patient�s problems and to find combinations of those things that discriminate clearly between the possibilities. The more information we have about the patient, and the more effective our tools for combining and matching within that information are, the less need there is to split hairs over the value of each piece of data.⁽²⁴⁾

These points are not at all surprising. Experience in many fields shows that thoroughness and organization can reap unexpected harvests of productivity. Achieving this goal in medicine demands a combinatorial approach that physicians, left to their own devices, cannot be expected to provide.

The benefits of a combinatorial approach cannot be achieved with software that seeks to replicate the judgmental approaches of expert clinicians. These approaches frequently involve rapid hypothesis formation from limited data followed by testing for confirmation or rule-out (or by treatment and observation of its results). By jumbling together the first and second stages of decision making, these approaches are fraught with error, waste and risk to the patient. The most qualified expert can proceed down a dangerous or expensive blind alley when relevant information is not taken into account. The same is true of the most sophisticated "expert system" or "artificial intelligence" software. For the first stage of decision making, the judgmental approaches of expert clinicians are a constraint to avoid, not an ideal to emulate, in software design. A far more rigorous approach to the first stage of medical decision making must be enforced.

In contrast, a crucial benefit of knowledge coupling software designed to implement a combinatorial approach is that it does not try to replicate or displace judgment; it incorporates experts' scientific judgments in the software's content, and defers practitioners' judgment to the the second stage of decision making. In the first stage, knowledge coupling software organizes and presents potentially relevant information, while filtering out what is clearly irrelevant. Then in the second stage the patient and caregiver must apply judgment and values in acting on the information presented and on any other information they wish to consider.

Using a combinatorial, knowledge coupling approach to organize the first stage of medical decision making leads to two further areas of inquiry: (1) given that patient care involves many decisions over time for patients who frequently have multiple, interacting problems and multiple caregivers, how are we to organize those discrete decisions over time into a coordinated, total process of problem solving, and (2) whose judgment and values are to control in the second stage of decision making? We turn to the first question in the next section, and the second question in part II.

E. Medical records and medical problem solving

In some cases, patients have only one significant medical problem, and a clearly correct diagnosis or treatment or both are readily apparent or quickly identifiable with knowledge coupling tools. But many cases, consuming a disproportionate amount of health care resources, are not so simple. In difficult cases, knowledge coupling software will usually suggest multiple diagnostic or treatment options, each with some findings in favor and some against. The different options must be regarded as hypotheses to be tested against further findings, which the software helps select. Indeed, sometimes no plausible diagnostic or management options will emerge for some time, and further tests or observation will be the only acceptable option.

In complex cases of this kind, quality of care depends on an organized and explicit structure for problem solving, ensuring that relevant information is collected, considered and acted upon. The medical record is the primary tool for achieving that goal. The record is where health care workers set forth their data, hypotheses, plans, results, analyses and conclusions. From this perspective, the medical record is comparable to the notebooks and manuscripts of a scientist. A scientist�s credibility and professional standing depends upon maintaining such documents accurately, completely, and in accordance with accepted formats. The same should be true of medical records.

In addition to the intellectual difficulties just described, complex cases present communication difficulties for which properly structured and maintained records are the only solution. Medical care may go on for weeks or months or years, numerous variables must be carefully followed, and ongoing decisions must take into account what has happened in the past. Multiple caregivers are usually involved, and coordinating their actions over time becomes both difficult and crucially important. The patient�s own awareness of the course of disease and treatment is vital. The medical decisions made or the performance of the health care personnel involved may be the subject of later scrutiny for a variety of purposes. From this perspective, the medical record is analogous in function and importance to a corporation�s audited financial statements, or a commercial airliner�s maintenance records. Such documents must be prepared and rigorously maintained in accordance with detailed professional standards. Serious legal consequences can ensue for workers and institutions failing to abide by those record-keeping standards. The same should be true in medicine.

Yet, medical records are notorious for their incompleteness and variability in format and content. Their poor quality reflects disorder in decision making and problem solving. We have already discussed disorder in medical decision making. In the larger process of medical problem solving, disorder becomes apparent when one compares what physicians usually do with the four basic steps that problem solving of any kind requires: gathering data, formulating problems, developing plans for each problem, and carrying out or modifying the plans in accordance with subsequent feedback. When one considers specifically what these four steps require in medical problem solving, it becomes apparent that the medical record is pivotal.

