The Curious Incident of the Dog in the Diagnosis


“Is there any point to which you would wish to draw my attention?”

“To the curious incident of the dog in the night-time.”

“The dog did nothing in the night-time.”

“That was the curious incident,” remarked Sherlock Holmes.

Sir Arthur Conan Doyle, Silver Blaze

We can learn a lot from Sherlock Holmes.

If there’s anybody who better personifies the ultimate diagnostician, I don’t know who. Sir Arthur Conan Doyle, creator of the Holmes canon, was himself a physician and purportedly based his famous detective on Dr. Joseph Bell — who, it was said, could glean a dozen esoteric facts of a patient’s background, history, and complaints from a single glance. (Holmes himself, of course, was not a medical man; that role was played by Watson, the earnest physician who carried his stethoscope wrapped inside his hat.)

Holmes didn’t diagnose illness. Instead, he diagnosed crimes. But the methods were the same, so much so that among the countless fictional characters based upon the Holmesian archetype, some have been crime-solvers (cf. Monk), yet others have been medical doctors (House is the best). Perhaps we shouldn’t model ourselves after the man, who was a single-minded addict and misanthrope, but when it comes to diagnosis — something we can’t escape in medicine — he knows whereof he speaks.


The diagnostic method

Holmes tells us in The Sign of Four that detection involves nothing more than three skills: observation, deduction, and knowledge. Let us consider what he means when faced with, for example, a complaint of chest pain.

Observation: we perceive a middle-aged male, alert and seated upright, rubbing at his sternum with a pained expression. His skin is slightly pale, his respiratory rate is slightly elevated, and he is hypertensive. He complains of “tight” 4/10 chest pain whenever he breathes. Upon auscultation we detect diffuse, bilateral, biphasic wheezing. We note a history of coronary artery disease, diabetes, and COPD.

Knowledge: Chest pain in adults indicates a high risk for acute coronary syndromes. Pallor, tachypnea, and hypertension are consistent with this diagnosis. Sharp, mild, pleuritic pain is not, nor is wheezing, all of which are more consistent with a primarily respiratory etiology. But we also know that MI often presents atypically, particularly in diabetics.

Deduction: Both cardiac (ACS) and respiratory (COPD exacerbation) diagnoses top our differential. An ECG and biomarkers are needed to further evolve the odds.


So what just happened? We observed using our medical assessment — the history, physical, and diagnostic tests — thus yielding a collection of facts and data. We took the set of background knowledge we already possessed, regarding pathophysiology, epidemiology, and hazard ratios, and used it to “fill in the blanks” and provide context to our assessment findings. Finally, we connected the dots together and used deduction to decide what we’re dealing with.

Holmes knew this method well. He might observe your tattered boots, and using knowledge he possessed of typical wear patterns in the various trades, deduce that you make your living as a longshoreman. Simplicity itself.

Why is this a useful model for diagnosis? Because it highlights the fact that these three skills are entirely distinct, though all quite essential. Observation requires skill with the physical exam, the ability to take a nuanced history, the acumen to interpret diagnostics — it’s simply the trait of being aware. (Holmes, succinctly: “Data, data, data! I can’t make bricks without clay.”) Knowledge is knowledge: it’s memorized facts, what you learn in school or from books, and it gives us the basis to understand the raw material we discover in our assessment. Finally, deduction is the mental capacity to analyze, discover patterns, weigh odds, use your imagination, and extract from the vast pool of observation and knowledge the particular pieces that are actually relevant. (Holmes: “… to recognize out of a number of facts which are incidental and which are vital.”)


The hidden danger

Here’s the rub: we’re almost too good at deduction. Humans are excellent at finding patterns in anything. If I leave you my tea-cup for long enough, you’ll undoubtedly find an image revealed in the leaves.

That’s good — but it’s an error. Because there’s not really any image in the tea leaves. But if you’re good at observing details, and have a strong imagination, you’ll still “deduce” many wonderful things from it. Call it apophenia: people want to connect the dots, even when there aren’t any. So we create connections that may be true, but are not always true. We develop stereotypes. Simplifications. False associations.

In medicine, we’re especially prone to this. Because we do know that the human body is interconnected, and that patterns are the rule rather than the exception. Indeed, a large part of developing experience and clinical judgment is increasing your catalog of mental connections. Crackles mean CHF. Irregular pulses mean A-fib. People with Foley catheters have UTIs. Homeless people are drunk. Toe pain is a nonsense complaint. We can’t avoid making the connections, because just like when Holmes examines your boots, those connections are essential to doing our job. But at the same time, we need to learn when to reign them in, or we enter an inescapable diagnostic tunnel after the first moments of patient contact.

It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts.

A Scandal in Bohemia

What’s the secret? Knowledge.

Sure, we gather the pieces from our assessment, and we automatically start to connect them together. We can’t help that; patterns jump out at us, we’re natural pattern-finding machines. But using our knowledge, we can look past those simplistic, eye-catching patterns, because knowledge tells us something more subtle: what’s missing.

She’s all false positives. See, that’s the trouble with naturals. They don’t see what’s missing.

Lie to Me, “Moral Waiver”

Okay, Friday night, a “man down” call for a homeless guy on the sidewalk. You’re already thinking: drunk. And the initial observations confirm it: he rouses sluggishly, slurs his speech, and pushes you away as he rolls back over. But then you open the mental box that you filled with this sort of thing in your training, and you reflect: where’s the bottle? And this is a strange spot — it’s cold, wet, public and unsheltered. And come to think of it, is that a medical alert bracelet? We should probably check this guy out a little more. Maybe take his blood sugar, look for any trauma, shake him awake and ask some questions.

The initial pattern recognition is there, but you don’t have to be a slave to it, because you know what else to look for. Even if five clues say one thing, if we don’t see five others that ought to be there, that tells us something different. Pertinent negatives, they call ’em in the business.

Maybe there’s nothing else; maybe the drunk is just drunk. But we’re too smart to make that kind of assumption. Because we know that getting it right doesn’t just mean registering the hits — it means checking off the misses, too.

Holmes would expect no less.




  1. Long after this post went up, I came across this paper, which hits upon some similar points.

Speak Your Mind