Very interesting article, his previous book 'The emperor of all maladies' is also highly recommended. A lot of medicine is not level III evidence or better backed.
As both a doctor (surgical attending) and programmer medicine feels not just like debugging a very complex non linear machine. Patient psychology is one of the hardest parts of the job. Is the stomach pain caused by a psychical problem or can it be functional?
The use of protocols (part of the science) can help in catching non typical presentations of illnesses, especially if the patient is not adequate in presentation (drunk, very young, old, trauma etc).
But sometimes medicine comes down to an art, when you have the feeling something does not add up, you have to dig. The same with (older) attendings with very attuned diagnostic skills due to lots of experience that can help solve non-obvious cases.
> medicine feels not just like debugging a very complex non linear machine
One way in which medicine is not like debugging is the hours spent by a physician vs a programmer. I (as a programmer) have sometimes spent 100 hours tracking down a bug whereas I've never seen a physician spend more than a few minutes trying to diagnose an illness.
Even with a gravely ill person, my experience has been that an attending physician has never spent more than about 15 minutes on the diagnosis (or to be a bit more generous, let's say several 15-minute re-evaluations as new information came to light, such as blood test results).
It's a strange contrast. I considered my bugs to be serious bugs, and it was worth 100 hours to diagnose it correctly. But my bugs are nothing compared to a human life. Why is our medical system OK with quick diagnosis? Does it all come down to money or would more time available to the physician make no difference because [medicine is art not science / some other reason]?
I don't think that's a fair comparison. For complex cases, a medical diagnosis would order a battery of tests in that first 15 minute consultation. Then, a different group of healthcare specialists would spend several hours running requested tests, and send the results back to the original Doctor. This would be the equivalent to the programmer setting up and running a number of tests to learn more about the program, which would take several hours of the programmer's time. The only difference between programming and medicine is that testing and diagnosis is done by the same person with programming.
It's easy to spend 100 hours when you know there has got to be something to find.
Medicine isn't like that. With experience you get a feeling for when you are missing something, and then invest the time to investigate in more detail. I could spend a 100 hours trying to understand why my grandma collapsed in the lounge, but 50% of the time I won't find anything at all.
The analogy is not apt. A closer analogy is not a programmer - the doctor is in IT support ... and is dealing with a unique combination of bespoke hardware and software for each case ... and has no makers documentation available.
I wouldn't say bespoke (which implies written from the ground up from scratch), but with possibly different environment and other configuration variables and, occasionally, slightly different shared library files.
And while there is no official documentation available, there is plenty of good guesses from hordes of brilliant reverse engineers.
Still, there are only a handful of error codes that are thrown, and you have to make your first guesses from those and the logging is pretty terrible when you do decide to dig a bit deeper.
And inserting your own code into the spaghetti mess is a bit of a crapshoot.
A lot of this is front-loaded in the 4 years of undergraduate training, 4 years of medical school, and 4-6 years of residency training that physicians do.
Physicians are trained to preload all possible diagnoses based on a certain set of initial symptoms--they have ideas about what the potential answer might be when you inform them of their symptoms, but in order to narrow that space down, additional workup needs to be performed (tests, imaging, etc) and until those results come back, anything else they say is speculation.
Very interesting comment. We do have to realize for semi-complex to complex diagnoses, many different doctors will be playing different parts. If you have a complex/rare form of cancer, it's common for a group of 3-6 physicians of varying specialty to sit in a room and discuss your case.
Additionally, when you work on a software-bug you probably try a bunch of different solutions. Try it to see if it works, repeat. There's only so much information on a patient, doc spends some time reviewing and makes a diagnosis/treatment decision/orders tests. Wait a day or a couple weeks for results, then repeat.
And doctor resources are scarcer than programmers. It would be great to have 2-4x as much time with a physician when you are sick, but $$$.
>I've never seen a physician spend more than a few minutes trying to diagnose an illness.
I have. The physician's debug loop just takes a lot longer than a programmer's. He orders tests, then waits days for them to come back before ordering more tests. This can go on for months. Then he prescribes something, and has you take it for a few weeks. Did that help? No? Try this instead. Etc.
Too bad it's all standardized now so the quality of care goes down for everyone. Doctors that actually care about making their patients better need to be able to deviate and research for themselves... and not be overwhelmed.
