Diagnostic thinking

By Marcello Cherchi, MD PhD

The first and most important step in treating a disease is figuring out what the disease is. In many cases it is not clear what disease is causing a patient’s symptoms, and in such cases we may temporize by attempting to treat the symptoms (sometimes referred to as “symptomatic management”). But if one can figure out what disease is at issue, then the likelihood of successfully managing that disease increases.

A physician’s task of securing a diagnosis can range in difficulty from trivial to extraordinarily challenging. While the public often entertains the notion that physicians can arrive at a diagnosis with something approaching the certainty of mathematics or physics, the reality of course is that biology, including the pathobiology of disease in humans, is a very messy and imprecise affair.

A practitioner uses a variety of cognitive approaches for this task, and these approaches have been scrutinized by physicians (Groopman 2007) and non-physicians (Kahneman, Sibony et al. 2021) alike. Physicians generally will not think of these cognitive approaches as distinct, though philosophers may identify several heuristics. Daniel Kahneman (p. 97) defines a heuristic as, “A simple procedure that helps find adequate, though often imperfect, answers to difficult questions” (Kahneman 2011).

One (disease) to one (symptom): Pathognomonic

A given finding (such as on physical examination or testing) is said to be pathognomonic of a specific disease or condition if there is a one-to-one relationship between that finding and that disease. Such simple correlations are helpful when available, but there are actually relatively few such relations in medicine.

One (disease) to many (symptoms): Occam’s razor

The “principle of parsimony” is a heuristic, a method for approaching a problem.  The idea is that if a problem has multiple potential solutions, them the simplest solution is likely to be the correct one.  This heuristic is sometimes referred to as Occam’s Razor, after the 14^th century Franciscan Friar, William of Ockham, even though the idea had been in use for centuries before that (McFadden 2023).  There are circumstances, such as in certain types of scientific modeling, in which Occam’s Razor may not be appropriate (Dubova et al. 2025).  But setting such special circumstances aside, the general application of Occam’s Razor in the practice of medicine is that if a patient presents with three new symptoms, then it is likely that those symptoms are all due to a single underlying diagnosis (sometimes called the “unifying diagnosis”), rather than three separate diseases that just happen to be starting at the same time.

A straightforward example of this from general medicine is that a patient who just developed fever, rust-colored sputum and a cough, is likely to have pneumonia. Is it possible that the patient has the flu (accounting for the fever), lung cancer (accounting for the rust colored sputum) and congestive heart failure (accounting for the cough)? Sure, it’s possible. But it’s less likely than the single unifying diagnosis of pneumonia.

An example of Occam’s Razor as applied to neurological symptoms is the triad of gait apraxia, urinary incontinence and dementia. When these three symptoms occur roughly simultaneously, or in close proximity, then the unifying diagnosis is likely to be normal pressure hydrocephalus, a condition in which the brain ventricles enlarge even though the intracranial pressure is normal. Is it possible that the patient has a severe peripheral sensory neuropathy (accounting for the gait problem), thoracolumbar spinal cord lesion (accounting for the urinary incontinence), and a rapidly progressive variant of Alzheimer’s disease? Sure, it possible. But it’s also less likely than the single unifying diagnosis of normal pressure hydrocephalus.

In neurology the logic of Occam’s Razor extends beyond mere symptoms (complaints reported by the patient) to the signs (what an observer finds on physical examination). If a patient has, for example, three findings on physical examination, and if you can “place” a single lesion somewhere in the nervous system that would account for all three findings, then the location of that lesion is more likely to be correct; more likely than there being three individual lesions in three separate locations. For instance, if a patient has a small pupil on one side, a droopy eyelid on the same side, and weakness or atrophy of the arm on the same side, then it is likely that the patient has a lesion at the apex of the lung on that side that is eroding into the sympathetic chain (accounting for the small pupil and the droopy eyelid on that side) and the brachial plexus (accounting for the arm weakness on that side). Is it possible that the patient has, for instance, a lesion of the superior branch of the oculomotor nerve (accounting for the droopy eyelid), a lesion of the trigeminal nerve (accounting for the small pupil) and a lesion in the brachial plexus? Sure, it’s possible. But it’s less likely than the single unifying diagnosis of a lesion at the lung apex — the most common cause of which is cancer involving the lung’s superior sulcus (sometimes referred to as a Pancoast tumor). Neurology trainees spend several years, and study thousands of cases, to learn to incorporate this skill of “localization” into their diagnostic thinking, and this approach is basically just Occam’s Razor as applied to neuroanatomical pathways, rather than to symptoms.

