Dr. Jerome Groopman, whose writing I generally enjoy, put out a book a couple of years ago called How Doctors Think. It examined, well, how doctors think, how they think they think, and what the future holds for diagnosing disease. It's a good book, but with some faulty assumptions. I'm not the guy to write about how decisions are made---I don't know enough about the field, a field which needs much more research. But most doctors do not, as is sometimes posited, make diagnoses via algorithm. Nor are we slavishly bound to statistical likelihood, as the use of likelihood ratios and, er, the like has some problems. What we do teach formally is the process of differential diagnosis.
Differential diagnosis (DDx) is the fun part of medical thinking, and hopefully the lessons learned about the process endure. When a resident or medical student presents a case to me, they often have an immediate feeling for what is wrong with the patient. This feeling may or may not coincide with reality. Getting a gestalt feeling for a case is important, but it is only a starting point. One of the gestalts I like my residents to get a feel for is whether a patient is really sick---I don't mean whether or not they have a cold or whatever, but do they appear seriously ill. There are parameters which can help determine this, but when someone comes in with a vague picture and you don't have access to sophisticated diagnostic equipment, it's good to be able to make that judgment.
One of the luxuries of being a teaching physician is being able to take the time to break down a case in a more formal manner and to develop a traditional differential diagnosis.
One way to do this is to present very little information to start
with. For example, if a student tells me a patient is anemic (has a low
red blood cell count) I may stop them and ask them what, very broadly,
can cause anemia.
- Loss of blood
- Decreased production of blood
- Destruction of red blood cells
- Sequestration of red blood cells
Then we can look at the evidence to make some guesses as to what may
be causing anemia in this patient. Are they bleeding? Do they have a
big spleen? In addition to these broad categories, we can look at
different categories of cause, or "etiology". What disease categories
can cause anemia?
- Neoplasia (cancer)
- Immune-mediated disease
This process often takes place implicitly, but I prefer to make it
explicit as often as possible. Once we've laid out the thought process,
we can look at the evidence presented by the patient herself and start
to cross out possibilities. For example, if the patient isn't obviously
bleeding, her stool is tested negative for occult blood, she no longer
menstruates, and has no evidence of internal bleeding, then acute blood
loss seems unlikely as an explanation. If the labs show normal red
blood cell morphology, but decreased kidney function, then decreased of
red cell production due to kidney failure seems more likely. If she has
poor kidney function, her red cells appear deformed, and her platelet
count is also low, this opens up another door.
Differential diagnosis stands in stark contrast to fake medical thinking. If you look at how some alternative practitioners think,
you see the evil, goatee'd twin of differential diagnosis, in which
someone ignores biology and comes up with elaborate explanations for
disease which bear little resemblance to reality.
Differential takes the implicit and makes it explicit. It forces us to
think of all possibilities and winnow them down to the most likely
answers. It's also a helluva lot of fun. Perhaps we will have a few
morning report cases here where we follow this process. If you want to
submit a case, go ahead, and I'll turn it into a post so we can all
play. If you do submit a case, please remember that privacy is a must,
both legally and ethically.
Thanks, I always enjoy your writing if you were ever to go over some typical cases that would be awesome. One of the other aspects which would be fascinating to read about wold be the common approaches of "it could be X or Y but just in case we should also rule out Z" Where X and Y are more probably but Z has a higher morbitidy/mortality. Obviously this approach arbitrarily narrows the differential to only three options...
Indoors or outdoors?
First response also gets into a more limited set of decision criteria (we don't, by definition, "diagnose") and there are boatloads of examples both from direct experience and from years of training examples.
The basic facts of Natasha Richardson's case are pretty much public, for instance.
If you are asking whether to post the case or email it, it's probably better to email it so I can futz with it then present it.
It's always humbling when you're discussing a patient who says that "my belly hurts" (or something similar) and within fifteen minutes you have a whiteboard literally full of very, very different possible diagnoses. It is also very satisfying to know that you can then narrow the list down to a few different etiologies with only a good history and physical.
That sounds a lot like an algorithm to me. Out of professional curiosity (I'm a computer scientist), what's your definition of "algorithm" that doesn't include the process you describe?
Your algorithm doesn't have steps in it for anemia specifically but it does have steps for asking "what causes anemia?", eliminate, repeat, etc.
This is not an algorithm you can currently write software for (very well), much less teach a computer to perform.
This should be an interesting series.
In fact, I'd like to volunteer myself. Right now the only "official" diagnosis I have is "other, etc., NOS" Any ideas?
Seriously, I'm looking forward to this.
One of the infinite varieties of how to build a differential does indeed incorporate that. It's one I learned early in med school, great for learning how to prioritize.
Basically you divide your diagnoses into tiers - 1, 1A, 2, and 3.
1 is your most likely diagnoses.
1a is your less likely but life threatening diagnoses (or less likely but otherwise must rule out, your Z diagnosis for example)
2 and 3 are each less likely diagnoses.
I think explicitly laying out a differential is a really important process in the education of a medical student. Often if it's not done, novices miss important things. Even when it's being done implicitly in the heads of more experienced doctors, laying it out explicitly provides an opportunity for students to "get into the heads" of their mentors.
