My favorite elective in medical school was my genetics rotation. I learned a lot of interesting and obscure facts, I made some good friends, and as luck would have it I was given the opportunity to write my first journal article. Part of what made it so enjoyable was the benign tolerance the staff had for my juvenile sense of humor. For example, in a field chock full of odd terms used to describe unusual clinical findings, a medical student might be forgiven for not getting them all straight in the span of a month. So I coined the term “schmechronekia” to describe any unusual feature, the proper name for which I could not remember. As in “the patient presents with clinodactyly, micrognathia and pronounced schmechronekia.” For some reason, they pretended to find this amusing.
Many years have passed since those salad days of mine, and I’ve managed to shore up my knowledge base a bit. But even within the ambit of my own specialty, it’s still impossible to know everything. It seems IBM wants to render such lacunae in my expertise obsolete. From the AP:
IBM is finally sending its Watson computer out into the working world.
The supercomputer system is best known for trouncing the world’s best “Jeopardy!” players on TV. But now one of the nation’s largest health insurers is hiring it to help diagnose medical problems and authorize treatments.
The insurer, WellPoint Inc., says Watson is a game-changer in health care.
IBM says the WellPoint application will relate a patient’s symptoms and history to textbooks and medical journals. In just a few seconds, it can present several possible diagnoses or treatments.
This does not delight me.
I should admit that I did not watch the “Jeopardy!” episodes wherein Watson made mincemeat of the flesh-based contestants. I gather that it got almost every answer correct (with a few really weird wrong answers thrown in), though I understand that its victory was based largely on being inhumanly quick on the buzzer. Still, it’s indisputable that it can access a vast store of information and come up with the correct answer with dizzying speed.
My negative reaction to getting the Watson treatment is multi-factorial. On its face, I don’t think any of us really relish the idea that a computer could do our professional thinking for us. That’s not to say it’s not true, but I don’t greet that prospect with unfettered joy.
However, this is my cue to talk about the “art” of medicine, which is another way of discussing how much fudging there is. The gray areas show up in various ways. There’s the diagnostic exam, for example. I can certainly think of times when I’ve looked into a kid’s ear and said “not infected” and had a colleague disagree. If you’re plugging information into a computer, it depends a lot on what you call it. Is that rash macular, or is it more of a patch? Is there pain when you press more around the navel, or up closer to the ribs? I wonder how Watson will accommodate that inescapable subjectivity. Or what if a patient doesn’t quite meet the textbook definition of an illness? I’ve had patients whose diagnoses would have been missed if strict adherence to the criteria in the text had been applied. I know my gut instincts aren’t entirely reliable, but they’re served me well enough. Does Watson have a gut?
In addition, any time you’re dealing with people there’s a certain degree of irrationality that has to be accepted. (Not on my part, of course. I’m always 100% rational.) With experience and familiarity with one’s patients and their families, a provider can get a sense of when “unnecessary” treatments really are helpful. Kid X almost certainly has a viral infection, let’s say, but also has really anxious parents. It may obviate a panicked late-night ED visit if I draw some tests I probably could do without, thereby reassuring parents that I’m on the look-out for something dire. Good luck plugging in the numbers for that cost-benefit analysis, yet considerations like those really do play a part in taking care of people. I try to be by-the-book as much as possible, but sometimes fudging is the right decision.
In the end, I wonder what WellPoint, Inc. will really use Watson for. If it’s just a useful way of winnowing down a differential diagnosis or choosing the best treatments for an unusual illness, why should I object? I worry that it may be used more as a way for WellPoint, Inc. to determine what it will pay for. (Call me cynical.) We all need to be paying attention to the problem of rising healthcare costs. But I wonder to what degree Watson will be overriding the human factor in medical decision-making, which can’t be discarded entirely when it’s humans that you’re taking care of.
*without all the money and fame, that is
The problem with your objection is that it’s like trying to criticize Wikipedia. “Wikipedia is wrong about (thing)!” “Oh well if you know so much then fix it!” “Wikipedia doesn’t have enough information about (thing)!” “Oh well if you know so much then add it!”
Similarly, an objection to everything you’ve said there is “you just didn’t write a detailed-enough procedure for diagnosis” or “your critera were not sufficiently defined” or things like that.
…wikipedia is edited by computers (someone human comes back and makes it sound more grammatical) and asbergers patients. know your sources.
Well, that’s part of the problem. How particularly defined does Watson need things to be? And what I may define as a “plaque,” some other provider may define as a “crust” or “scale.”
…which is, incidentally, part of why electronic health records is going to be such an impossible job. Not only do doctors describe the same thing different ways, you’ll get plenty who just will NOT agree to use someone else’s way.
MI is MI is MI. and if all else fails, you CALL THE OTHER DOC.
A big point of medical records is Stopping Killing Patients who have allergies to medicines/foods. It’s kinda hard to screw that up.
