Atul Gawande wrote about 'non-human' tools in medicine [0]. He advocated using checklists but he has also written about other tools. One I remember was a flow chart to predict some sickness. It outperformed humans solidly.
The biggest problem with checklists, flow charts and software (one older example is MYCIN [1]) is adaption. If the medical practitioners don't want the systems they are going to fail. He also writes a lot about that in the book. He tried to introduce it into hospitals but the professionals often ignored the lists or just checked the boxes without actually checking the condition.
I personally imagine that a practitioner with a good system works the best or like Frederick P. Brooks said [2]:
> If indeed our objective is to build computer systems that solve very challenging problems, my thesis is that IA > AI that is, that intelligence amplifying systems can, at any given level of available systems technology, beat AI systems. That is, a machine and a mind can beat a mind-imitating machine working by itself.
What I would like to see more in these discussions is the possibility of using computers for triage in medicine...too often, these kinds of "will computers replace us" discussions focus on the computers doing everything...but how about a computer being trained to do as much of the "mechanical" work of a doctor's judgment as possible? Leaving the humans to focus on the 5%
An example I have in mind is Apgar's score, an almost stupidly simple index of a newborn's baby health, but a system devised by an anesthesiologist who realized that doctors' traditional method of judgment (check to see if the baby is crying "weakly") left many babies at risk of dying needlessly. http://en.wikipedia.org/wiki/Apgar_score
I think if the profession stopped worrying about "replacing" doctors and rather, augmenting them, I think the discussions would be more beneficial.
Furthermore, the things the author of the NYT piece complains about are exactly the things that even the simplest machine learning algorithms handle easily. I'm worried articles like this will hold back the adoption of effective analytics in medicine, since doctors have no understanding of modern machine learning technology and as a result make wildly incorrect assumptions about the limitations of our ability to use data to support clinical decisions.
I think that people often miss why people go to doctors in the first place and why they trust them. In society a doctor was historically always a very well respected individual. I think it's worth reflecting on that. Why were they respected? Did it help the patient that they were respected?
Considering the impact of placebo medicines, I can't help but wonder if there's an element of psychology that needs to be noticed and addressed here (and whether it is actually key that a doctor is respected (for the right reasons, obviously. Specifically: that they will be able to address the problem addressed to them)). My hypothesis is that a good doctor is someone who asks the right questions but also looks for questions that need to be asked for that individual's needs. Example: the patient has the symptom of a broken arm, but why is that. Is there something more here? Is there some form of domestic violence involved? Or... do they have a weakness in the opposite arm, and when cycling they are more likely to fall off.
One of my doctor friends said to me that they'd recently shown that the better the Doctor communicates with a patient, the lower the chance of lawsuits. I like to call it 'managing expectations', from the whole project managementy world although someone will shoot me for saying that!
It may well be the case that on paper checklists work best, but I assert that there's a lot more personality and social interaction to a doctor than anyone initially realises. The challenge in this space is marrying the two concepts.
Doctors are expensive. Computers are cheap. The more we can replace doctors, teachers, taxi drivers and other baumol workers with machines, the more consumer surplus is generated.
How would improved diagnosis bring down the price of medicine? Medicine doesn't only consist of diagnosis, there's also treatment and infrastructure. It's obvious that any improvement in diagnosis of illness cuts down on misdirected treatment and improves public health, saving money would be a secondary effect.
Besides, the fact that healthcare in the US is so expensive compared to the rest of the Western world is due to political reasons; we know how to bring down the cost of healthcare here without affecting outcomes by doing what the rest of the civilized world does.
How would improved diagnosis bring down the price of medicine?
I didn't say improved diagnosis would. I said replacing humans with machines would. If the process were 1 hour with a diagnostic bot and 1 hour of treatment with a doctor, that will allow 2 patients to be treated in 2 hours of a doctor's time (increasing supply and reducing price).
Your political baiting is completely orthogonal to the point. If you play politics and tweak the payment system (e.g. adopting India's excellent capitalism-based system), a doctor still spends 1 hour diagnosing and 1 hour treating. 1 patient is treated rather than 2.
Consumers can't consume if they don't have the money. Every time you fire someone and he can't get job for time X, you are reducing overall consumption during that time X.
Instead the money if concentrated in fewer hands and there is an upper limit to what a human can consume (e.g. a single rich person can't eat more than hundred middle class people).
If someone is spending less on healthcare, they will instead direct that money into buying perhaps restaurant meals. More demand for restaurant meals means that prices and profit margins in that industry will go up which will cause more restaurants to open, who will in turn employ more chefs and waiters causing their wages to go up and causing them to buy more things (like healthcare).
I don't think healthcare expenditure would have reduced cost, because health is not a negotiable goods.
