How do physicians make the necessary judgments and decisions when faced with diagnostic and therapeutic uncertainty and choices? Are there rules or algorithms by which clinical judgments and decisions are made? Certainly how a physician reasons has an impact on the types of judgments and decisions he or she makes. Beginning in the late eighteenth and early nineteenth centuries, clinical practitioners endeavored to make medical judgments and decisions more rational (Engelhardt, 1979). Their endeavors did produce fruit, especially in the twentieth century.
For the biomedical practitioner, clinical judgments and decisions are objective and modeled after the judging and decision making processes of the natural sciences. "The assumption that clinical reasoning is applied scientific reasoning," claims Mattingly, "underlies nearly all research on clinical reasoning in medical fields, and the informal perceptions of practicing health professionals" (1998, p. 275). For humanistic practitioners, clinical judgments and decisions reflect a subjective reasoning process, which includes the patient's personal information and values and which also involves the patient's narration of the illness experience.
Although there are profound differences between biomedical and humanistic or humane practitioners, the general outline of the process for clinical judgments and decisions are to some extent similar. That process, or "journey" as Engelhardt (1979) calls it, begins with collecting data and making observations and is followed by hypotheses formation and testing, after which judgments and decisions concerning the patient's disease state and the best way to proceed therapeutically must be made by both the physician and patient. The difference between biomedical and humane practitioners concerns the role, if any, of logic or intuition in the process of judging and deciding the best course of action, as detailed in the previous chapter. Often the debate revolves around whether the physician must adhere to strict guidelines or can utilize gut feelings. But as some commentators note, clinical judgments and decision making are complex notions and "in their rich and full sense are freighted with values, including ethical and moral values" (Engelhardt, 1979, p. xxii). In this chapter, the nature of clinical judgment, which is often considered informal in nature, is examined first, followed by the clinical decision making, which is generally modeled formally.
7.1 Clinical Judging
What is meant by judging and judgment, especially from an epistemological perspective? And specifically, what is a clinical or medical judgment? Generally, judgment involves an evaluation or assessment of evidence, data, or observations, in order to discern or decide a path of action, which is discussed in terms of decision making in the next section. According to Bernard Lonergan (1992), judgment is embedded in a cognitional scheme and is an answer to a reflective question, which results from a reflective insight.
In Lonergan's original cognitional scheme there are three levels of operations. The first is the level of presentations or experience. It is concerned with the data and observations of experience, i.e. with the given. The next level is that of intelligence or understanding, in which one has an insight into the intelligibility of the data and observations. This level involves the events associated with thinking and reasoning, as discussed in the last chapter. At this level, one answers questions of fact, e.g. who, what, when, etc. The final level is that of reflection on the insights into the intelligibility of data and observations. At this level, the reflective question, "Is it so?," is addressed. To answer that question, requires a reflective insight into the truth or falsity of the questions of fact. According to Lonergan, "judgment is the last act in the series that begins from presentations and advances through understandings and formulations ultimately to reach reflection and affirmation or denial" (1992, p. 301).
Between the question for reflection and the judgment that represents its answer is reflective understanding and insight. That reflective understanding or insight is the result of "marshalling and weighing the evidence," i.e. one comes to grasp "the sufficiency of the evidence." For Lonergan, "to pronounce judgment without that reflective grasp is merely to guess" (1992, p. 304). But how does one know when one grasps the evidence sufficiently to make a judgment. Lonergan proposes several possibilities (Tekippe, 1996).
The first is that there are no further relevant questions to be answered. Once sufficient insight into understanding is grasped, one is justified to make the appropriate judgment. Akin to this is one's expert knowledge into the problem at hand. As one continues to learn about the problem, one becomes more informed as to what constitutes proper and sufficient evidence to answer the question for reflection. Next, there is the satisfaction of one's intellectual curiosity. Humans are, by nature, creatures who, in principle, have an unrestricted or a pure desire to know. When such desire is sated, to pronounce judgment is then generally warranted. Finally, humans display a capacity (unless otherwise impaired) to determine wisely when the evidence is sufficient to justify a judgment.