Many (diseases) to many (symptoms): Hickam’s dictum

Occam’s Razor, however, is merely a heuristic. It does not guarantee you that the simplest solution will be the correct one; it merely says the simplest solution is more likely to be the correct one. When one is faced with several different possible solutions, it is also sometimes difficult to figure out which one is “simplest.” An obvious dilemma could be: Simplest in what way?

But probably the biggest challenge that medicine poses to Occam’s Razor is the reality is that people actually do get sick with more than one disease — a situation sometimes referred to as “Hickam’s dictum,” after Dr. John Bamber Hickam (appointed of medicine at the University of Indiana in 1958) who, according to apocryphal accounts, opined that, “A man can have as many diseases as he damn well pleases.” Medical literature abounds with case reports of patients presenting with multiple diagnoses, sometimes even multiple rare diagnoses.

Many (diseases) to one (symptom)

If three different symptoms are attributable to three individual diseases (in keeping with Hickam’s dictum), that’s already somewhat difficult to figure out. Things get even trickier if a patient has one symptom being caused by three different diseases — a problem commonly encountered in our practice.

For people who have never experienced problems with their balance, the process of maintaining normal equilibrium seems natural, perhaps inevitable. But maintaining balance on two feet is a complicated process, relying on the integrity of multiple areas of the nervous system, all working in synchrony with one another. Consequently, lesions in a variety of places can manifest with the sole symptom of disturbed equilibrium. These tend to be the cases that find their way into our practice, where we’re expected to figure them out.

An example of this would be a patient with diabetic peripheral neuropathy who insidiously develops progressive supranuclear palsy and then additionally is struck with benign paroxysmal positional vertigo. Such a patient may present to our office with complaints of falling and feeling unsteady. In order to maintain balance, the brain needs sensory input from the feet, but that input can be impaired if diabetes damages sensory nerves (diabetic peripheral neuropathy). Progressive supranuclear palsy is a neurodegenerative disease that insidiously chisels away at one’s postural reflexes. Benign paroxysmal positional vertigo is an inner ear problem that provokes vertiginous symptoms triggered by changes in position of the head with respect to gravity. To be sure, these three diseases can present with other symptoms as well; for instance, diabetic peripheral neuropathy can cause pain; progressive supranuclear palsy can cause neck stiffness; benign paroxysmal positional vertigo can cause nausea. But each of these diseases individually can provoke a sensation of unsteadiness in isolation. This situation — in which multiple diseases cause a single symptom — eludes both Occam’s razor (one disease causes many symptoms) and Hickam’s dictum (many diseases cause many symptoms).

False positives and false negatives

Any evaluative history, procedure or technique in medicine has false positives (incorrect appearance of pathology) and false negatives (incorrect appearance of health). When I say “everything,” I am referring to a patient’s history, physical examination, blood work, diagnostic imaging, diagnostic testing, pathology — everything. Even tests previously thought to be ultra-precise, such as genetic testing, have proven to be less so.

Breaking down a complex problem

In clinical practice a physician may hear a patient’s history, do a physical examination, and have a “gestalt” that the available information is most compatible with a specific disease. While such thinking often leads to a correct diagnosis, it is also notoriously subject to a wide variety of biases.

Because arriving at a diagnosis is a complex process, it is often helpful to break down that task into smaller problems, each of which is easier. In medicine this is sometimes done through promulgation of diagnostic criteria, such as, “If you have these two features in the history, and these three features on physical examination, and these two test results, then the likely diagnosis is such-and-such.” This approach is sometimes given a fancy name, such as the “mediating assessments protocol” described by Kahneman and colleagues (Kahneman, Sibony et al. 2021). The rationale is that each smaller problem can be evaluated, if possible in isolation from the other problems, before looking at the sum total (of the results of all the little problems) to arrive at a diagnosis. This approach reduces (though does not eliminate) the possibility of bias, and is also helpful because of the issue of false positives and false negatives mentioned earlier.

References