Slashdot Mirror
Computers Shown To Be Better Than Docs At Diagnosing, Prescribing Treatment
Lucas123 writes "Applying the same technology used for voice recognition and credit card fraud detection to medical treatments could cut healthcare costs and improve patient outcomes by almost 50%, according to new research. Scientists at Indiana University found that using patient data with machine-learning algorithms can drastically improve both the cost and quality of healthcare through simulation modeling.The artificial intelligence models used for diagnosing and treating patients obtained a 30% to 35% increase in positive patient outcomes, the research found. This is not the first time AI has been used to diagnose and suggest treatments. Last year, IBM announced that its Watson supercomputer would be used in evaluating evidence-based cancer treatment options for physicians, driving the decision-making process down to a matter of seconds."
You might think it's funny, but Watson is still sensitive about screwing up on Jeopardy a couple years ago.
There's no -1 for "I don't get it."
I need money when an honest mistake is made
An expected outcome. First machines become good and cheap at performing manual labor, then it's lowly qualified jobs such as sorting stuff or basic accounting.
In a few years, liberal professions will fall. Our salaries (I'm a doctor) have been diving as more and more people around the world can afford a career and achieve a good enough level to perform as a doctor or an engineer.
Creative and risk-taking careers will resist for a longer time.
We can hope for a future of working machines and humans enjoying themselves. The other option will be cheap-ass humans with no way of earning a living whatsoever.
Big Pharma driven medical reps, motivated by insurance/lawsuit dodging and prescription kick-backs, or a heartless robot driven by a growing database of real-world evidence?
Best path remains: don't get sick.
I'm sure this is not what diagnosticians would like to consider, but I think the simple truth is that computers will negate our need for human diagnosis in the very near future. I'm sure we'll hear the same "The computers will be great *assistance* to actual doctors" but I don't think that's actually true. I know the computer is likely doing a better job, there should possibly be a few overseers but the vast majority of doctors shouldn't be needed, and frankly I wouldn't want them second guessing what's likely to be a much more accurate and unbiased diagnosis given by the machine.
I find this interesting, I was wondering when we'd reach the point where the accumulation of knowledge available in any given field exceeded the ability of the human mind to completely grasp in a useful manner. It's going to reach a situation where multiple experts on a given subject with a fair idea about related subjects are going to be the only unit capable of actually doing anything sooner rather than later - apparently in medicine at least computers have come to the rescue.
I suppose with the many specialisations in every area we're already there, the question is can we repeat the improved returns in areas like physics and chemistry.
I have 2 kids and my first one was always sick until a year ago. He was on antibiotic at least once a month
After a while I picked up on a few things
There is a simple 5 point check to see if the llllness is bacterial or viral
Once he diagnoses the doctor does not choose the drug. The physicians desk reference aka pdr tells the doctor what to prescribe
The dosage is determined by the drug companies depending on the child's weight
The pdr will have every known disease, illness, condition and the drugs or procedure required to treat it
In the early 1970s, Mycin achieved 69% accuracy of prescribing a "correct" treatment for a patient's condition, which was deemed better than human specialists.
The linked articles don't seem to include the absolute accuracy that it achieves, just the relative accuracy against doctors. I wonder if we've come any further than the basic expert system rules allowed 40 years ago.
An expert system is only as good as the information it is fed. Until we get machines that can quickly scan a human body and tell us everything there is to know about it, we will still need doctors to talk to patients, vet what they say, observe what didn't mention and ask followup questions.
Virtual Doctor you got leprosy
So, who do we consult for a second opinion? HP? Oracle?
I don't care what this article says, this is nothing new. My in-home psychiatrist in the 1990s was great at diagnosing my issues. Perhaps you were a patient of his, too? His name was Dr. Sbaitso.
"This one goes in your mouth, this one goes in your butt."
-JWR
The problem with doctors is their ego. Being a nurse for quite some time Ive seen countless doctors come through the hospital that do not treat a patient properly, ignore patient complaints, dont treat problems aggresively enough, are too passive, and so on and the vast majority of them refuse to do otherwise because of their ego. They are doctors, no one should question them and they hate it when you do and when you do they dig their heels in and dont budge. Doctors are also human so they get tired of their job, they get lazy, they stop caring after awhile, they worry more about what they will do later than the problem at hand, they stay up too late.