Although Lonergan provides a precise and unambiguous analysis of general judgment, Engelhardt (1979) claims that the notion of clinical judgment is ambiguous. It can either refer (1) to the "capacity" to make judgments or discernments or to draw conclusions concerning the patient's disease state and to determine what steps must be taken therapeutically or (2) to the "experiential origins" of that capacity. For biomedical practitioners, the origins of clinical judgment are logical and include scientific reasoning. Generally the capacity for such judgments is based on rules and algorithms. Such is not the case for humanistic practitioners. "One finds physicians asserting," claims Engelhardt, "that one can make adequate clinical judgments only on the basis of actual experience, not simply on the basis of general principles of physiology, pharmacology, pathology, or even on the basis of a reconstruction of past clinical judgments" (1979, p. xii). In other words, the capacity for clinical judgment is not simply reducible to rules or algorithms. Rather, it depends upon a tacit or an intuitive dimension.
Consequently, there is a sharp divide between biomedical practitioners and humanistic or humane practitioners over on the origins of clinical judgment and the capacity by which judgment is made. On the one hand judgment reflects the outcome of objective scientific reasoning, while on the other hand it reflects the subjective outcome of intuitive reasoning. In this section, clinical judging is examined in terms of the objective and subjective dichotomy and the art and science dichotomy, as well as with respect to its tacit dimension and the role of phronetic and narrative reasoning. Finally, I discuss the notion of a good clinical judgment.
7.1.1 Objective or Subjective?
In a piercing analysis of the intuitive or subjective dimension of clinical judgment, Paul Meehl (1954) demonstrated that in terms of clinical prediction statistical or actuarial methods outperform intuitive or clinical methods. The statistical method involves comparing the aggregate of clinical data to an actuarial table to determine how best the patient responds to therapy. The clinical method forgoes any use of such tables. "Clinicians," observed Meehl, "often hold the view that no equation or table could possibly duplicate the rich experience of the sensitive worker" (1954, p. 26).
Meehl's analysis of the empirical data, however, revealed that "special powers" like intuition do not function effectively in the clinic. Only through statistical analysis could the efficacy or "clinical usefulness" of a therapeutic procedure be determined accurately. "Out of the welter of diverse cases, with mixed data and complex judgments, you simply cannot tell," according to Meehl, "whether your use of a procedure is paying off or not" (1954, p. 136). Most clinicians have interpreted Meehl's assessment as a crippling blow to subjective clinical judgment and as a clarion call to replace it with the objectivity of actuarial tables and other mathematical models (Baron, 1988; Katsikopoulos et al., 2007).'
The reaction to Meehl's critique, as well as to research on judgment by others, such as Egon Brunswick, Kenneth Hammond, and P.J. Hoffman, involved attempts to model clinical judgment and efforts to test the validity of those models empirically (Goldstein and Hogarth, 1997). Most models are based on a notion of human judgment as "a matter of combining pieces of information that are weighed according to their importance" (Doherty and Brehmer, 1997, p. 547). The simplest model to account for the combining and weighing of information and evidence is a linear one. This model also performs the best in terms of predicting a practitioner's diagnostic judgment. For example, Lewis Goldberg (1971), in a controlled study, demonstrated that the linear model accounts for Meehl's data on clinician's judgment for assessing patients' psychotic or neurotic states, as compared to other nonlinear models such as conjunctive, disjunctive, exponential, and logarithmic.
What makes the linear model so powerful is that it captures the "vicarious" nature of a patient's communication of symptoms and of the clinician's use of them in diagnostic judgment (Hammond, 1955).2 Although linear models are powerful, they are severely limited. Borrowing a term from mineralogy, Hammond (1955) claimed that such models are "paramorphic." According to Hoffman, "the mathematical description of judgment is inevitably incomplete, for there are other properties of judgment still undescribed, and it is not known how completely or how accurately the underlying process has been represented" (1960, p. 125). But, the paramorphic nature of models is not all bad since models do aid in "describing, predicting, and understanding human judgment" (Doherty and Brehmer, 1997, p. 546).
In an attempt to simulate clinical judgment, in order to enlist the aids of computers in medical practice, John Gedye embedded it in the clinical encounter between the patient and the physician. "A clinical encounter," according to Gedye, "is thus an occasion for the exercise of clinical judgment, and since it is generally accepted that this utilizes a clinician's finest sensitivities, it might seem that any attempt to formalize such an activity would be a move back into a world of inflexible concepts" (1979, p. 95). However, he recognized that inflexible concepts are often needed to provide a patient with an unambiguous assessment of his or her illness, as long as these concepts are "appropriate" for a patient's specific needs.