If anyone becomes permanently unemployable, he will become poor. If there is demand for restaurant meals, prices and profit margins will increase and poorer people will be unable to afford it.
Each person that becomes poor, is another consumer less and the rest of people will become richer. Once you have waiters/chefs that have super high wages, they'll be automated and removed permanently from their job. Repeat until there are only few producers/consumers and the rest are living on a fringe.
Totally agree. Of course there is going to be more backlash if this is described as a "replacement." What's wrong with just offering doctors better tools to do their jobs? We offer them surgical tools, why not software tools? That's a less threatening description...but also a less compelling headline.
I despise this attitude that people go to the doctor to figure out what is making them sick. Most of time, when people visit general practice doctors, they already know what's wrong with them. But we have to wait hours or days for an appointment, drive through traffic, sit in a waiting room for hours, and pay a bunch of money so that some jerk can spend five minutes typing our symptoms into his laptop.
I get it: they're gatekeepers to prevent people from abusing or misusing medicines. But wouldn't it be better to allow people access to medicine based on objective measurements taken over a long period of time rather than making people interview for it?
When I'm sitting in a waiting room, I'm thinking about what I need to say and do to convince the doctor that I'm sick. If I'm polite and smile is he going to believe me when I tell him I'm in pain? If I don't have a fever is he going to believe I had one two hours ago? I know that he gets paid whether he treats me or not and that he knows every treatment is a potential lawsuit if he makes a mistake. He has every incentive to play it safe and find some reason to avoid treating me.
The other day, I read an article about a woman in Houston. Emergency room doctors avoided treating her for a stroke because her boyfriend mentioned that she has occasionally smoked marijuana. Classifying her as a drug user allowed them to just give her some Zoloft, push her out the door, and collect the check. She'll never be the same again. And this happened in one of the best hospitals in one of the best medical centers in the world. The problem is systemic.
your characterization of visits to the doctor is spot on: they are gatekeepers to medication and treatment and they have a warped incentive structure that biases them to under-treat anyone they don't "believe".
on top of what you describe, i find most doctors to be biased towards "i know better than you" behavior. i have found that being an intelligent person makes dealing with doctors especially difficult since you need to be strategic to convince them and at the same time appear to not know too much about your own health problems. i have had incompetent resident doctors argue with me just because i know more about my illness than they do, and their ego is such that they cannot admit to themselves or me that they are wrong.
any technology that leads to me spending less time interacting with doctors is a huge win.
Classifying her as a drug user allowed them to just give her some Zoloft, push her out the door, and collect the check. She'll never be the same again. And this happened in one of the best hospitals in one of the best medical centers in the world.
Apparently not.
It's worth remembering as we discuss this that the US healthcare system is exceptionally poor by first world standards. There are all kinds of systemic corruption and perverse incentives that result not from clinical practice but from the legal and insurance systems, and no amount of computer assistance is going to fix those things alone.
"... At the Mayo Clinic’s Transform symposium this month in Rochester, Minn., I heard Eric Dishman, a general manager at Intel, explain how he had used data to individualize his own cancer care. More than a decade ago, when he was only partly responding to chemotherapy for a rare kidney cancer, he used a step monitor to help figure out what provoked his pain and then worked with a physical therapist to treat it. More recently, scientists were able to analyze the genetic sequence of his tumor, identifying a medicine for treatment. He is now cancer free..."
If you find the work in cancer interesting, it's worth looking into a few commercial outfits that are doing this type of testing:
* Genomic Health ($GHDX) with OncotypeDX
* Foundation Medicine ($FMI) with FoundationOne
Both of these companies have tests on the market that analyze tumor genomes (sequencing or microarray-based) and provide information to oncologists to better treat their patients. Information like drug efficacy and tumor aggressiveness (requiring chemo-/radiotherapy).
As someone who has been in and out of the US medical system, I strongly agree with the opening statement by Dr. Wadhwa.
In my experience there are a lot of incompetent and immoral doctors. As a patient I need a trusted opinion, for I myself have not spent the last ten years in training to become a doctor. So who or what can I turn to? The medical system in this country feels too badly broken. I need another option.
Seriously, what is the cause for this in American medicine? Every time I go in for something simple, I have to go through about 3 layers of administrative people before even a nurse practitioner sees me. I'd be really glad if I could interact with a computer-based system in the mean time.
The doctor starts on time. Patient one requires slightly more time than scheduled...repeat for other patients. Also doctors have other tasks than seeing patients: reviewing lab work (part of seeing patients?), pharmaceutical sales reps, signing paperwork, etc.
Automation would be nice, but not sure how you could remove all the time sinks in US Medicine.
and some patients miss appointments, and nobody plans to be ill so there are last-minute "squeeze-me-in" appointments, etc. etc. Some practices manage this much better than others though.
