How about fewer doctor’s offices fending patients off to physicians assistants? I cannot tell you how many times I’ve been asked if I would accept an appointment with X, Y or Z, only to learn the suggested practitioner is a PA. The idea that someone with an undergraduate degree is as capable as a physician or nurse practitioner would be laughable if it weren’t so dangerous to patients.
19
The AI technology is targeting diagnosis directly, and care only indirectly - to the extent that care is linked to diagnosis. In much of healthcare, the so-called "art" lies in recognizing that the basis of scientific method in healthcare - setting all other things (but treatments) equal - needs to be moderated by an understanding of how any given patient with a diagnosis is not necessarily equal to others with the same diagnosis. We do not always do X because of diagnosis Y. Sometimes we do X despite Y. If the issue is to treat a disease like type II diabetes with attendant risks for co-emerging multiply-determined problems like a broad array of cardio-vascular conditions, that is one thing. But if we are treating a PERSON with type II diabetes and risks for that other stuff - that is a far more complex and dynamic problem - one for which health care providers are best equipped and the smartest AI is less-well equipped. So if we could offload some of the diagnosis and plowing through the electronic health record so that providers could care for people with X, rather than providing treatment for X, that might be good. And if the AI can help us detect low vs high probability trajectories and assist in managing care longer-term --that might not be so bad. After all - doctors go to medical school, not charm school - and many patient prefer telehealth to face-to-face engagement even if the provider is engaging.
5
Computer aids to help physicians have been around for a while. Dr. Larry Weed who taught me at the University of Vermont College of medicine had a program that worked well. But unfortunately it was not used extensively. I am now teaching Emergency Medicine in Elmira, New York and would be glad to test, or use one of the AI programs in our practice. So if you are out there and have a system you would like to test let me know.
5
AI, Physician assistants, Nurse practitioners, or any other augment and extension of the physician will be tremendously beneficial to medical/surgical specialists. Why treat 20 patients per day when you can virtually or by extension treat 150-200. You'll still need a physician to take on the risk of decision making, and you'll still need physicians to explain and convince people to proceed with additional testing and procedures. In theory cost should go down, but access to care will increase and the tech companies that create and monopolize AI will keep costs inflated. If people want lower costs, it comes in only 2 ways- 1. Ration non essential care; 2. Reduce the layers of redundancy or waste in the system, ie millions of workers who handle insurance, hospital, and office administration.
5
Interesting...
Six of the ten most-recommended comments focused on use/misuse of the title of "Physician's Assistant"...
Meanwhile...
https://it.toolbox.com/articles/artificial-intelligence-drives-medical-imaging-to-new-territory
“...It’s only a matter of time before every X-ray machine is connected to the cloud and one human doctor per hospital puts his hand on your shoulder when he reads you the output from the AI algorithm,” says Nanalyze on its blog about medical imaging start-ups integrating AI into their strategies...
.....
“...One of the best-funded companies, Butterfly Network, is working to apply machine-learning algorithms in order to help patients conduct ultrasounds using their own iPhones. The company proposes to fit all of the imaging equipment required for ultrasounds, known as the iQ, onto a single silicon chip...
“...The breakthrough technology works by tethering together thousands of ultrasonic speakers to create a three-dimensional picture of images inside the body, and has already received clearance for both obstetric and cardiac exams...
“...The result is an ultrasound that can be conducted by a patient, vastly expanding the portability and affordability of the medical technology...
PS
https://nypost.com/2019/02/05/health-device-firm-leases-2-floors-near-madison-square-park/
3
Who is going mourn the skilled professions, when they are conquered by AI? Nobody mourned the plight of shipbuilders and machinists.
AI is already better than radiologists at reading charts and finding warning signs. Who among us would place our health and well-being in the hands of inferior diagnosticians?
Hard to be sympathetic towards an industry that employs so many underhand tactics to pad bills and sent patients into bankruptcy. Healthcare is run for the benefit of shareholders and CEOs.
11
I had a miscorrect diagnosis a couple years ago for a condition that was probably made worse as a result (I was finally diagnosed, but only after an ER visit!). My wife died a few years ago after a long, complex illness, which was not correctly identified for several months. I will always wonder if she'd still be alive if she had received the correct treatment early on. I welcome anything that can help reduce this problem.