Gedye also argued that clinical judgments are to be made from arguments that are "hypergnostic," i.e. arguments in which the conclusions extend "beyond" the clinical data and observations.' What grounds this hypergnostic leap is the similarity between two cases that share a variety of features: "the solubility of the hypergnostic problem may depend on having, or finding, an appropriate representation of the data, appropriate in the sense that it manifests pertinent criteria of nearness" (1979, p. 110). Gedye warned, however, that not all clinical judgments are hypergnostic in nature and may require further analysis and research.
7.1.2 Art or Science?
The discussion over clinical judgment often takes the form of a debate over the art or science of clinical judgment. Eliot Sober criticized the debate over whether clinical judgment is an art or a science, by distinguishing four dichotomies in the debate that he claims are fictitious. The first is advocated by clinicians who view clinical judgment as an art: "The skilled clinician is capable of achieving an intuitive insight that is inherently non-logical" (1979, p. 30). Sober rejected this assertion, claiming that clinical judgment involves the same non-mysterious problem solving skills as any other professional discipline. The next dichotomy is that artful clinical judgment takes into consideration the patient's unique personal information and not simply the generalized features of the disease state. Again, Sober rejected this position. He contended that the patient's uniqueness is overplayed but that science, although abstracting from the unique, cannot represent the concrete world completely with its abstractions.
The third dichotomy is that the art of clinical judgment takes into consideration the patient's emotional state. Sober, however, argued that emotions can also function cognitively as a source of important information concerning the patient's illness. The final dichotomy is the distinction between art of clinical judgment concerned with the qualitative and its science with the quantitative. Again, he rejected this dichotomy. "Inferences [derived from clinical judgments] using purely qualitative concepts," according to Sober, "can be just as precise as the most finely honed mathematics" (1979, p. 36).
Sober advocates an informational approach to clinical judgment, composed of both logical and psychological features. "Clinical judgment," for Sober, "is to be understood as occurring within an information-processing system, which has as its input a specification of observed characteristics of the patient and perhaps some laboratory data, and has as output a differential diagnosis" (1979, p. 32). Clinical judgment, then, is a skill that involves both art and science.
Alvan Feinstein also advocated a moderate position with respect to the dichotomy of art and science, concerning clinical judgment. Although traditional attempts to define clinical judgment in terms of scientific rationality have failed because of the complexity of human observations and data and attendant decisions, Feinstein attempted to overcome this failure by distinguishing between the various kinds of observations and cognitive activities involved in clinical judgment. "By dividing the observational data into descriptions of disease, illness, and host, and by analyzing the therapeutic and the environmental decisions separately," claimed Feinstein, "clinicians can discern the ingredients of clinical judgment" (1967, p. 28).
The artful dimension of clinical data, according to Feinstein, is relegated to those observations pertaining to the description of the illness and the host and to environmental decisions. However, the scientific dimension of clinical judgment is consigned to those observations concerning the disease and to therapeutic decisions. "This aspect of clinical judgment," opined Feinstein, "is a product of the clinician's mind, of his cultivated intellect and knowledge" (1967, p. 29). Thus, he challenged clinicians to incorporate scientific methodology into clinical judgment. "Clinical medicine, therefore, like most other human activities" concluded Feinstein, "is an indivisible mixture of both art and science" (1967, p. 295).1
7.1.3 Tacit Dimension
Gilbert Goldman provided a robust defense of the tacit dimension for clinical judgment. To that end, Goldman defined clinical judgment as "the mental processes involved in all stages at which the clinician collects and interprets data; formulates a problem statement, confirms and refutes diagnostic hypotheses; considers, plans, and implements possible diagnostic and therapeutic options, tests, and interventions; and evaluates likelihoods and outcomes" (1990, p. 48). According to Goldman, the dominant view of clinical judgment is that it is based exclusively upon an explicit form of knowledge that can be reduced to rules, formal models, and computer simulation. However, he argued that this view has failed (Goldman, 1991). The reason is that there is a tacit dimension to clinical judgment. This dimension consists of "knowledge which is possessed and utilized on an implicit, or subsidiary, level without conscious awareness" (Goldman, 1990, p. 50).
