Recently I tweeted about an article dealing with artificial intelligence (AI) in healthcare (as one does). The author noted that an AI can often be well trained to deal with the vast majority of standard cases it might encounter. For example, a symptom checker can typically offer the “correct” follow-up given a specific set of findings (I put the word correct in quotation marks because so few things are agreed upon in healthcare). Yet with respect to artificial intelligence, the author accurately notes, it is the edge cases that we must focus on.
The Cambridge Dictionary defines an edge case as “a problem or situation, especially in computer programming, that only happens at the highest or lowest end of a range of possible values or in extreme situations.” I find this description satisfying because I was once a software developer (FKA programmer) back in the day (fine, it was decades ago!). Yet, since I hardly program anything nowadays, perhaps I can try to define edge case in healthcare via a common medical school fable.
In healthcare, we often refer to obscure or uncommon diseases as zebras. Why? As medical students are taught, when you hear hooves, think horse, not zebra. Another way of saying this: common things happen commonly. As a young trainee, diseases both rare and commonplace have equal footing in our minds. Hence, students are often just as likely to think brain tumor as sinus infection. The zebra, the uncommon diagnosis, is medicine’s edge case.
Since I’m telling stories from my past, let me give a shoutout to the former Pediatric Intensive Care Unit (PICU) director during my residency, Dr. Ashok Sarnaik. Dr. Sarnaik was not one to mince words nor suffer fools, which made him a bit (ok, a lot) intimidating. One day during rounds in the PICU, I didn’t answer one of his questions to his satisfaction. I remember exactly what he said to me: “Craig, I can teach a monkey to intubate one of these kids. Only humans can think. I need you to think!” Ouch, right?
What the PICU director was telling me (and all of my colleagues) that day was that physicians need to be prepared to look beyond the obvious and commonplace when the evidence or circumstances demand it. While it may be true that common things happen commonly and that we should think about horses when we hear hooves, there are still zebras out there! They live among us, and they present to doctors and nurses without identifying themselves. Naturally, it’s appropriate to focus on the most likely cause for a patient’s concerns, but if we aren’t prepared to contemplate the more obscure answer, we’ve failed.
When I read about the edge case, I was thinking about zebras. Algorithms can get very good at finding horses when they “hear” hooves, but they’re not so good at the uncommon zebras. Is that ok? Probably not if you’re the zebra in need of diagnosis. We spend an inordinate amount of time, training, money, and resources in medicine to ensure we don’t miss the obscure answer. Missing an edge case in a computer program might lead to an unsatisfied user or inaccurate result, but in healthcare, it can be the difference between life and death.
So how do we in healthcare find the sweet spot between completely ignoring the edge cases and focusing exclusively on them? I think a principal answer is a combination of teamwork and ensuring everyone can work at the top of their license. Think about a child presenting to a clinic with complaints of sore throat, fever, and malaise. Since common things happen commonly, the nurse who is rooming the patient and taking the initial history should be appropriately thinking about strep throat. That nurse may be permitted to initiate a workup by performing a rapid strep test.
When the physician, who has more specialized training, comes to examine the child, they may notice more subtle findings such as a stiff neck or a bit of drooling. This may be a patient with a retropharyngeal abscess – indeed a zebra if ever there was one. Working at the top of our license helps us get the most appropriate level of practitioner to address the problem at hand. Further, teamwork isn’t just with medical assistants, nurses, therapists, and physicians. It’s also seen when a primary care doctor works with specialists to get the patient to the right resource as quickly as possible.
In medicine, we need to almost practice Orwellian double think: realize that while the hooves we hear most likely belong to horses, we can never rule out zebras. We must simultaneously ignore and concentrate on the edge cases. It’s not easy, but clinicians do it with every single patient that they see. Now if we can only train the algorithms to do the same. Maybe someday!