4
When I have a medical concern, first I consult with Dr. Google. Last summer, Dr. Google helped me diagnose Lyme disease in myself. I read a few articles and confirmed that it's a serious problem, so later that day, I was in the minor emergency clinic with a human doctor who confirmed the diagnosis.
Dr. Google is the augmented medicine that I can access. I’m sure Dr. Google is used by millions of people every day and it’s extremely useful.
New augmented A.I. medical systems are tools to help doctors diagnose and treat disease. It is like a doctor’s consultant, confirming an opinion, or providing alternates worthy of consideration. But it is nothing more than a tool and the human doctor remains responsible for final decisions.
A great thing about A.I. is the feedback loop. Today, doctors follow their patient’s outcomes to learn and improve. A.I. systems also constantly use the feedback loop to learn and improve. The difference is that a doctor has a limited data set; A.I. has a worldwide data set.
Medicine is expensive and millions of people lack access to medical care due to high costs. A.I. will help doctors work smarter and faster, and that will help to improve quality of care and reduce costs. It’s a win-win for doctors and patients, and it can’t come soon enough.
9
When all of the jobs go to robots
What will people do?
This isn’t going to end well.
6
> "as good as a doctor"
In China. Where traditional medicine, such as energy healing, is considered viable. No thank you.
A "neural network" is nothing more than a collection of interconnected information. It is *not* an AI. We do not have an AI at this time.
For now, we have some informatiin that may guess better than Chinese doctors. When we do have a real AI, I'll wait and see if it's more accurate than American doctors.
4
A bit of AI history: 1970s-era AI focused on "expert systems", systems of logical rules and hard-coded deductions about facts entered into the system. A prominent example of it was Stanford's MYCIN, an expert system for identifying bacteria that caused infections. MYCIN asked a series of questions about symptoms and came up with both a diagnosis and a suggested treatment. It was never used in practice, not because it didn't work, but because of both the ethical dilemmas presented by computer diagnosis and also the difficulty of deploying a computer-based system in medical clinics of the time.
1
First Issue: The headline in the hyperlink describe's AI's success as Physician Assistants: I am a Physician Assistant, which is a specific position of a highly trained healthcare provider that works with Physicians, see our own patients, make diagnoses, prescribe medicine and perform procedures. We are not being replaced by AI, as the headline will have you believe
Next: When comparing the two arms of the study, diagnoses made by AI vs diagnoses made by human physicians, who determines who is correct? I would imagine a human physician. How does this take the human bias out of the data?
Finally: We try our best to practice evidence based medicine, based on years of research, data and numbers, but there will always be an art to medicine, based on individual circumstances, social factors, insurance, etc. I work in an Emergency Department, and a plan I make for a homeless patient with no resources will often differ drastically than the plan for a well to do person with a home, a primary care doctor, and a ride to make a follow up appointment, even if they have the same diagnosis
18
@Jesse
It isn't intended to replace doctors, but if it could quickly make a suggestion to a doctor they may occasionally go "Oh, I hadn't thought of that, let's do this additional test to rule out that possibility." All of my doctors are at Weill Cornell, a top hospital, and my GP brushed off my symptoms for a decade and I had to diagnose myself by demanding certain tests. When I went back for my annual checkup she had already forgotten the disease I had been diagnosed with by her 6 months prior. I have the ability to stay on top of my doctors and demand better care, but everyone on the planet may not be so fortunate. Having more safety mechanisms in place is only a good thing.
10
@Jesse
I'd love to know how the art of medical practice and/or diagnosis is tied to, or influenced by, or determined by insurance.
I don't think you meant that in your string of words. Or I certainly hope you did not.
If you did, it is astonishing to me, as someone who trained in the medical field, albeit long ago.
And if I ever thought my physicians were engaged in the art of medicine based on which insurance coverage I carried, I'd be outraged. And out of there.
The two most most important contributions to airline safety are the autopilot, which is a network of computers that flies the plane and (depending on the craft) lands it; and the computer-driven GPS networks used by air-traffic controllers to keep planes from crashing into each other.
Decades ago, the resistance to both those technologies--by pilots, controllers & even airlines--was just as ignorant and just as self-serving as the resistance today by medical professionals to AI diagnostics. Like the resisters decades ago, the resisters today don't want to save lives, their want to save their jobs.