Goldman gave the example for the tacit dimension of clinical judgment in terms of a surgeon who knows exactly how much force to exert when suturing. The tacit dimension of clinical judgment involves skills, which may be physical in nature, as with the example of the surgeon, or cognitive or mental in nature, as in clinical judgments. Importantly, the tacit dimension is complementary to the explicit dimension, in that it represents the "knowing how" that grounds the explicit dimension of "knowing what" (Goldman, 1990). According to Goldman, the tacit dimension consists of the "routines which complement the explicit rules of practice, which tell [the physician] which rules to employ when, and which case requires the use of which information" (1990, p. 53).
Although Michael Scriven (1979) acknowledged the importance of tacit or implicit reasoning in clinical judgment, he claimed that such judgment is not simply a matter of tacit knowing. In like manner, logical reasoning in terms of rules or algorithms is also important; but, again, clinical judgment is not reducible to logical reasoning. According to Scriven, "in clinical inference leading to clinical judgment, we operate from such rough guidelines and these cannot be adequately formulized either as statistical or as exact generalizations" (1979, p. 15). In other words, clinical judgment is neither an intuitive faculty nor a logical faculty; rather, it is a skill.
Scriven claimed that the skill needed for clinical judgment is based on logic different from traditional mathematics. This logic is one of "considerations" What Scriven meant by this is a logic that can combine the multifaceted dimensions of information required to make a clinical judgment. Besides relevant generalizations, his "logic of considerations" also incorporates "many relevant estimated values of variables." The result is an epistemology he called "the theory of weak knowledge." It is an epistemology that includes "possibilities and approximations" in its epistemic base. In conclusion, Scriven contended that although statistical or actuarial methods do outperform intuitive or clinical methods it is not necessarily the case that all such objective methods will prevail all the time.
7.1.4 Phronetic and Narrative Reasoning
Recently, Aristotle's notion of phronesis or practical reason has also been used to explicate and defend clinical judgment from a humanistic or humane perspective (Jonsen and Toulmin, 1988; Pellegrino and Thomasma, 1981a). For example, Kathryn Montgomery defines clinical judgment in terms of "the practical reasoning or phronesis that enables physicians to fit their knowledge and experience to the circumstances of each patient" (2006, p. 33). Montgomery contrasts this type of practical reasoning with that of scientific reasoning, for the latter is concerned with obtaining truths of a universal sort while the former is concerned with the truth of the individual patient presenting to the physician. Clinical judgment based on practical reasoning leads to the best course of action for a patient given the specific conditions for that patient and not for some statistical mean given some generic set of conditions.
Duff Waring (2000), however, challenges the claim that clinical judgment is the result of phronetic reasoning. Waring argues that it is best described as a result of techne, for the practice of clinical judgment as techne leads to the production of health. Aristotelian phronesis, on the other hand, is concerned with "living well in general." According to Waring, clinical judgment may be analogous to phronesis but it does exemplify it.
Humane practitioners also utilize narrative reasoning, when making clinical judgments. For example, Montgomery examines the role of narrative reasoning versus scientific reasoning in clinical judgment with respect to generalization and particularization. Although generalization, especially for epidemiological statistics, is important for the science of medicine, particularization, in terms of the patient's individual values and concerns, is critical for sound clinical judgment. "Understanding the particulars," asserts Montgomery, "despite the inexact relevance of biological science and statistical epidemiology to the circumstances of one person's illness, is medicine's chief moral and intellectual task" (2006, p. 86).
The particulars of a patient's illness are best determined though narrative reasoning, according to humane practitioners. Moreover, anecdotal case studies are not incidental for medical practice but essential. Thus, the individual patient is not peripheral but central to clinical judgment and to medical practice. And, for that patient what is important is what Montgomery calls the "individual cause," which addresses why the patient became sick in the first place. That cause is best determined through narrative, not scientific, clinical judgment.
7.1.5 Good Clinical Judgment
What makes for good clinical judgment? There are several criteria that have been proposed to determine such judgment. Arthur Elstein (1976), for example, identified several features of a good clinical judgment. The first is "affective sensitivity." "Sometimes good judgment," noted Elstein, "is said to be displayed when a physician is sensitive to the emotional needs of a patient as well as to the psychological and social problems that frequently arise in coping with a grave illness or as a consequence of certain therapies" (1976, p. 698). Good judgment, therefore, requires the physician to take into account more than simply the clinical data concerning the patient's physiological or pathological condition.
