Chapter 2: Diagnostic Hypothesis Generation

Hypotheses and Cues

( Case 1, Case 2 , Case 3– Case 4, Case 16)

In our daily life, we are constantly generating hypotheses about our environment—about our visual images, how the physical world operates, our expectations of events, and our perceptions of people. These hypotheses provide a framework for interpreting all of our unstructured experiences. Diagnosis, a special case of unstructured problem solving, is initiated when a physician evokes, formulates, or triggers one or more hypotheses from a set of cues that emanate from an encounter with a patient.[ 18], [ 19], [ 20], [ 21]–[ 22 ] The cues that initiate hypothesis generation are varied. Sometimes a single symptom, such as dysuria, or a single physical finding, such as prominent facial bones, triggers a diagnostic possibility.

At other times, even a single laboratory result, such as an elevated serum calcium concentration, does the same. Most often, however, the cues are multiple: The patient’s age, sex, race, appearance, and presenting complaints constitute a familiar set. We formulate hypotheses at first contact, and we continue to evoke new hypotheses as long as we fail to satisfy ourselves that we have the “right answer.” Although hypothesis generation usually is the first step in diagnosis, it continues as we refine our existing diagnostic hypotheses and eliminate those hypotheses that are no longer tenable.

Diagnostic hypotheses take on many forms and lie on a spectrum from quite general to highly specific. Forms include disease entities, syndromes, involvement of an organ system, or even such notions such as “healthy,” “sick,” or “desperately ill.” Along the scale of specificity, hypotheses might range from a vague notion such as infection, to more specific entities such as gram-negative sepsis, to highly specific disorders such as meningococcal meningitis.

The Cognitive Basis of Hypothesis Generation

( Case 1, Case 3, Case 6, Case 23, Case 57, Case 63)

The process of hypothesis generation is best understood in the framework of modern cognitive science, which holds that the brain is an information processor that manipulates semantically meaningful “chunks” or packets of information. Such information chunks are represented in memory, but theories conflict about the nature of their storage. No matter how the information is represented, access to it evokes or generates a hypothesis about the state of a patient.

Hypotheses are generated rapidly—probably tentatively at first as candidates for acceptable hypotheses, and then, if they are consistent with existing data, they are accepted as plausible explanations for a finding or a set of findings. Quite likely, only a small number of hypotheses remain active at any given time. Given the limited ability of short-term memory to manipulate only 5 to 10 items at a given time, one can presume that this memory restriction also pertains to diagnostic hypotheses.[ 23],[ 24] If so, many hypotheses must be quite evanescent as others take their place, even though discarded hypotheses can and do re-emerge at a later stage in the process.

Some of the factors known to be important in the generation of diagnostic hypotheses include disease prevalence, heuristics (rules of thumb), and the gravity or seriousness of a patient’s condition. Triggering hypotheses according to a disease’s (or condition’s) prevalence presumably is an optimal approach, but it is uneconomical as a cognitive function because it requires considerable memory storage and processing, including checking for consistency against available clinical data.[ 25],[ 26] Instead of this cumbersome approach, we often rely on heuristics to evoke hypotheses.[ 27] One commonly used short-cut is designated the representativeness heuristic, an approach that relies on the resemblance of a set of findings to those of some well-defined clinical entity.

The findings of simultaneous cough, dyspnea, and travel to California might trigger the hypothesis “coccidioidomycosis,” for example, even though the prevalence of other diseases that cause both symptoms is far greater than that of the fungal infection. Another commonly used short-cut is the availability heuristic.[ 27],[ 28] This approach is a function of familiarity with a given clinical entity, usually because a certain pattern of findings evokes a readily recallable, particularly striking clinical entity.[ 29] The triggering of the hypothesis “pheochromocytoma” in response to the finding of a sudden, severe increase in blood pressure is such an example. As with the representative heuristic, however, there is no guarantee that a hypothesis evoked by the availability heuristic accurately reflects disease prevalence.