A number of publications discuss these four steps of medical problem-solving, and the corresponding parts of the medical record, in detail.⁽²⁰⁾ ^{(52) (53) (54) (55) (56) (57)} The key element is problem-orientation � organizing all information (other than an initial database of screening information) around a complete list of patient problems. The basic components of complete records organized on this basis are shown in Table III.

Table III Components of the Problem-Oriented Medical Record

Record components	Problem solving steps
Defined database	Gathering an initial database of patient-specific information � predetermined history, physical examination and laboratory findings plus additional predetermined findings specific to investigating both the patient�s complaint(s) and significant abnormalities uncovered.
Problem list	Formulating a complete list of problems, based on the patient�s complaint(s) and the results of coupling the database findings with general medical knowledge.
Initial plans	Formulating initial plans for diagnosis or management of each problem on the problem list, based on coupling the problems with general medical knowledge. Each initial plan should explicitly reflect the basis and status of the problem, the disability associated with the problem, the goal to be achieved, the parameters to be followed, the treatments to be instituted, further investigative steps, and complications to watch for.
Progress notes	Follow-up actions for each problem, including monitoring the problems and responses to treatments, reassessing and revising the problem list and plans, and taking further diagnostic or therapeutic action if needed, all recorded in structured progress notes and flowsheets, each labeled by the problem to which it relates.

Physicians vary widely in the extent of the initial database they choose to gather; they often formulate problems as hypotheses rather than documented findings; problem lists are often incomplete or absent altogether; plans for diagnosis or treatment of a problem may be erroneous or premature because the problem is formulated incorrectly or interrelationships among different problems and procedures are not considered; plans for some problems may be neglected, or elaborate treatments may be undertaken when no treatment at all would be the better plan, the rationale for plans may be unstated; needed information may not be gathered or information buried in a source-oriented record may be overlooked; diagnostic and treatment hypotheses may not be modified when needed; communication among caregivers and the patient is difficult and may omit crucial information.

These and other compromises of quality are inevitable when caregivers are left to their own devices. In contrast, enforcing a problem-oriented approach to medical record keeping naturally brings rigor to the problem-solving process. As stated by Ian Lawson, "the POMR [problem-oriented medical record] . . . compels relationships and interdependencies as conditions of physician conduct. And now, through computer technology, it draws us into a thorough-going consistency, which is unfamiliar to most natures and threatening to our tolerated caprices."⁽⁵⁸⁾

Without corrective feedback loops, health care, like any complex activity, runs out of control. A problem-oriented structure promotes corrective feedback at multiple levels. For example:

� A complete problem list confronts the caregiver at every encounter with a total picture of the patient's particular combination of problems. This is a necessary safeguard against the tunnel vision of specialists and the fallible categories and generalizations of medical "knowledge." Interactions among multiple medical problems and their treatments, for example, are easily overlooked if a complete problem list is not maintained. Moreover, the discipline of formulating problems highlights discrepancies between physicians' hypotheses and supportable conclusions.

� The necessity for consistent, comprehensive data gathering becomes obvious to all when a complete problem list is required, because the completeness of the problem list is a function of the data collected. Yet, the need for a carefully defined and consistently collected initial database is still not recognized. The November 1997 AMA/HCFA Documentation Guidelines for Evaluation and Management Services, for example, provide: "The extent of history of present illness, review of systems and past, family and/or social history that is obtained and documented is dependent upon clinical judgment . . . " (p. 5, emphasis added), and recent proposed revisions (pp. 3, 5) continue this deficiency.⁽⁵⁹⁾

� Using the problem list to organize ongoing data about the patient's condition and medical procedures (in contrast to the traditional, source-oriented organizational scheme) permits rapid comprehension of voluminous records, whether in paper or electronic form, by caregivers, patients and third parties. "Few people understand the complexity and volume of data that need to be addressed each time a patient encounter takes place," as Ken Bartholomew writes. "The volume of data on a chronic patient becomes so large that it becomes unmanageable, and therefore lost, just as the volume of data in the medical literature is already unmanageable and lost to the average practitioner."⁽⁶⁰⁾ Records with a problem-oriented structure address this difficulty by placing all items of information in their logical context.

� Problem-oriented records give patients themselves a total picture of their condition and their progress, which is essential for them to develop the understanding and the behaviors that coping with illness often requires. As Ken Bartholomew has described in his practice, for example, when patients are regularly confronted with their own medical records documenting precisely how their lack of medical progress correlates with their lack of therapeutic compliance, they are more likely to take responsibility and less likely to blame the provider.⁽⁶¹⁾

� Population-based outcome comparisons are of limited use for an individual patient care decision unless the population data is drawn from intelligible, complete medical records. Such records make possible detailed analysis to identify specific subpopulations of individuals whose medical profiles are most comparable, and whose outcomes are thus most relevant, to the individual patient.