Mr. Saunders,
Of course it’s about payment. that’s how wellpoint makes money. But there’s a big difference between “from what you say, we think you ought to consider this,” where a doctor might not know what recommended procedure is, and “we think you ought to consider this cheaper option” when you know perfectly well what’s going on.
I do visual processing — yeah, it may be different NOW based on what you write in… but if you give me snapshots, I can start telling you 30% of doctors class this as A, and 70% as B, use these three tests and go with your gut if they’re inconclusive.
Frankly, I got to beef with that. If this is how it ends up being used, then more’s the better.
… I have confidence that we can run this in such a way that you guys don’t need to be Spock to get the best results. OTOH, I think that doctors who have a lot of hypochondriacs, or really weird patients, may see their payments go down.
The real win is when we can get backfeedback, and actually have the system learn, and raise redflags for research “I(the computer)’ve been told that 40% of people have heart attacks with these symptoms, but I’m picking up a pronounced trend based on ethnicity. Someone want to do a real study?”
Obviously, it depends what it is used for. I agree that determining insurance coverage, etc., solely on the basis of Watson’s output would be a huge mistake. But I am MUCH more excited about the possibilities of this than you are. My kid has an extremely rare syndrome. I was lucky to get a diagnosis relatively early. Many other parents of kids with the same syndrome did not get diagnoses until much much later. Why? Because doctors did not know what to look for. Unless a doctor happened to run across the syndrome in their education, AND remember it correctly, there may be a false result. I was told several times that some aspect of my kid’s health, say aspect X, “couldn’t be” the case, because they’d seen that before, and he presented differently. Eventually, it turned out that X was indeed the case. This was true of both his overall diagnosis (which almost every doctor told me before the genetics results were in that he couldn’t possibly have it) and several manifestations of his diagnosis. Indeed, I was incorrectly told that I couldn’t expect one specific problem as a reaction to anesthesia. This was a mistake which had very serious ramifications. It was a rare reaction, but it could happen with kids with his problems, and it did indeed happen. I later met with several doctors with more expertise in this area were indeed aware of this possible reaction – but not the doctor who had performed the surgery. Doctors relying on their gut were mostly right. But in many many instances, they proved to be wrong, and wrong largely because of their inability to know all the details of extremely rare medical occurrences.
I agree no computer is sophisticated to replace some of the subtleties of doctor patient interactions, and that experience is irreplaceable. But the brain has limits. There are too many diseases and problems. The doctor of course should have the final say. ANd hopefully the doctors could understand by the limitations of the information they enter, just what are limitations of the answers Watson has. But this tool could be invaluable in getting people the right diagnoses and the right treatment early.
Rare diseases are just that…rare. But if you add up all the rare diseases in the world, that makes a good number of people who could benefit from something like this.
I should probably have shuffled things around a bit in the main post to make my primary concerns more clear. Given that the company that’s rolling Watson out is an insurance company, and insurance companies are not typically the repository for diagnostic insight, my major concern is that WellPoint, Inc. might be using the tech more to determine what services are justified/worth paying for than anything else. Most providers don’t look to their patients’ insurance companies as a source of reliable medical management information, so I’m curious just how WellPoint plans to make Watson available, and to whom.
I agree that in the case of your child, having an ultra-smart diagnostic computer may have helped. But if many of his features were atypical for his diagnosis, would Watson not also be similarly misled? Again, I’m curious to see exactly how the technology will be employed as a practical matter, and I’m willing to admit the possibility that it will be a fantastic boon to medical care.
“You see, it’s not me that’s denying these services. I mean, if it were up to me I’d be happy to pay you. But, well, the XR-2200 says that we can’t do it. And, well, Computer Says No.”
…And the Computer is your friend.
Yes, agreed it should be in better hands than an insurance company. Unless the insurance company really were concerned only about containing the costs of misdiagnoses (which I doubt), it does presage Bad Things. It would be much lovelier in an academic setting, with contributions from all experts on diseases.
His features were atypical, but not unheard of. Except most docs hadn’t heard of the less typical possibilities, and they are listed in journal articles, but not textbooks. A sufficiently detailed computer might have been able to help. And less typical features had nothing to do with the anesthesia problem, which was far more serious. The surgeon and anesthesiologist just hadn’t had the experience of a kid with certain neuro issues reacting a certain way – and so didn’t think it could happen. That cannot be that unusual an occurrence.
… the insurance company around here is part of a university. I think the whole thing is a nonprofit.
That said, Part of the whole thing will be compiling stats to see which doctors are poor performers (not doing well in making sure their diabetes patients don’t hit the ER), and I’m sure there will be some sort of remediation if the doctor’s too bad.
(my experience on this is limited to talks and what I hear on the lunchtable.)
But can Watson insult people and abuse Vicodin?