11
There is a serious logical fallacy underlying this study: Who determined the "correct" diagnoses that the AI system was trained on, and against which it, and the panels of physicians, were scored? Panels of physicians, no doubt, who, as we know, frequently make mistakes. This is not like training an AI system to interpret, say, sonar data, when we know during training whether or not there's a submarine down there. Expert radiologists can disagree; that's why it's not necessarily malpractice when a planned appendectomy reveals a patient with a healthy appendix.
BTW, as a Physician Assistant, I take the controversy here about the ambiguous headline as further evidence that we should have been called Physician Associates all along.
9
@jan for a few years in the 1960s and 1970s the graduates of Duke,myself included, were Physician Associates. then it went by the wayside. interesting...
2
@jan
Looks like the editors heeded your comment. The headline now reads: "AI Shows Promise Assisting Physicians.
2
I would have loved this 25 years ago, before having 5 painful surgeries for a disease I didn't have. In the past seven years as symptoms shifted into high gear, I've seen 4 gynecologists, 4 orthopedic specialists, 2 primary care physicians, 2 physical therapists, 2 masseurs, 2 acupuncturists, 2 pain specialists, 1 gastroenterologist, 1 chiropractor and a Feldenkrais practitioner. That's 21 practitioners who all got it wrong. Can I have my time and cash returned please? BTW, the right answer is: Fibromyalgia.
8
Here we go. The continuing pursuit of maximizing profit in the medical “industry” substituting technology for wise, personal care, which has been deemed “too expensive” to invest in.
3
@Claudia
That's funny, I drew a completely different conclusion from the article. I find doctors to be semi-competent, at best, when it comes to diagnosing disease they don't often come across. So anything that betters the odds of a patient leaving the doctor's office knowing what ails him, including a computer diagnosis, is a vast improvement, in my book.
9
A. I. computers are inherently objective, at least now, and not tainted by our de facto criminal health care system of be rich, don't get sick and/or don't have a bad life event (pre ACA) like big pharmaceutical/hmo execs. or many doctors.
Doctors or specialists can save your life or help you but by and large they tell you what you already know, can't help you or worse harm you and/or rip you off.
I have had more luck with PAs., pharmacists, physical therapists, nurses, internet etc. then doctors when dealing with non life threatening conditions.
The latter are at the bottom of the food chain in our de facto criminal health care system and by and large just want to help you.
5
@Paul, "by and large" ....lol
2
@katiatt- thank you for your reply. Yes I put "the good guys" ie doctors who save your life or really help you at about 20%.
Re bye and large, the opposite with the non doctors, they are low on the de facto criminal medical chain and are just looking to help you imo.
1
So the Chinese are less stringent on sharing data...if the insurance companies in the US were not lobbying to be able to uninsure millions, Americans would be more inclined to share. To do so now could literally cost you your life.
3
I think that using "Physician Assistant" in caps implies specifically my profession and is not what this article is about. As an assistant to the physician is obviously a " different kettle of fish" and I resent the caps as I have been a practicing PA , providing diagnosis and treatment for 37 years. The headline is misleading and I am surprised someone at the Times did not catch this error.
8
So now Physician Assistants are going the way of fast food workers; replaced by technology. What's going to drive our economy when we've all been replaced by technology? Who's going to have any money to spend?
3
Physicians have already given up their profession by being bought and sold by hospital systems and their executives. Physicians have already handed over their responsibilities to nurse practicioners and physician assistants. So as employed line workers they will be minimized more by AI because it will be cheaper. There is no one more to blame than the physicians themselves. We will all know that the healthcare we recieve is better because the healthcare system tells us so with their marketing. Oh the humanity.
8
This article shows little understanding of what a human physician assistant really does besides run tests. Most are very good communicators, listeners and filters for patients’ pain, sickness and suffering complaints that don’t always get passed to the doctor who can appear somewhat intimidating to the patient. They usually have valuable empathy and understanding skills critical in diagnosis and the time to use them. Somehow I just don’t see this in an AI anything.
6
Reports like this are expected from the media. Little supporting data, but endless projections of robots taking over diagnoses from supposedly incompetent physicians. I am a radiologist and I have seen much wild speculation. When I read the actual papers closely, they are close to useless.