Having watched a fair bit of this from the sidelines, I suspect we are heading into a bit of an existential crisis in medical training and clinical practice, because our ability to generate potentially meaningful data is already outstripping the time and ability to analyse it (and train for that). The difference is accelerating.
Of course this doesn't mean that your putative attendings diagnostic skills are any worse - just that they are going to be more and more likely to miss things that actually were inferable from the data, when the case is not simple. At least, without a possibly profound change in practice and training.
To quote the article: "Mukherjee’s laws of medicine are laws of uncertainty, imprecision, and incompleteness. ...".
I don't know if these ideas constitute science "laws", but of course the practice of medicine is bounded by such factors. Then again what discipline of any importance is not also similarly constrained? A common theme on HN is the inevitability of software flaws, the uncertainties and corner cases inherent in programming tools, like compilers, render satisfactory outcomes a matter of probability, not at all unlike medical diagnosis and treatment.
Nothing new, medicine has been regarded in the last century or so as an art, with a basis in science to the extent it exists. Truth is that a great many phenomena have received little or no study, yet unstudied problems still show up in the office every day and must somehow be addressed.
Good practitioners welcome scientific contribution and eager to employ "evidence-based" procedures. Trouble is there's not much guidance forthcoming, leaving those in the front lines doing what they can with the inadequate tools in the toolbox.
It's worth adding that a substantial part of the probabilistic nature of the work is attributable to the limitless variability of human illness. I've long expressed the idea in a terse form: The reality is no two people ever have exactly the same disease. The subtle differences are what make diagnosis and treatment a constant challenge.
Put another way, abstraction is at once a powerful benefit and potential source of error. Declaring a patient's condition falls into a particular diagnosis lends a valuable clue. However such categorization may easily blind us to the small, crucially important distinctions among sufferers with the same diagnosis. We can easily be fooled by our own cleverness. Point is that the intuition, the talent, of a fine practitioner will always defy "explanations", and that makes it an art.
What's worse is that modern day society expects physicians to be able to act with some certainty. And when outcomes don't turn out the way everybody wishes it would have, the target of scapegoating is the physician.
If you saw a person falling in front of a car, and you stopped to help him but in the process stepped on the man's glasses, would you believe it to be fair for the person who fell to blame you for the broken glasses?
And in a world where stepping on someone's glasses can bankrupt a well to do family, that pettiness of yours and all that think as you do causes people to die under the tires of a car... because would be helpers do not think they can afford to do what is right.
I will note that Siddhartha Mukherjee is also the author the _The Emperor of All Maladies_, perhaps the greatest popular science book on cancer that is scientifically and historically accurate and very engaging. I highly recommend it (and will be picking up this book on the basis of Dr. Mukherjee's previous work).
Interesting. For a while I've felt that Medicine isn't the science it's cracked up to be - or at least our knowledge of the state of the art is never as certain as presented by science. It seems to have many false paths of thought similar to economics.
In the spirit of confirmation bias, I recently listened to a great podcast relevant to the topic [0] which discusses why this is. It's more than just lazy statistics.
Very interest article (yeah, it scrolls for me). And it mirrors some of my concerns with only basing yourself on EBM
Most studies are flat. They are concerned with "condition X and drug Y". They go to a finer granularity usually when it's obvious (like treatment A works ok but it is obviously a worse choice if patient has condition B). Getting too deep in conditions is a slow process
Added to that the elements of diagnostics, which more often than not is a "Plato's Cave" view of the patient's condition
"The first law is that a strong intuition is more powerful than a weak test. By this, he means that common things occur commonly and uncommon things, uncommonly... “If you hear thundering hoofbeats, think horses, not zebras — unless you happen to be in Africa.”"
Yet doctors routinely do harm - chemotherapy, amputations etc. Its part of many therapies to make a patient hurt. So that's a noble goal, but modern medicine left it behind a century ago.
As both a doctor (surgical attending) and programmer medicine feels not just like debugging a very complex non linear machine. Patient psychology is one of the hardest parts of the job. Is the stomach pain caused by a psychical problem or can it be functional? The use of protocols (part of the science) can help in catching non typical presentations of illnesses, especially if the patient is not adequate in presentation (drunk, very young, old, trauma etc).
But sometimes medicine comes down to an art, when you have the feeling something does not add up, you have to dig. The same with (older) attendings with very attuned diagnostic skills due to lots of experience that can help solve non-obvious cases.