The second feature of a good judgment involves the physician's ability or capacity to evaluate competing principles, in order to determine which principle applies in a given case or if another principle should prevail. In other words, a physician, in order to display good judgment, may need to think outside the traditional clinical box. Elstein provided an example of a patient suffering from both congestive heart failure and significant blood loss. One requires removal of fluid, the other addition. "A physician with good judgment," according to Elstein, "knows how to reconcile these apparently competing demands" (1976, p. 698). The final feature of a good clinical judgment is an ability to select an adequate diagnostic hypothesis or therapeutic protocol. For example, a physician displays good judgment when presaging difficulties associated with different therapeutic protocols and ameliorating them for the patient.
Engelhardt (1981) proposed that good clinical judgment also involves the fewest costs or risks to the patient, both in terms of diagnosis and therapy, with respect to morbidity, pain and suffering, and financial expenditures. Clinical judgment consists of both a correct diagnosis of a "medical problem" and its resolution in terms of an appropriate therapeutic modality. "Good clinical judgment," for Engelhardt, "requires, then, the reliable weighting of the probable diagnostic significance of various clinical findings while taking into account the significance for the patient of various possible adverse outcomes" (1981, p. 314).
The foundation of good clinical judgment, according to Engelhard, is prudence, especially in terms of what a patient values vis-a-vis health and sickness. Such judgment assists a patient in negotiating "the geography" of medical problems and their solutions. Prudence allows both patient and physician to choose between various competing values. A good clinical judgment then is a complex process that results in the most appropriate clinical outcome for the patient. Finally, it is "a creative process in the sense of requiring changing responses, given different patient evaluations of the significance of such possible outcomes" (Engelhardt, 1981, p. 314).
Narrative reasoning, as already alluded to, is also considered essential for good clinical judgment. For example, Montgomery claims that the very features biomedicine denies in its attempts to be a science are the ones needed for sound clinical judgment and good medical practice. These features include "appreciation of the individual person and the anecdotal event, recognition of a person's pain, attention to feelings, an awareness of one's emotional life and participation in the lives of others, and knowledge of the provisional nature of clinical knowing" (Montgomery, 2006, p. 174). Narrative reasoning provides the best means for accessing this information. It is through the interpretation of the patient's illness story that physicians are best able to understand the suffering associated with illness and then to make the appropriate clinical judgment concerning what best to do therapeutically. Trisha Greenhalgh (1999) makes a similar claim for the role of narrative reasoning in clinical judgment. She argues for a "narrative-interpretive paradigm" to make sense of not only the objective clinical data but also the subjective data of the illness experience.'
7.2 Clinical Decision Making
Whereas judging pertains to evaluation of evidence, decision making involves action on that judgment. After weighing or judging the evidence, one then decides on the best course of action. One simply does not evaluate the evidence and then generally does nothing. Rather, judgment of evidence often calls forth some type of action based on that judgment. For example, to collect laboratory evidence on a patient's condition and then evaluate it leads to some type of therapeutic action. To stop short of deciding on an action after making a judgment fails to complete the full operations intelligence calls forth.
Lonergan (1979) revised his tripartite cognitional structure articulated in Insight to include a fourth level, the level of decision. Once one makes a judgment about the evidence, then a decisive action generally follows. Importantly such action is the level at which freedom occurs, a freedom that involves responsibility on the part of the self-conscious knower. For Lonergan, only through our decisions are we authenticated: "One has to have found out for oneself that one has to decide for oneself what one is to make of oneself; one has to have proved oneself equal to that moment of existential decision; and one has to have kept on proving it in all subsequent decisions, if one is to be an authentic human person" (1979, p. 121).
Although decision making in general has an important existential dimension, clinical decision making is founded on more formal decision analysis procedures, which are examined in the first part of this section. I then look at various decision models that have been proposed for clinical decision making, followed by a clinical example illustrating clinical decision making. Finally, the procedure of tree pruning in order to make decision making manageable is examined, concluding with a discussion of the advantages and disadvantages of applying formal decision analysis to the clinic.