How do you feel about cases without a resolution (i.e. that the ultimate etiology is still unknown?)
aren't those the best?
The ultimate purpose of a differential diagnosis is for differential treatment. For some things a differential diagnosis is not necessary if it doesn't result in differential treatment.
Once I had pneumonia, diagnosed by chest sounds, weakness, and slight delirium. The doc suggested a chest x-ray and I asked what for, would it affect the treatment? He said no, so I didnât get one. I was in my late 30âs, never smoked, had been in that practice for 15+ years and the symptoms had come on suddenly. The treatment was going to be oral antibiotics no matter what the x-ray showed, so the x-ray wasn't needed.
Much of what CAM does and pretends to be âdiagnosingâ something really isnât. If all a chiropractor is going to do is âcrack your backâ, what does he/she need an x-ray for? The use of x-rays and technological devices such as E-meters is simply part of the placebo process. Make it complicated and sciencey and it will be a better placebo.
lylebot "That sounds a lot like an algorithm to me. Out of professional curiosity (I'm a computer scientist), what's your definition of "algorithm" that doesn't include the process you describe?"
Colin "This is not an algorithm you can currently write software for (very well), much less teach a computer to perform."
OH,OH, I'm married to an engineer who specializes in algorithms, so maybe I can take a stab.
You don't have to teach a computer to do it, you already have a really powerful one in your head. An algorithm does not have to be computer driven it is just a series of defined steps taken to solve a problem that can be repeated. Sort of like a decision tree. Different field, same process.
Here would be a medical example for cough. http://familydoctor.org/online/famdocen/home/tools/symptom/516.html
micheleinmichigan, yeah I'm a computer engineer who writes software. :) The point was DDx is an algorithm but you can't write it in software, much less teach a computer to perform.
Current practice of "artificial intelligence" is, well, not intelligent. The Turing Test is, frankly, overrated and insufficient (see Searle's argument for starters). But now I'm way off topic.
I remember my doctor once saying to me. I didn't expect a younger woman like you to have two unrelated illness.
I came in for a feeling of a lump in my throat, (which turned out to be hashimoto's thyroiditis), a nagging cough and out of breath (which turned out to be asthma.) fatigue and aches and pain (whichever).
What is the process for contributing symptoms to one or multiple diseases?
This is also an argument for doctors and patients to have some, well, patience, with each other, because it did take a few months of return visits, test, etc. to figure things out.
speaking as someone who's not a medical professional but watches them on TV, how much of a resemblance does the DDx process on "House" bear to reality? Putting aside the dramatic conventions of "the ducklings do all the procedures and there are no nurses or techs" and "all test results come back at warp speed" -- and of course the snark -- is the actual give-and-take of diagnosis anything like the scripted version? The show does have medical consultants, but I always wonder how far reality reaches before the demands of the script take over.
I've never found house to be in any way realistic. They usually get the basic medical facts wrong, although the thinking process is at least familiar.
Young male patient (20s). Presenting large purplish bruise on left side of torso 3 weeks after blunt collision (got tackled). Complains of acute pain upon deep inhalation. WBC elevated. Breathing sounds normal, no rattling. Heart rate normal, blood pressure normal, temperature slightly elevated. Urea elevated, bilirubin normal. Stiff and painful to lift left arm more than 15 degrees of arc above shoulder plane.
Toaster - no medical training here, but couldn't this be a broken rib?
The House 'Algorithm':
Patient presents with mysterious yet worrisome symptom(s)
House pulls out whiteboard.
Duckling 1: Could be sarcoidosis. (editor's note: it's NEVER sarcoidosis)
House: *smackdown* Run these tests. Go!
Ducklings: All tests came back normal.
House: *smackdown* Go do absurd illegal investigation of patient's home and fancy impressive looking tests.
Ducklings: We found a penguin, a box and a ball of twine in their house. Fancy impressive looking tests told us nothing, but patient had a Very Scary Symptom in impressive machine. Now we have to figure out what is wrong, or they will die in 10 minutes.
House: *smackdown* Hmmm. Patient told me they were abusing heroin, but they were lying. Everybody lies, you know. Could be cancer. I'll go bug Wilson. You go run these extraneous tests in the meantime.
House to Wilson: make me a sammich, bitch
Wilson: Don't you have a patient?
A Miracle Happens.
House is struck with Inspiration.
House to ducklings: start them on this treatment. Also, you are all morons.
Is this the thinking process?
"The House Algorithm"
You left out the brain biopsy: rectal bleeding, do a brain biopsy; cough, do a brain biopsy; Wilson coughs, do a brain biopsy.
And the toxic medication. The more toxic the better. If there is only one disease (with only 12 cases known worldwide)for which medication x is a legitimate treatment; and med x kills half those who receive it and gives all the survivors ovarian cancer (including, remarkably, the male survivors) then this will be the first medication House orders. At the wrong dose, route and frequency.
In the House MD frame of mind...
Differential: Is it Lupus?