Ive also seen countless times people who swear by these doctors and will accuse of them of fault simply because the doctor has been nice to them. A doctor who is friendly can do no wrong in a patients eyes. Or a patient has some common problem that even a moron could treat, then when the doctor fixes them suddenly they become godlike.
Personally I would trust the diagnosis of a machine that has more information programmed into it than a thousand doctors do collectively from allover the world.
A machine can be a doctor because thats how doctors are trained, they are trained to be machines. They treat the problem, they do not treat the person. Treating the person is a nurses job. And treating the problem is simply nothing more than deduction. If patient has multiple problems then you take those numbers and it will lead you to the correct answer. Thats all being a doctor is, you look at the signs and symptoms, then you add them alltogether and you get the answer as to what the cause is. Its all a forumla and nothing more, there is no great mystery to it.
Serious. Logic kind of dictates this to have happened.
It does not mean that doctors are worse than the system; on the contrary: doctors are those fantastic deliverers of input to this expert system.
Of course, a medical system that has all sorts of medical information, treatments, medicines, and potentially millions of case histories provided by human doctors will evaluate chances, correlations and success probabilities much much faster and much better than the family GP. So what's new in here? Maybe that the computer systems are now available to actually do these calculations.
Apart from doctors who will understandably not want to be rendered obsolete (and they won't be -- the computer can only prescribe a treatment, not administer it!), the main objection that would be raised to this is "What if the computer makes a mistake?" For some reason, people are really bad at understanding that even though the computer might make a mistake, it will make mistakes at a lower rate than a human. This is the same problem with computer-driven automobiles. Yes, the computer might screw something up and kill somebody, but this should happen at a much lower rate than caused by human drivers -- however, because the rate isn't EXACTLY ZERO it is seen as completely unacceptable, even though this is an irrational position to maintain.
If you suck as a doctor or driver then these things are true. Unfortunately most people suck at everything.
I really good doctor can beat the odds just like a really good driver can. It's not even about being better than average, it's about actually being good at what you do. Very few people fall in to that group.
This sort of thing is just what big pharma wants, no human interaction and careful consideration, just a pill dispenser...symptom a + symptom B == Pill 2...
How much you wanna bet this thing always prescribes expensive non generic drugs and never tries the 50-70 year old known treatments that are usually the first steps in treatment before new expensive drugs are prescribed.
Also, anyone notice the change in medical advertising and communications, they never say "ask your doctor" any more, its ask your prescriber, or ask your provider...like they want to dis-intermediate doctors and are getting the public ready.
And will the system consider the patients age/cost to treat/insurance level/likelihood of patient paying future insurance premiums to make up for expenses?
It will if you program it to. Things like this are tools. As a relatively young doctor (resident) I welcome things like this. Every doctor I know uses reference material, some are printed on dead trees and some are electronic. Today, there's not much difference. But the point it is that there's too much medical knowledge for one person to keep it all in their head at one time. If something like this were to come to market it wouldn't be replacing doctors, it would be augmenting them. Machines do what we tell them to, always have and (hopefully) always will. False rivalries like this completely miss the point. I would love to have a computer algorithm that could correctly diagnose 99% of the time even if it were flagrently wrong the other 1%. That's why humans are in the loop.
... is that the computer doesn't have, literally, hands-on experience.
For example... you have stomach pain. Okay. Where? Does this hurt? Does it hurt if I poke this? Do you have a fever? How high? ("Oh, I don't know, maybe XX" ...)...
In other words, you still need someone who *knows* what to do to help diagnose simply to *get* the information. You need someone trained to know where to poke, where to tap, where to see if something hurts. And then, of course, to try to wheedle the real information out rather than the confusing answers patients (note that I am not a doctor, I am a patient) give. :)
As a diagnostic tool, I think it would be a huge benefit. An individual simply can't keep up to date on everything, don't know everything, and can't research everything. Having a diagnostic tool that *does* do all that would be a huge benefit. There would be options and diagnoses that perhaps the doctor had not thought about. Additionally, the diagnostic tool - I hope! - would not just say "Oh, it's this" but would perhaps say "Well, it could be this, and you need to find out this information first ... or it could be this, and for that you need to do these tests..."
The idea that a patient, combined with a non-trained non-doctor, can somehow come up with all the right answers/information and a machine can simply diagnose it on the spot seems like a very simplistic view. Perhaps that works with colds, the flu, or warts... but for more complex things... it seems diagnoses tend to work on a "it could be this; let's do more specific tests to find out if it is" type of investigation.