In reports like this, the word "accuracy" is tossed around as if it is meaningful. It is not. Experts look at numbers such as false positive fraction, false negative fraction, and most importantly the area under the Receiver-operator curve. NO one mentions these real measures in these articles. There are issues regarding the prevalence of the disease in the population. Never mentioned. There are issues of what diagnoses are chosen. Some are quite easy. Others are difficult, particularly when an unusual disease may have symptoms shared with many other diseases.
The real danger in these articles is that administrators and politicians (who control medicine now) will begin to believe what the manufacturers of these systems are selling -that they are ready to be put into operation now. Hardly. AI cannot even organize the medical record for us so the we don't have to page through endless useless information to find what we need. That would be a doable, useful AI project, but it is not sexy.
Other physicians writing here have noted also - we have seen this before with other systems over the past 20 years that promised the world and are no longer used.
Readers - don't believe any of this
12
@Dr. anonymous - yes, and I also wonder if it takes into account base rates as per Baye's theorem
1
@Dr. anonymous/Agreed. I hope AI will become seamless enough to reduce redundancy and improve time management, but it is a long way off. Every residency I interviewed at stated they would be implementing EMRs, and 7 years out, some of them have begun implementing EMRs, but mixed with paper charts as well. A mess. You cannot teach a machine, at this point, (who knows what the future holds?) the art of medicine. a great deal of medicine is empathy, reassurance, the human connection.
1
@Dr. anonymous This clearly strikes a nerve in you. Radiology is actually the most promising Application of AI and hopefully you and your contemporaries will be out of business sooner rather than later
1
Calling neural networks and other Machine Learning techniques “Artificial Intelligence” contributes to the misleading hype around them. Although neural networks were originally intended to model neurons in the human brain, that goal no longer constrains their design. Machine learning, although it is considered part of Computer Science subfield of Artificial Intelligence, is more accurately described as Computer Science’s own version of statistical modeling and inference. Neural networks, like all machine learning techniques, have strengths and weaknesses, and despite impressive improvements in recent decades, they are still pretty far from achieving anything like human intelligence.
6
All I can say is that it's about time that the triilion dollars that the US health care "system" has wasted on the collection of electronic medical records is finally being put to some good use.
4
Why are we spending so much $$ on developing AI when all of that $$, or most of it anyway, could be spent on much better educating our entire American population about what it means to have good health knowledge in the first place.
I'm sorry, but I don't see AI in the current ramping up of it's application as being the end all be all the already fabulously wealthy want it to be. I see it as just another way to flip people out of their jobs, without providing any monetary compensation for their demise.
3
@RLC
Even people who are fully informed of the health implications of their acrivities do not modify the harmful ones, as very few are able to apply rational processes to their habits. Hundreds of millions are spent encouraging people to drink alcohol, smoke, eat junk food, and sit around.
A better diagnositic tool wouldn't "flip" anyone out of a job, either. Likely make them more effective.
2
@seattle expat- If I can't figure it out myself, or parts of it, I want a physician to diagnose me, thank you. And if he's stumped? He can confer with another human DO, MD. Incidentally, we already have AI. It is called Google. And we all have access to it- professionals and laymen alike.
1
The A.I. probably has a bad bedside manner.
3
My stayed in the US, while I am living in Berlin, Germany.
I know it became a joke to compare both health insurance and health care systems, but you would not imagine how the Germans handle misdiagnosis.
When unsure about a diagnosis, they simply consult two other doctors. No need for AI and whatnot. And it is all paid by their, or our, health care system.
Don't get me wrong. AI is great, it is the future, but misdiagnosis handled by an AI is just another diagnosis which is, in reality, even harder to explain and track. All AI solutions, especially those that base on trained AI's, are, in case you didn't know it yet, act often like a blackbox. The magically trained AI blackbox then spits out a diagnosis or pattern or solution, and you are hoping that it fits.
I therefore have my doubts that AI will ever be a better misdiagnosis help then, say, just consulting multiple doctors.
7
AI in medicine, there has been talk about this since 1983. Still will have same vacuous talk in 2055.
I can out diagnose any moronic AI device!
MD surgeon
4
This all sounds wonderful. However, maybe it's me, but I have to wonder if they brought this to the US would the technology built into the software and system be mining data about Americans that is then sent back to China?
The Chinese company Huawei when found to be doing just that, sadly should make us question if this will do the same.