7.2.1 Decision Analysis
Whereas clinical judgment depends on implicit or tacit dimensions of human cognitive and emotional resources, clinical decision making involves more formal strategies such as flow charts and algorithms, especially assisted by computer technology. In other words, whereas clinical judgment is concerned with questions of understanding pertaining to the patient's illness and whether that understanding is accurate based on the evaluation of the clinical evidence, clinical decision making is concerned with the decisions about what action to take and whether it is the best one to take for the patient. The questions that animate decision making are: "How do people decide on a course of action? How do people choose what to do next, especially in the face of uncertain consequences and conflicting goals?" (Goldstein and Hogarth, 1997, p. 4). These questions have stimulated a vast literature on medical decision making, in an attempt to answer them.
"Sound clinical decisions," according to Jerome Kassirer and colleagues, "depend upon the integration of a variety of facts regarding a patient's condition with an extensive store of medical knowledge" (Kassirer et al., 1988, p. 212). This general approach is divided into a number of steps. For example, David Ransohoff and Alvan Feinstein (1976) identified five of them. The first is precise articulation of the clinical problem, often in a hypothetical format, followed by construction of a mathematical model of it. The form of the model is generally in terms of a decision tree, composed of branches connected by decision and chance nodes. The next step is to assign objective or subjective probabilities to uncertain events within the decision tree. Peter Doubilet and Barbara McNeil (1988) divided these probabilities into objective probability values, those that are based on previous evidence, and subjective probability values, those that depend on the physician's expert opinion or judgment. The third step is to assign a "utility" value for each expected outcome. These values are often based upon the personal value or preferences of either the patient or physician. The next step is to determine the expected value for each branch of the decision tree, by calculating the product between the probability and utility values for each branch of the decision tree. The final step is to choose the branch with the highest expected utility.
The goal of formal decision analysis, then, is to maximize the expected value of a decision. It is important to note that in humanistic models it is imperative to factor into a decision the patient's values or preferences concerning the outcome. Even though one branch of a decision tree yields the highest expected value, it may be rejected because of the patient's values. For example, a clinical practitioner may assign a utility of 0 to death as an outcome, while the patient may not. Moreover, Doubilet and McNeil (1988) and other advocates of decision analysis have added an additional step of sensitivity analysis, which involves altering systematically assumptions and values within the first three steps to determine how sensitive a decision is to variation of these assumptions and values. This step is important since clinical decisions are based on uncertainty, and physicians are unlikely to trust their clinical judgment unless the decision based on this formal style of analysis demonstrates that it can account for uncertainty.
7.2.2 Decision Models
Within the last several decades, a variety of models have been proposed to account for clinical decision making. Deborah Zarin and Stephen Pauker (1984) provided a general scheme for most models. In their scheme, the first three steps of decision making correspond to the following inputs: structuring the problem in terms of a decision tree, probabilities or likelihoods of outcome, and values or utilities of outcome. These three inputs result in a decision and consequent action via an integrative process. Zarin and Pauker then identified four possible types of models, which "differ from one another in (1) which of the two participants (doctor or patient) is the source of each input, and (2) the source of the integrative process that is used" (1984, pp. 185-186).
The first model is the classic or traditional paternalistic model, in which the physician is the source for all the inputs and integrative process. The next two models incorporate the patient in an effort to satisfy the doctrine of informed consent. In the first of these models, the physician informs the patient of the inputs and possible outcomes; but, the patient remains a passive agent and the physician is still the source for the integrative process. In the next model, the physician informs the patient but it is the patient who decides what inputs to use and who is the source of the integrative process. In the final model, the physician is responsible for the first two inputs, the structure of the decision tree and probability of outcomes, and also is responsible for informing the patient about them. The patient is then responsible for the third input or the value or utility of the outcome. The physician is the source of the integrative process. For Zarin and Pauker, the last model is the best for clinical decision making since it involves the expert knowledge of both the patient and physician.
7.2.3 Example
Jerome Kassirer (1976) has provided a superb example of clinical decision making, based on an actual clinical case. The patient was a twenty-four year old female, who had both kidneys removed several years earlier because of bilateral hypernephromas. Recently, she received a kidney transplant, underwent a splenectomy, and was treated for Klebsiella sepsis and pneumonia. She was admitted to the hospital because of vomiting and diarrhea, and she was found to have a fever (104.2°F) and rales in the left lung. Her symptoms worsened, and "she had severe left upper quadrant abdominal pain radiating to the left shoulder, generalized abdominal tenderness, diminished bowel sounds, splinting of the left chest, and poor movement of the left diaphragm" (Kassirer, 1976, p. 156). Her white blood cell count was 8,900. The initial diagnosis was subdiaphragmatic or subphrenic abscess, which is an accumulation of purulent exudates or pus below the diaphragm.
