Can a Computer Replace Your Doctor?
nytimes.com
nytimes.com
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
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.
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.
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 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.
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.
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.
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.
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.
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..."
* 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).
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.
The English NICE recommend treatment based on cost effectiveness.
The "friends and family" test should be used by all English hospitals and that might give you some information.
I'm not sure how you find out whether a particular surgeon is an incompetant butcher.
only if it's pre-programmed to be > 40 min late for each appointment
Automation would be nice, but not sure how you could remove all the time sinks in US Medicine.
/not a doctor, just observations as an end user.
It annoys me, as a process-oriented thinker, that so many offices never get better at handling average daily volumes.
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.
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...
Medicine could become kind of like this.
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.
A blod post on this topic: https://medium.com/@iamhealee/you-should-use-symptom-checker...
The one glaring exception was the doctor I had in Mexico, where the medical care is fantastic and inexpensive.
(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.
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?
edit: clarity
Betteridge's Law of Headlines states: Any headline that asks a question can be answered with No.