3
As a healthcare provider I can see pros and cons. The differential diagnoses from a machine would obviously be much wider for example whether a lesion on a CT scan is inflammatory, neoplastic, or infectious in appearance. But then we have to be prepared for the ramifications: a search for each one of these would be incredibly costly and time-consuming and harrowing for the patient as well as for the provider, because one needs a good clinical history to pursue the few probable diagnoses rather than going on a hunting expedition.
5
I find it interesting that this is being touted as the “next great thing”. When I was in the military, I did work on AI in the mid 80’s with a colleague on this same topic. At that time we found that algorithms were typically more accurate than humans when diagnosing a wide range of illnesses. As technology has improved, and processing power has increased exponentially, this can only get better, as these systems can process massive amounts of patient data. While a single health care provider depends on their personal knowledge and experience, AI systems can draw on unlimited amounts of data to make diagnoses.
But, let’s accept the reality. We worked on similar systems 25 years ago, and we saw great promise, but we still seem to be having the same conversations as to whether AI can be “trusted”. It’s not about the technology. The fact is that our health care providers feel threatened and have pushed back on these advances. Until they accept that AI can be a valuable tool, progress will be hindered.
9
The next thing that has to happen is the digital standardization of the physical exam. Patients will go to an independent center for their physical. Heart sounds, breath sounds, pulses, neurologic findings, and abdominal ultrasounds will be acquired and digitized. Inspection will be enhanced by digital photography for computer analysis.
The patient will arrive at their physicians office where laboratory, radiologic, and physical analysis will be compiled. The physician will manually confirm any positive or suspicious findings.
Until we get a standardized, objective physical analysis we'll never fully incorporate AI into medicine
3
As a physician it is exiciting to read about AI technology in medicine. I can see it being used as a diagnostic tool to guide me in decisions. To note, physician assistants (PAs) are health care professionals that diagnose treat in many aspects of healthcare including operating in surgery. it is one thing to consider using AI technology but it is insulting and presumptuous to consider that AI would replace clinicians.
21
This can only be a boon for patients, as doctors are widely claimed to have difficulty forming a coherent patient narrative from digitized medical records.
1
Data is being turned into money the world over, as just one individual in a sea of 7 billion each one of ours data is not worth much but in the big picture data is very valuable. Interesting times ahead for societies as we decide who can collect and how that data can be used and who gets to profit from it.
Machine learning or A.I. Is about to invade areas of work or labor that were historically immune from machine competition. Some very powerful industries are about to suffer from disruption that in the past only affected lower paid less powerful workers. I anticipate a battle royal over these advances as the powerful try to protect their pay checks.
5
The introduction of AI into the mainstream medical practices is already happening,,albeit on limited areas-like Cardiac stress tests or monitoring,looking at the images generated by CT scanner and MRIs We arrive to our destinations in a plane by use of multiple tools of technology that were not voted upon! While it’s true that sole reliance on technology that’s not perfected and widely tested can have serious consequences like the spectacular crashes of Air France in Atlantic and recently Indonesian airlines in October last year. In rural areas with scantily existing medical services if can be augmented by improving health technology then we all need to embrace it. Papers like this article if done in larger number of places and involve more people,then we will see the net benefits and scope for improvements. Let’s not be luddites and reject new changes that can help humanity
2
Soon there will be robotic surgeons doing heart transplants, much more reliably than any human. Just like self-driving cars, lives will be saved; but, many will resist the changes.
The author did not discuss accountability.
Machines won't be held accountable for wrong decisions - people will. The professional status of health professionals is as much about their accountability as about their capabilities.
A common perception of AI in medicine, as reflected in the comments on this article, seems to be situations of either doctors or machines. This is not what Dr Zhang, who is quoted in the article, says. He says that machines will augment what physicians can do.
Machines can do things that physicians can't.
Computers have always been able to read very large volumes of data. The new algorithms now enable them to draw conclusions from very complex relationships in the data. People are good at drawing complex conclusions but not at considering very large volumes of data. Physicians are also scarce and expensive.
The only statement that I may not agree with is the suggestion that machines be used for initial screening. That is already a decision that can be harmful and it should be made by a physician, or physician assistant.
Until machines will be held accountable, which may never happen, medical decisions should be made by humans.
We should appreciate the collaboration of machines and humans.