Perhaps Drs could use a tool like this similar to how weather predictions are done. Run the model(s), and use human analysis to weigh the data.
"This mission is too important to allow you to jeopardize it." -- HAL
In my experience, most doctors don't really think about problems, but just look at historical remedies and prescribe those particular drugs. They just regurgitate what is in the books. If you have something new or multiple problems (I've been dealing with this for a year now, with no relief in sight), I think I'd rather listen to watson. These guys have cost me thousands of dollars and all of my vacation time and I'm worse than when the problems started.
Electronic medical records are unfortunately not up to where they should be. It baffles my mind that pertinent information is still kept in a paper based chart bigger than a phone book for the majority of health centres. My hospital is trying to make a big movement, with a large optimization project. All patient armbands are as they were before, except they also come with a barcode. The idea is regular meds are linked to a patient barcode, and when both armband and med packet are scanned medication errors are prevented. Equally this system serves to immediately chart what the patient received and when.
This is important, as histories and results are all fed into the same system. The gold standard of treatment is listed, along with alternative treatment options, in a drag a drop like system. While this has the potential to create "brainless" medicine, it also prevents things from slipping through the cracks. For example, say a patient comes into the ER with abdominal pain. The system will spit out a differential diagnosis, in order of probability and list the required labs, tests, medications and interventions for each diagnosis. As test results come back, the diagnosis list diminishes. Equally it makes sure patients are receiving that dose of antibiotics that evidence based decision making has determined is the best course of action in most circumstances. It also knows that this organization has switched to a new antibiotic as the other previously used four months ago was known to cause minor complications in some patients, but knows that your patient has an allergy to this new medication so it warns you and provides a list of alternatives.
While this system is still in the process of going live, I am very hopeful it helps the healthcare system. I don't believe it will fix the system however I do believe medication errors, outside of emergency situations such as cardiac arrest, will be reduced.
Add in a strong AI and a large majority of actual diagnosis, rather than suggested diagnosis, may be accurately determined.
The big difference here is in technologies.
Mycin used a bunch of rules to arrive at medical decisions. These had to be manually entered by medical experts which is time consuming, difficult, and outdated by the time it is completed. It was a narrowly focused, brittle system that strived to emulate the logic of it's designers.
The newer approaches (like the kind from this article) build their own probabilistic models of medical diagnosis and treatment based on statistical analysis of data. Lots and lots of data. Data that would have taken hundreds of years to process in the 1970s. From this, they learn what works and what doesn't and under what circumstances independent of the knowledge of its designers. In effect, it can be better than its creators.
My God can beat up your God. Just kidding...don't take offense. I know there's no God.
Probably because its a summary of a press release, rather than the actual paper. This was a computer simulation exercise - a model. No patients were actually treated. There was no prescribing. There were no superior outcomes. Its all hypothetical.
Could an AI outperform an MD? Sure, could happen and probably will eventually. But this model doesn't show that. It just shows that its easier to model health care in silicon than the real world. Remember we can cure cancer in mice - models aren't the same thing as reality.
+--------------------- You idiot! I told you we were facing the wrong way!
Some may still have jobs if they can hang on to customers who want real in person or over the phone service.
I regret going to medical school
As is often the case with new medical tools designed to avoid common fatal mistakes.
WTF does your comment have to do with airports?
It has been reported that some scientists were found dead after self administering a dangerous drug based on some computer algorithm. Could it be that these are same scientists?
Most doctors are pretty bad. In fact, most that do not perform surgery are no more than technicians who have a lot of student loan debt.
Cheap, automation is one reason we need healthcare. Before escalators, elevators & cars, we all got much much more exercise. Before mass produced candy bars & fast food, we ate much much better.
The UN had a computer based medical diagnosis tool for use in Africa more than 20 years ago. That said, its a helpful, useful tool. I trust Docs to use it, and understand that they will take the advice of the computer as advice, while keeping clear thinking and all of their other knowledge into consideration. It can probably help a lot though.
I've read similar stories about set steps being better able to deal with certain medical emergencies than just 'doing what's standard', and thought the IBM Big Blue computer and the way it operates would be a boon to medical sciences...and this would have been an amazing episode of House: where he competes against something like a medical version of Big Blue. =)
Is why I will always want a doctor rather than a machine at this point. As another poster put, the machine augmenting the doctor is what I'd prefer. Give the doctor an earlier version of a tricorder + the current equivalent of an Enterprise database, let the doctor deduce it thereafter.