I think that is the more interesting question: why don't practices get better over time at managing this? There are places I go: hair salon, doctor, the vet that are always 30 minutes behind. As a one-off that makes sense, but as a pattern stretching over years it means they don't adapt to account for this. If, on average, appointments take 2 minutes longer than they think then change the appointment times to account for that. Do _something_.
It annoys me, as a process-oriented thinker, that so many offices never get better at handling average daily volumes.
Easily, definitely. I've never had a primary care physician that couldn't have been replaced by a "Dr Kiosk™"; which would almost certainly have been superior in nearly every way.
Of all mainstream professions (including software developers) medical doctors have the highest average IQ. They understood very well how dangerous for their mortgages and standard of living a computer doctor can be. The MD lobby in US have fought tooth and nail since the 1980s to stall advance in this area.
Atul Gawande's checklist manifesto emphasizes the need for processes while solving complex problems like surgery. A doctor can forget to wash his hand while a machine cannot forget. So that is a place where computer assisted techniques make sense just like the case of flying a plane where the autopilot can take care of most things except when things go wrong and humans have to take over. I don't think anyone with adequate experience in healthcare and technology would make wild claims about replacing doctors with machines. There are plenty of places where machines can aid doctors and simplify processes at scales thereby letting doctors focus on more complex things. That's how the automobile industry evolved and so will healthcare. But doctor's will always be needed. They would just be doing different things or existing things in a different way.
There are plenty of people in this part of the world (Eg. India) where people are just getting to the point of trusting a ATM machine with cash withdrawal. Assuming machines become better than doctors at asking patients the right kind of questions, knowing when they are lying or just not able to communicate their problem, there is still a significant barrier and probably a generation of cultural changes that would make it acceptable to have a machine as your primary care giver. Diagnostics have been using complex technology for a while, so there is no doubt about technology improving the decision making process. However diagnosis is just one and probably a small part of the healthcare process. Actually improving the patients health requires following a treatment plan and adherence to lifestyle and medical processes. Most people just fail to comply due to behavioral reasons. This is where technology and so call 'AI' can play a role. By understanding individual human behavior patterns and guiding people to do small changes that would improve their health outcomes. We have been working on fixing this problem for people with chronic conditions like diabetes at http://www.janacare.com.
I would most certainly trust a well programmed computer to replace my doctor.
The examples given in the piece seem more to do with not taking a single measurement as an absolute diagnosis. Which I'd never program a computer to do anyway...
Personally I'd compare it to meteorology. Meteorologists used to interpret images from satellites and I suppose they made predictions based on a bit of science, but also a lot of heuristics. Now (still a supposition, I'm just a layman about this) computers do most of the heavy work. Yet meteorologists still exist : their work consists in choosing the computer, operating it, understanding the output and other stuff like that. You can't just give a supercomputer to someone and expect him to predict the weather. It's still a job for an expert.
Extending the metaphor. A computer might not predict a perfect storm developing because the model isn't perfect. Without the human there to see it, you miss something disastrous.
A doctor would be the same thing. Sometimes things just don't follow the rules we expect them to. Medicine is full of instances where diseases cause obscure complications. I mean the series House MD is based on these, there wouldn't be a show without thousands of these instances.
I think taking human error out of mundane doctoring would be great, and allowing people to specialize into the more complex areas where creativity is as much a part of the diagnosis as anything else then I think we would greatly benefit.
What's sad is if you read many of the cases used in the show House, the patients IRL died. The condition was found in the autopsy, because the doctors just kept following routine.
It can, but not now. Maybe in 10 - 15 years (or more).
The numerous available online (mobile apps) symptom checkers are a good indicator of the current state of how a computer can solve your health/medical issue. Computers need a lot more systematic data which is currently not available.
I was looking forward with amusement to reading such an article. Mainly because I realized that all the arguments the doctor would make would come down to decision making. He could pose vague rhetorical questions ("do you treat the MRI or the patient?") but at the end of the day it's about collecting data, analyzing it, making a decision (about diagnosis etc), and informing the patient. All the things that he discusses, therefore, are about justifying a human doing it instead of a computer. A human currently has the advantage in certain "soft" undefinable aspects of decision making, which we can call intuition. But if this is eventually formalized into a algorithm whose performance can be measured and quantified vis a vis human doctor performance, it will be like computer image recognition. In some cases it will be obvious that the computer has erred, but for the majority of cases it will be able to notice things humans don't, because of the sheer amount of collective experience around the world as well as correlations that humans may not have been able to detect.