Still another short-cut used in hypothesis generation is related to the physician’s ever-present vigilance for life-threatening manifestations or complications of a disease. Repeatedly, physicians engaged in the diagnostic process change from generating hypotheses based on any of the mechanisms described earlier and instead evoke hypotheses for these diagnostic imperatives when early diagnosis and treatment is critically important for a patient’s well-being.

In the midst of the process of generating hypotheses based on prevalence, representativeness, or availability, physicians often evoke hypotheses that identify life-threatening manifestations or complications. Such hypotheses (sepsis, shock, pulmonary edema, acute myocardial infarction, hyperkalemia) may be generated without regard to prevalence, but they focus on the value of alertness to serious events while the “routine” part of the diagnostic process is underway. This type of medical rule of thumb may be merely a special case of known heuristic mechanisms such as availability.

Hypotheses as a Context

( Case 7, Case 8)

Given that maximum uncertainty characterizes the initial state of a diagnostic encounter, hypotheses form an essential function: They frame, or constrain, a patient’s problem and provide a context (or problem space) for further diagnostic reasoning and exploration.[ 30],[ 31] Each diagnostic hypothesis evokes a template of possible clinical findings against which a given patient’s findings can be compared.

The diagnostic hypothesis “nephrotic syndrome,” for example, mandates the presence of heavy proteinuria, typically includes hypoalbuminemia, edema, and hyperlipidemia, and encompasses an exceptionally large array of syndrome characteristics that include predisposing factors (diabetes mellitus, amyloidosis, systemic lupus erythematosus), short-term complications (venous thrombosis), long-term complications (accelerated atherosclerosis), pathophysiologic associations (sodium intake and edema formation), and histopathologic correlations (“spikes” on silver stain in one of the cases—membranous nephropathy). Thus, when the nephrotic syndrome becomes a hypothesis, its many characteristics become a framework against which a patient’s findings are assessed. Within this framework, or context, new data are gathered and assessed and hypotheses are preserved, rejected, or refined.

The value of the context lies in its capacity to guide the subsequent diagnostic process.[ 32 ] The context helps the physician to formulate appropriate questions as he or she takes a history of the present illness, directs certain specific aspects of the physical examination, and identifies tests that might provide additional relevant clinical data. Evidence suggests that physicians do not simply gather data without regard to diagnostic hypotheses, and that they do not simply accumulate facts until a diagnosis becomes evident. Rather, they gather relevant data within a defined context. The context serves as a guide for predicting which information might be useful to gather, which tests might be helpful, and which diagnostic procedures deserve further attention.

Expertise and Error

( Case 3, Case 9, Case 54, Case 58, Case 66)

Clinical experience and expertise clearly enhance the quality of hypotheses generated. Knowledge of the various clusters of cues that should trigger certain hypotheses and knowledge of the characteristics of diseases and syndromes that become the context for further diagnostic resolution facilitate the process of hypothesis generation.[ 19] “Book knowledge” is insufficient for optimal hypothesis generation, in part because diseases and syndromes vary far more in their attributes (combinations of clinical findings at onset, clinical course) than those characterized in “classic” textbook descriptions. Indeed, experience with one patient after another with a given disease or syndrome produces the enriched model of a disease or syndrome against which we measure new cases.

The process of hypothesis generation, however, is imperfect. When a patient’s disease is common and its manifestations are typical, when a patient’s clinical findings are representative of a certain disease, and when one or more striking clinical features point toward a specific diagnostic entity, the correct diagnosis often emerges quickly. Neither disease prevalence nor the heuristic solutions described before guarantee that the correct diagnosis will be generated initially, nor do they guarantee that the correct hypothesis will ever be evoked.

Both rare diseases and common diseases with atypical manifestations can be overlooked, and perceptual errors (e.g., failure to recognize that a patient has the classic physical features of acromegaly) can lead to faulty or insufficient hypothesis triggering. No special reasoning skills will suffice to trigger diagnostic hypotheses if the physician does not have sufficient knowledge about disease entities or about the full range of expected manifestations of these entities. A lack of either makes hypothesis generation at best faulty and at worst totally lacking.