One thing I would like to see in these AI systems is occasional variations in their recommendations (for each given symptom set) for experimental purposes. It would be easy to code, and the results could be used to help choose the best outcomes. If two different treatments appear equivalent in outcome, the AIs should prescribe them randomly in a 50/50 ratio, modifying the ratio as new evidence arrives.
(This sort of thing is analyzed by the online advertising industry as a "multi-armed bandit")
Doctors are like car mechanics, and vary in quality every bit as much. Most doctors are ONLY good at that which they commonly experience. Anyone attending a GP with 'interesting' symptoms that hasn't first done a little online research is a complete fool. Saying "put all your trust in the doctor" is like saying "put all your trust in a teacher" or "put all your trust in a police person".
Of course, getting a computer to teach or do policing duties is VERY difficult. Getting a computer to diagnose illness, on the other hand, is completely trivial.
One, the computer can literally know EVERYTHING. This is a good start, and a massive advantage over a Human doctor. Now the patients symptoms need to be linked to the giant database of illnesses. This simply involves a questionnaire as long as is required. This approach is also a massive advantage over the Human doctor, who will be trying to wrap up the diagnosis in as short a time as possible. The computer can just keep asking questions while more input seems useful.
Of course, some of the symptoms the computer needs to know about may be better or only available from a physical examination by the doctor. Clearly, being rational people, we would like to see the Human and machine systems fused together optimally.
People who worship doctors and surgeons in the same way as people who worship Catholic priests and Bishops are a nightmare for society. Americans, in particular, are trained to elevate medical personnel to some god-like status, rather than see them purely as mechanics of the Human body. This is GOOD for the status, and thus the salaries of doctors, but lousy for the efficiency of the medical community. When your medical facilities exist purely for reasons of obscene profit and personal gain, true quality of engineering won't exist. Instead, you'll get something designed to look good, not be good.
Considering my luck with Western Medicine, I would have been better off with a dice roll anyway. I look forward to AI doctors, especially since they'll probably be a mite cheaper than the real ones, and not constantly late for their own damn appointments.
In Soviet Russia, dot slashes YOU!
... the love?
Meh. You can be replaced by sufficiently advanced algorithms. I'm mathematician, FFS. There are already automated theorem provers which can solve undergrad-level problems. As computers evolve, they'll be just as good as people and loads cheaper than people at everything we do, up to and including the creation of art. If we progress to the point where all of our jobs can be done by computers... what should we do? At the point where artificial intelligence becomes genuine intelligence, it will rapidly outpace human intelligence. This is evolution. We are breeding our replacements.
Hmmm...what if there is an irreducibly non-algorithmic component to human intelligence? That would invalidate the Church-Turing thesis that your assertion seems to be relying on. Roger Penrose thinks it would at least, and he makes a pretty good argument for that position, and against the Strong AI you are hypothesizing. He also defends his argument pretty well, too..
I have cancer.
http://www.snpp.com/episodes/4F21
"The wars of the future will not be fought on the battlefield or at sea.
They will be fought in space, or possibly on top of a very tall
mountain. In either case, most of the actual fighting will be done by
small robots. And as you go forth today remember always your duty is
clear: To build and maintain those robots. Thank you."
-Styopa
I call bullshit. There is no such diagnostic software.
Here are the names of some companies producing the "non existing software":
DiagnosisPro
Isabel Health Care
Here's a recent article on the New York times about the subject, Isabel Health Care has been around since at least 2006, when we recommended the software to a local hospital.
There is no complete database of every possible disease with typical and atypical presentations.
Complete in that it contains every disease known, and no you do not need agreement about the symptoms to enter them in a database. The reasoning process handles that part, you simply enter "some reports say these are symptoms, some other reports these other symptoms, and moreover it is always possible that a few symptoms maybe absent".
From the NYT article, here is one example of a "non such database":
Since the 1980s, Massachusetts General Hospital has been developing and refining DXplain, a program that provides a ranked list of clinical diagnoses from a set of symptoms and laboratory data.
And Derivatives; don't forget the market genius of Derivatives - oh! - and Subprime Loans too!