Although the diagnosis was uncertain the clinical decision facing the attending physicians was whether to operate or not, in order to resolve the abscess by draining the pus. The decision tree contained two main branches: the first represented surgery, the other not. At the time of the uncertain diagnosis, the probability the patient was suffering from a subphrenic abscess that could be resolved through surgery was 0.3. According to Kassirer, this probability meant that "we have estimated that 30% of patients with a clinical picture comparable to that shown by this patient on this date would have a subphrenic abscess and 70% would not" (1976, pp. 157-158).
Both main branches also bifurcated, with the first branch representing a surgically correctable abscess and the other a non-surgically correctable one. The probabilities for resolving the abscess were based on evidence published in the literature. In order to determine the best decision, the utilities were next calculated. The best outcome, a spontaneously resolved abscess by non-surgical protocol, was assigned arbitrary units of 100, while the worst, death, 0 units. Other outcomes ranged between these two values. Based on these utilities, the expected value for surgery was 62.5 units, while for non-surgery 81.1 units.
The clinical decision made in this case was to treat the patient non-surgically with antibiotics and fluids, along with nasogastric suction. Although the patient improved initially, after several days her symptoms became worse. As Kassirer narrates, "the new data available from the evolving clinical course, the change [increase] in white cell count, the results of echography, scan, and plain film markedly increased the probability of a surgically accessible lesion" (1976, p. 159). The probabilities were revised to reflect the changing clinical picture, even though there was no precedent within the published literature. Based on the same utilities as before, the outcome of the new decision tree differed from the original. Now the expected value of surgically resolving the abscess was 38.9 units, while non-surgical were 25.9 units. Even though surgery was the best decision, Kassirer points out that it is not without serious risks given the deteriorating condition of the patient.
7.2.4 Pruning Decision Trees
Clinical decision trees can become quite large and complicated (Kassirer, 1976). In response, physicians and patients may only focus on a sub-branch of it and ignore others. Consequently these trees are often pruned by removing sub-branches, through the physician's clinical judgment. "Branches can be pruned," claims Kassirer, "only if it is obvious from inspection that the probability and utility of the outcome are such that their combination will yield a value that will contribute little or not at all to the expected value of the outcome" (1976, p. 161).
Even though the pruning process is often carried out at an intuitive or instinctive level, there are principles that can be used to make pruning more logical in nature (Schwartz et al., 1973). The main principle concerns the degree of the probabilities and risks. "If both the probability of a given event and the risks associated with it are extremely high," claim Schwartz and colleagues, "the branch cannot be pruned. In contrast, if both the probability and the risks are low, branches can be pruned with impunity. The decision to prune or not to prune," they add, "when there is a low probability but a relatively high risk is more difficult" (Schwartz et al., 1973, pp. 461-463). Besides the probabilities and risks, the values or utilities associated with the branches, especially from the patient's perspective, must also be factored into whether to prune or not.
7.2.5 Advantages of Decision Analysis
Given the level of uncertainty in medicine and the decisions that must be made often on incomplete information, decision analysis provides several advantages for making clinical decisions. "Advocates," according to Stephen Eraker and Peter Polister, "have claimed that decision analysis enhances effective decision making by providing for logical, systematic analysis and by prescribing a course of action that will conform most fully to the decision maker's own goals, expectations, and values" (1988, p. 380).
Specifically, Eraker and Polister identified three advantages to decision analysis for clinical decision making. The first is that decision analysis is explicit in terms of its overall structuring of the clinical problem, especially with respect to the formation of a decision tree. "With the decision analysis framework," observed Eraker and Polister, "one can identify the location, extent, and importance of any areas of disagreement, and ascertain if any such disagreements have a significant impact on the indicated decision" (1988, p. 382). The next advantage is the quantitative nature of decision analysis, in terms of the probabilities and values. Such quantification provides a more objective means for evaluating the various clinical decisions. The final advantage is the prescriptive nature of decision analysis, which provides the best option in terms of what diagnosis to make or therapy to follow.