These illustrate but some of the ways in which properly designed medical records are necessary for rigor and feedback in medical practice. The broader point to understand is that problem-oriented medical records are needed to satisfy generic requirements of orderly problem solving. Those requirements, like combinatorial analysis, are elementary and inescapable. Yet they are rarely satisfied or even understood in medicine, because they impose a "thorough-going consistency" that is "unfamiliar to most natures and threatening to our tolerated caprices."

Some readers may fear that imposing consistency and comprehensiveness would expand the already intrusive powers wielded by doctors and health care institutions. It may seem, for example, that carefully investigating every complaint of fatigue or depression, or formulating a plan for every problem on a complete problem list, exacerbates what Ivan Illich called the medicalization of life.⁽⁶²⁾ Medicine's increasing ability to identify abnormalities that produce no current symptoms or suffering, to employ advanced treatments of uncertain worth, and to expose the manifold connections between physical, mental and social conditions, increases the potential for problematic interventions and wasted resources. Equally important, the expanding reach of medical science "removes a huge range of human experience from the realm of personal wisdom and individual understanding."⁽⁶³⁾ The answer to these concerns, however, as we discuss in parts II and IV, is to respect the patient's autonomy in the second stage of medical decision making. Patient autonomy helps create the checks and balances that medicine requires. Autonomous patients will employ information tools to gain individual understanding and to resist providers who would pursue elaborate medical interventions of dubious benefit. At the same time, those patients will demand an orderly, consistent, comprehensive approach to coupling their own needs with medical knowledge. They will recognize that such a standard of care must be enforced if they are to be protected, and if acceptable standards of quality and economy in medicine are ever to be achieved.

Key messages

� Health care organizations should focus not only on outcomes but also on clinical inputs, because outcomes can be understood and improved only when well-defined inputs are consistently applied.

� Cognitive inputs to medical decision making can be improved with habitual use of software tools that make it possible to comprehend patient uniqueness and avoid harmful, unnecessary trial and error.

� Software tools, if properly designed and habitually employed, permit combinatorial analysis of numerous, simple, inexpensive observations, tests and procedures, linking them with medical knowledge to identify all individually relevant options, and the pros and cons of each, for unique patients.

� The actions of caregivers and the medical knowledge they employ must be subject to effective feedback, which requires complete medical records with a problem-oriented structure.

� Just as new systems are needed to improve cognitive inputs to medical decision making, so new systems are needed to assure the skillful, reliable performance of manual tasks and procedures.