This represents my field (Biomedical Informatics) so poorly, I actually feel shame reading the CW article, the IU article, and the journal article. From the CW article: "This is not the first time artificial intelligence has been brought to bear on healthcare." Not a bad statement, until they cite Watson (and an ongoing research project which has still produced nothing of value) as their evidence. How about the fact that the journal that published the IU study is called....wait for it...."The Journal of Artificial Intelligence in Medicine" and has been around for almost 20 years? How about MYCIN, eMYCIN (a generalized form of MYCIN which could be applied to any domain), INTERNIST, DxPLAIN, ISABEL, or the whole host of fully developed diagnostic expert systems which were developed over the past 50 years and which can outperform physicians? Maybe you ought to think about _why_ these aren't used? From the IU article: "..we believe that the most effective long-term path could be combining artificial intelligence with human clinicians." Holy crap! IF ONLY THERE WAS A 50 YEAR OLD FIELD OF STUDY CALLED "CLINICAL DECISION SUPPORT" PREDICATED ON THIS EXACT IDEA! I'm so glad myself and my colleagues have the support of these two geniuses from IU. Same article: "The framework here easily out-performs the current treatment-as-usual, case-rate/fee-for-service models of health care." That's also fantastically obvious. Replacing the fee for service model has been in the works for decades. It is a well-known and well-studied fact that we order a ton of unnecessary tests/procedures and this is strongly tied to the fact that we compensate healthcare providers based on the number of tests/procedures performed, and not based on patient outcomes; this brings me to my next point... From the actual journal article: "the goal here (i.e. optimality) is defined as maximizing patient improvement while minimizing treatment costs." Nice one. Why didn't we think of it earlier? Oh. That's right. Ask the Oregon Health Plan how "cost effectiveness" gets translated in the popular media (I'll give you a hint: death panels). Look. I can keep going all day on this naive research and reporting. Suffice it to say, you can't compare current doctor's performance to any model that assumes 0 liability (the AI model does not feel compelled to order tests to cover its ass), no fee-for-service setup (the AI model has no financial incentive to order additional tests and procedures...and in fact has the opposite incentive since it is measured in terms of unit cost), and an explicit cost-effectiveness evaluation model (something that will earn a flat-out revolt if you actually try it), UNLESS (and this is a BIG unless) you are also going to propose a realistic plan for getting rid of medical legal liability, fee-for-service, and widespread negative public opinion about cost-effectiveness measures in healthcare. To make it clear, I think the research is useful and should be explored further. I think researchers whoring themselves out for attention, failing to acknowledge limitations in their study, and allowing their institution to print sensationalist crap about their work is despicable.
Dr. Lexus: "Don't wanna sound like a dick or nothin', but it says on your chart that you're fucked up. Ah, you talk like a fag, and your shit's all retarded. What I'd do, is just like... ha ha... like... aha... you know, like, you know what I mean, like... haha... Don't worry scro'! There are plenty of 'tards out there living really kick ass lives. My first wife was 'tarded. She's a pilot now." - from "Idiocracy" (2006)
I seem some dumb crap being said here. Sadly, Slashdot updated their site and I can't reply to posts because of IE compatibility issues. (And yes, I don't have a choice but IE.)
***
Farmers did not steal Mosanto's corn. Their fields were infected. Mosanto's defense was that there was too much GMO to have just been from infection. However, we've seen plenty of evidence that supposition was wrong. Canada and Europe which have outlawed GMO grain are still fighting to eradicate it. (And will likely loose.)
But the dumbest one i hear is that the use of glysophate is proof of theft.
GMO corn is glysophate resistant. That means it has greater resistance than typical corn. But Roundup existed and was used LONGGGGGG before GMO products were on the market.
It's just now, you can spray a lot more without risk to your crop.
I'm a doctor, just retired, and I can say that most ailments don't need any treatment. Time, diet, sunshine, exercise, sleep, social support, will cure most of what ails us. I'd wager that if the health care system disappeared, the overall mortality of Americans would decrease (see below). We have a nation of hypochondriacs and a professional class that would like to keep the status quo. Beware the medical industrial complex!
http://healthwyze.org/index.php/component/content/article/502-death-rates-drop-when-doctors-go-on-strike.html
So if I just get interviewed by a nurse that will enter my "symptoms" into the diagnosing computer I could game the system to prescribe pain medication and opiates...
Profit!!
This one goes in your mouth, and this one goes in your butt. No - wait THIS one goes in your mouth and THIS one in your butt...
...
At the end of the day though, it's up to the doctor to make the decision to accept or reject the outcome. I don't think you can program human intuition.