7.2.6 Criticisms of Decision Analysis
Although decision analysis continues to influence diagnostic and therapeutic decisions, there have been several criticisms levied against it. In comments on Kassirer's 1976 paper, for example, Ransohoff and Feinstein (1976) identified several problems with the strategies of decision analysis. The first is that the decision tree must include all the possible outcomes and actions or it will distort the clinical problem. "If these additional courses of action are possible and reasonable but are not considered in a decision analysis," according to Ransohoff and Feinstein, "then the structure [of the decision tree] is unsatisfactory because the problem has not been completely evaluated and the results may therefore be misleading" (1976, p. 166). The next set of problems concerns the estimation of the probabilities for the various outcome branches. Most probabilities in the literature may not be appropriate for the particular patient under treatment and that quantifying such probabilities may be difficult at best for the physician (Ransohoff and Feinstein, 1976).
The final set of problems revolves around assigning utility values. The first problem is "that many important outcome values are intangible and are therefore not easily measured" (Ransohoff and Feinstein, 1976, p. 166). The second problem is the comparison of possible outcomes, which have different attributes and require different scales for measurement. "This double task of converting intangible and multi-attribute outcomes into meaningful numbers," according to Ransohoff and Feinstein, "creates a major difficulty that is inherent in decision analysis and that cannot be managed readily if at all" (1976, p. 167). The final problem is who determines the utility values. Should it be the patient, the physician, hospital or HMO administrators, insurance company executives, or society at large such as politicians? Each of these would most likely assign a different utility value to a particular outcome (Ransohoff and Feinstein, 1976).
There are other problems with the quantitative approach of decision analysis. For example, Patrick Croskerry points out that clinical decision making is a complex process: "there are often too many variables or unknowns in the clinical situation, too many ethical and financial restrictions, or too many other resource limitations to even allow a simple quantitative approach to guide each clinical decision" (2005, p. R5). Logical rules associated with decision analysis cannot capture the complexity of many clinical decisions.
Croskerry (2005) also identified several hard wiring problems with the quantitative nature of decision analysis. The first is instinctive or behavioral in nature. The reasoning process that underlies decision making is adaptive in nature and reflects evolutionary pressures. Thus, much of human decision making depends upon the hard wiring selected through natural selection. In addition, personality and gender also influence clinical decision making. There are various styles of decision making that reflect a clinician's personality or gender. For example, anesthesiologists are by lot withdrawn compared to surgeons and may acquiescence to a surgeon's clinical decision concerning an operation (Croskerry, 2005).
Heuristics and biases also play an important role in clinical decision making (Croskerry, 2005; Tversky and Kahneman, 1974). "A variety of studies in the clinical setting," according to Croskerry, "have repeatedly demonstrated the importance of heuristics and biases in information processing and establishing a diagnosis" (2005, p. R3). Heuristics are the rules of thumb that permit a clinician to include or factor intuitions into a decision. For Amos Tversky and Daniel Kahneman, "heuristic principles... reduce the complex tasks of assessing probabilities and predicting values to simpler judgmental operations" (1974, p. 1124). However, there are a number of biases, around forty in all at the time of Croskerry's article, which can distort a clinical decision. These biases include, for example, ascertainment bias, ego bias, gender bias, outcome bias, and overconfidence bias. Given the complexity of clinical decision making, Croskerry concludes that "one approach does not fit all... There will always be a gradient of decision-making that parallels the degree of uncertainty associated with the wide variety of patient conditions, and which are to some extent discipline-specific" (2005, p. R6).
7.3 Summary
Clinical judgment and decision making are important epistemological components of both the biomedical and humanistic or humane models of medical practice. For biomedical practitioners, clinical judgment and decision making are based on scientific reasoning. This often leads to a paternalistic position for physicians, who find it too difficult or too time consuming to translate the technical dimension of medical language and concepts that under gird clinical judgment and decision making into language and concepts the patient can understand in order to participate in the judging and decision making processes. Of course, this paternalism not only plays an important role in the origination of the quality-of-care crisis but also exacerbates it.
Humane practitioners, on the other hand, endeavor to include the patient as an active agent in the clinical judging and decision making processes. By including the patient in these processes, the physician and patient can communicate more effectively. Narrative-based medicine has been championed as a means for promoting more effective communication that leads to an enhanced quality-of-care.