The T. J. Hooper, 60 F.2d 737, 740 (2d Cir. 1932).
LeGrand J. Competition, cooperation or control? Tales from the British National Health Service. Health Aff 1999;18(1):27-39. p. 31 ("Perhaps the most striking conclusion to arise from the evidence [on the effects of recent NHS reforms] is how little overall measurable change there seems to have been").
Vladeck B. Plenty of nothing � a report from the Medicare commission. N Engl J Med 1999;340:1503-1506 ("members of the commission could not even reach consensus among themselves as to what [Medicare�s] problems are, or in what priority they should be addressed").
Thorpe K. The rising number of uninsured workers: an approaching crisis in health care financing. Washington, D.C.: National Coalition on Health Care, 1997.
Chassin M, Galvin R. The urgent need to improve health care quality. JAMA 1998;280:1000-1005.
Chassin M, Assessing strategies for quality improvement. Health Aff 1997;16:151-161 (p. 156) ("At present, the notion that the market can stimulate real quality improvement is no more than a theory, because virtually all of the competition occurring in the marketplace revolves around price").
Shuster M, McGlynn E, Brook R. Why the quality of U.S. health care must be improved. National Coalition on Health Care, Washington, D.C., 1997. p. 13-14).
Doyle JC, Unnecessary hysterectomies: study of 6,248 operations in thirty-five hospitals during 1948. JAMA 1953;151:360-365.
Millenson M. "Miracle and wonder": the AMA embraces quality measurement, Health Aff 1997;16;3:183-94. p. 186.
Shekelle P, Kahan J, Bernstein S, Leape L, Camberg C, Park R. The reproducibility of a method to identify the overuse and underuse of medical procedures. N Engl J Med 1998;338:1888-1895 (for elective hysterectomies in seven managed care organizations, the proportion found inappropriate by different expert panels ranged from 24 to 52%).
Bernstein S, McGlynn E, Siu A, et al. The appropriateness of hysterectomy: a comparison of care in seven plans. JAMA 1993;269:2398-2402 (finding that 16% of hysterectomies in managed care organizations were performed for inappropriate reasons and that 25% were of questionable benefit).
Millay E. Upon this age, that never speaks its mind. In: Collected Sonnets. New York: Harper & Row, York, 1988. p. 140.
Keljo D, Squires R. Clinical problem-solving: just in time. N Engl J Med 1996; 334:46-48.
N Eng J Med [letters] 1996;334:1404-1405.
Wilson J, Foster D, Williams textbook of endocrinology. 8th ed. New York: Saunders, 1992. p. 522.
Arieff A, Defronzo R. Fluid electrolyte and acid base disorders. New York: Churchill Livingstone, 1985. p. 432.
Ibsen H, Clemmensen O. Eruptive nevi in Addison?s disease. Arch Dermatol 1990;126:1239-1240.
Kassirer J. A report card on computer-assisted diagnosis ? the grade: C. N Engl J Med. 1994;330:1824-1825.
Werner E, Webster G. Shugerman A. et al. Performance of four computer-based diagnostic systems. N Eng J Med 1994;330:1792-1796.
Weed LL et al. Knowledge Coupling: New Premises and Tools for Medical Care and Education. New York: Springer-Verlag, 1991. p. 12, 38, 53-54.
Hoffer E. N Engl J Med [letter] 1996;334:1404.
De Loos W. N Engl J Med [letter] 1996;334:1403.
Knowledge Coupling, ibid. note 20. p. 38, 54.
Weed CC. The philosophy, interpretation and use of problem-knowledge couplers [unpublished monograph]. Burlington (VT): PKC Corporation, 1982-1998. Available from: URL: http://www.pkc.com.
Bartholomew K. The perspective of a practitioner. In Weed LL, Knowledge Coupling: New Premises and Tools for Medical Care and Education. New York: Springer-Verlag, 1991. p. 240.
Bacon F. Novum Organum: Aphorisms Concerning the Interpretation of Nature and the Kingdom of Man, First Book, para. 19 (1620) (Montague trans.). Available from: URL: http://history.hanover.edu/texts/bacon/aphor.htm
Grove W, Meehl P. Comparative efficiency of informal (subjective, impressionistic) and formal (mechanical, algorithmic) prediction procedures: the clinical-statistical controversy. Psychology, Public Policy and Law 1996;2:293-323. p. 316.
Klein G. Sources of Power. Cambridge (MA): MIT Press, 1998.
Spengler P, et al. Clinical judgment research is finally analyzed by meta-analysis. Papers presented at symposium at American Psychological Association meeting, August 1998. (List of papers available from:URL: http://www.sciencenews.org/ sn_arc98/8_22_98/Fob3ref.htm.)
Working Group on Biomedical Computing, Advisory Committee to the Director, National Institutes of Health (US). The Biomedical Information Science and Technology Initiative.1999. Available from: URL: http://www.nih.gov/welcome/director/060399.htm.
Dawes R. Rational choice in an uncertain world. New York: Harcourt Brace Jovanovich, 1988.
Dawes, R, Faust D, Meehl P. Clinical versus actuarial judgment. Science 1989;243:1668-1674.
Dawes, R. The robust beauty of improper linear models in decision making. Am Psychol 1979;34:571-582.
Tversky A, Kahneman D. Judgment under uncertainty: heuristics and biases. Science 1974;185:1124-1131.
Wason P, Johnson-Laird P. Psychology of Reasoning. Cambridge (MA): Harvard U. Press, 1972.
Knowledge Coupling, ibid. note 20. pp. 8, 37-42, 212, 226-27.
Evans R, Pestotnik S, Classen D, et al. A computer-assisted management program for antibiotics and other antiinfective agents. N Engl J Med 1998.338:232-238.
Johnson M, Langton K, Haynes R, Mathieu A. Effects of computer-based clinical decision support systems of clinician performance and patient outcome. Ann Intern Med 1994;120:135-142.
Bartholomew K., ibid note 25. p. 258.
Whiteheald A. An introduction to mathematics. London: Williams & Norgate, 1911. p. 61. Quoted in Strachey C. ed. Systems analysis in programming, computers and computation readings from Scientific American. San Francisco (CA): W. H. Freeman & Co., 1971.
Burger C. The use of problem knowledge couplers in a primary care practice, Health Care Information Management 1997;II(4):13-26.
Weed LL New connections between medical knowledge and patient care, BMJ 1997;315:231-35
Yee W. Knowledge coupling: support for psychiatric decision making. In Miller M, Hammond K, Hile M. Mental health computing. New York: Springer Verlag; 1996.
Weed LL, Weed L. Reengineering medicine. Federation Bulletin:The Journal of Medical Licensure and Discipline, 1994;81:147-183.
Weed LL. Reengineering medicine: questions and answers. Federation Bulletin: The Journal of Medical Licensure and Discipline 1995;82:24-36.
Zimny N. Making information accessible and useful to practicing clinicians. Problem-knowledge coupling. Int J Technol Assess Health Care 1992;8:109-117.
Weed, LL. Problem-Knowledge Coupling. Mt. Sinai J Med 1985;52(2):94-98
Huth E. The underused medical literature, Ann Intern Med 1989;110:99-100.
Covell D, Uman G, Manning P. Information needs in office practice: are they being met? Ann Intern Med. 1985;103:596-599.
Bartholomew K, ibid. note 25, pp. 240-41.
Cross H. Computers in a private practice in Hampden, Maine. In Knowledge Coupling, ibid. note 20, p. 285.
Weed LL. Medical records that guide and teach [Parts I and II]; N Engl J Med 1968;278:593-600, 652-657. Reprinted in M.D. Computing 1993;10(2):1-15.
Weed LL. Medical Records, Medical Education and Patient Care. Cleveland: Case Western Reserve University Press, 1969.
Bjorn J, Cross H. The Problem-Oriented Private Practice of Medicine: A System for Comprehensive Health Care. Chicago: Modern Healthcare Press (McGraw-Hill), 1970.
Hurst W, Walker K, eds. The problem-oriented system. New York: MEDCOM Press, 1972.
Hurst W, Walker K, eds. Applying the problem-oriented system. New York: MEDCOM Press, 1973.
Van Horn E, Weed CC, A problem oriented approach to the computerized patient record, Burlington (VT): PKC Corporation, 1998. Available from: URL: http://www.pkc.com.
Lawson I. Comments on the POMR. In Driggs MF, ed. Problem-Directed and Medical Information Systems. New York: Inter-Continental Medical Book Corporation, 1974.
Available from: URL: http://www.ama-assn.org/emupdate/cpt.htm.
Bartholomew K, ibid. note 25. p. 273.
Barthlomew K, ibid.note 25. p. 265.
Illich I. Medical Nemesis. New York: Pantheon, 1976.
Goodwin, J. Geriatrics and the limits of modern medicine. New Eng J Med 1999;340:1283-1285.