(I should pause here and note the danger of confirming nontestable theories though, as Karl Popper once wrote, "Once, in 1919, I reported to him a case which to me did not seem particularly Adlerian, but which he found no difficulty in analyzing in terms of his theory of inferiority feelings, Although he had not even seen the child. Slightly shocked, I asked him how he could be so sure. "Because of my thousandfold experience," he replied; whereupon I could not help saying: "And with this new case, I suppose, your experience has become thousand-and-one-fold."" http://www.stephenjaygould.org/ctrl/popper_falsification.htm...)
Go back and look at all the arguments the doctor makes. Once you remove the vague handwaving, you're still left with the basic question: what are you measuring, analyzing and deciding? Can computer aided methods do it better and more accurately, and in the areas they can't, why not, exactly? - and can this be improved? If so, it's a matter of time.
I am not saying a computer can replace a nurse's bedside manner or a doctor's ability to make the final call but computer aided medicine would seem to be as superior to regular medicine as computer aided chessplaying has long been shown to be superior to both grandmasters and computers alone.
Even the maxim he quoted, that doctors should only measure things that may affect the diagnosis or treatment, is a heiristic not unlike the chess heuristics that guide grandmasters' thinking. But how often does a tactical line or particular "hidden" property of a position override that heuristic? And similarly, a computer can help uncover hidden correlations that a simple human rule may miss. After all, medicine thought for millennia that washing hands was just a ritual with no medical benefit until bacteria were discovered.
Agree that technology enabling better health care delivery is a great thing and still completely under-delivered. But as for technology replacing doctors...
I think in general it is misunderstood what doctors actually do... ie that they perform some kind of analysis to produce some kind of output. What they actually do, is take responsibility for the well being of a complete stranger. There are certainly a number of doctors that don't understand this, and have some other kind of agenda like making money, getting through another day etc... but ultimately, a doctor's 'special' role is to take responsibility. This is not really measured or paid for or even expected, but I would say that the success of a healthcare system relates to how much it helps or interferes with this process. People don't really get this, because there are very few jobs like this, and we are used to not having anyone take responsibility for anything. Government and business are masters of diluting responsibility in an ocean of bureaucracy. Accountability/transparency are not a substitute for responsibility either.
I don't know how a computer is going to take responsibility for my health, or my child's health. It seems a frankly ridiculous idea to me, like replacing the world's violinists with robots.
There is probably a role for a 'computer' to deal with nuisance (as in, currently a nuisance for the person who has to take time off work, go to a medical centre, wait for the doctor) medical problems like routine blood tests, rewriting a script, routine screening. These jobs are already being taken on by non-doctor health practitioners like physician assistants, nurse practitioners. This is a reimbursement issue really, and it can be done because:
1. There is strong evidence base for the activity (eg doing a pap smear for pre-cancerous cervical lesions, or vaccinating a baby).
2. It is easy to measure the success or failure of the outcome (eg the pap smear is done, or not done).
3. There is a very small risk of legal action. (so that the intervention can be indemnified in such a way that it is still economically viable).
Much of medicine is not like this however. There is not a clear evidence base. There are not easy to measure outcomes. The consequences when things go wrong can be significant for the patient and the doctor/treating team. Health systems are already ruined by an irrational focus on optimizing the wrong thing eg surgical waiting times, time to hospital discharge, profits, so lets not make the problem worse with technology that is apparently better at the outcome we've decided to apply to the cornucopia of human disease and dysfunction.
Disclaimer: I'm a doctor (in Australia). I'm sorry for the waiting times. The truth is, I don't have the skills to invent an adaptive appointment queuing algorithm. Isn't that an engineering problem?
I can give the answer without reading the article: No, it can't. Next time you have a heart attack or cancer or a broken bone, see how much your laptop or your iPhone can help you.
Perhaps the title should be changed then. If it has become something that is so well known and reviled, even when pointed out, then maybe we should change that. Perhaps a script that edits these titles out? I think Betteridge's law is true maybe 80% of the time (sorry no data here). If so, then thats enough to just screen out the bait titles.
The biggest problem with checklists, flow charts and software (one older example is MYCIN [1]) is adaption. If the medical practitioners don't want the systems they are going to fail. He also writes a lot about that in the book. He tried to introduce it into hospitals but the professionals often ignored the lists or just checked the boxes without actually checking the condition.
I personally imagine that a practitioner with a good system works the best or like Frederick P. Brooks said [2]:
> If indeed our objective is to build computer systems that solve very challenging problems, my thesis is that IA > AI that is, that intelligence amplifying systems can, at any given level of available systems technology, beat AI systems. That is, a machine and a mind can beat a mind-imitating machine working by itself.
[0]: http://en.wikipedia.org/wiki/The_Checklist_Manifesto
[1]: http://en.wikipedia.org/wiki/Mycin
[2]: http://www.cs.unc.edu/~brooks/Toolsmith-CACM.pdf