Rapid Responses:

Read all Rapid Responses

Yes, but who will pay the IT costs: Matthew B Teolis
bmj.com, 2 Aug 2000 [Full text]

This Article

Read responses to this article

Services

Email this article to a friend

What's new

Latest blogs

Find BMJ on:

Services

Tools

Email to friend

Print this page

Online poll

Find out more >>

Resources

Rapid responses for this article

Yes, but who will pay the IT costs Matthew B Teolis

Print issues

Function	Problem area	Old approaches	New approaches
Medical decision making	Basis for decision making	The mind of the educated physician	Software tools and patient choices
	Structuring and recording the decision process	Broad discretion for physicians� "clinical judgment"; manual medical records varying in structure and completeness	Defined, explicit structure for physicians and others; complete electronic medical records with a problem-oriented structure
Execution of decisions	Individual performance	Educational credentialing and apprenticeships, certifying general knowledge and skill	Credentialing based on periodic demonstrations of actual competence in discrete skills
	System performance	Individual responsibility for errors	System improvements to support individual performance

Coupler Sets for Specialty Services	Use
Screening Set	For complete health history, mental health screening, wellness, physical exam, and laboratory screening
Senior Assessment and Disease Management Set	To support care planning, case management, disease prevention, OASIS reporting, and patient education for senior and disabled populations
Occupational Health Set	To equip providers to identify and manage work-related injuries, exposures, and disease, as well as to navigate the complex rule sets surrounding specific work environments
Behavioral Care Set	For assessing potential behavioral manifestations of medical problems, supporting the determination of DSM-IV diagnosis, and guiding patient/provider strategies for disease management
Chronic Disease Management Set	For managing such conditions as diabetes, COPD, asthma, depression, and hypertension