8
I want to know...of the 1 in 20 how many will die? Because in the UK they cannot diagnose sepsis and many have died from this, regardless of age of the patient. I think it is sheer incompetence. And since if anything we are in an age of increased diagnostic tools this should be improving, even without AI.
3
Whether out of implicit prejudices or ideological blinkers, let's not run down this great initiative to bring at least a semblance of competent medical care to isolated and far flung areas of the world that haven't seen a qualified medic in eons. Such an AI system when used as primary care in such remote places would prove to be a boon in achieving totally accessible health care for all.
Most minor disorders would get solved at this juncture, leaving only the more complex ones to be taken care of at the next higher echelon, manned by qualified medical personnel. Moreover, such a primary care tool will also enable a better chance to diagnose maladies in such isolated regions where people are accustomed to just resignedly "carry on" with their ailments. Now, they have a chance to live a fuller life, thanks to a simple screen, some circuitry, and solar powered batteries.
3
The rules and limitations governing the use of AI on human beings are... ?
I don’t remember getting to vote on AI and its integration into society at all, and I’m getting very tired of seeing commercials and articles which seem to either diminish or entirely ignore the extremely dangerous potential of using machines and machine learning on human beings.
It’d just be nice if we could perhaps, at least one time, not let the private sector do literally whatever it wants with a new technology before stopping to think that maybe the private sector has successfully normalized it to Americans without bothering to introduce or discuss potential harms. How do you teach a machine to learn? Use a machine on a human being like there’s anything in the law that could possibly justify it.
2
Almost every Sunday the NYTIMES Magazine publishes a column about doctors failing many times to find the right diagnosis for a patient who eventually finds a doctor who is correct. I have long speculated that if AI can win at chess against every grandmaster in the world then they can be programmed to make a correct diagnosis for human illnesses. That day is fast coming. Doctors will use this as a tool, not a replacement for people. And it will be a very valuable tol for many people who suffer while trying to find the correct diagnosis for their illness.
30
Wow this headline might freak out the PAs and future PAs out there, another job lost to the machines
8
Why not get rid of the doctor and see a robot with A.I.
6
60 minutes already did a report showing algorithms doing a better job than the best physicians in the country in making differentials using IBMs big blue. These algorithms have been around for years. It is only due to the corrupt medical system in this country that gets rich off the sick, the poor and the dying that these basic algorithms have not been employed. AI is just the next iteration. This software will never be used here because of the billions of dollars that corrupt physicians and hospital administrators will loose. They have created impregnable barriers to entry to maintain their salaries when most of them are little more than an inefficient poorly designed algorithm. Greed, stupidity and fear will keep our healthcare system the worse in the developed world. What does it say about doctors' skills and salaries when algorithms have been doing a better job than them for the last 10 years, and the best coders are still working on an AI that can drive a truck?
15
Please realize that the term Physician Assistant applies to a specific group of health professionals who have been formally recognized since the 1960’s. Ignorance of this history does not give me much confidence in the writer’s predictions of the future! I’m also disappointed in the knowledge base of the science editor who should have caught this.
21
This physician assistant thanks you!
1
Doctors begin having to train their replacements.
https://www.businessinsider.com/signs-youre-about-to-be-fired-2015-9
3
Wonderful! Another opportunity to replace humans--and humanity--with computers. Factory workers...farm laborers...lawyers...restaurant workers...truckers...PA's. I can't wait to just sit back and let computers take care of everything for me. Sounds like "It's a Wonderful Life" for the 21st century. All I'll do all day is sit at home and commune with my fellow, similarly disposed humans. On Facebook, I guess.
Not.
Count me a one who will take my chances with my fellow man, not some coder writing algorithms next door or around the world. Do we really want computers to completely drive our lives? Maybe I should ask Alexa or Google.
11
As a HUMAN physician assistant, I am deeply dismayed and offended at this headline. Physician assistants are intensely trained healthcare providers who can do nearly any task or procedure that a physician can do. The PA profession grows by leaps and bounds each year and is helping to fill desperate healthcare shortages across the country. PAs are generally in collaborative relationships with physicians, not supervisory. While the “assistant” moniker is inaccurate, it is the accepted terminology. Using the term “physician assistant” to refer to the AI in this article is unfortunate, and I encourage you to find another term.
27
@Hunter Stidham
I don’t think healthcare corporate gonna replace you, in fact in long term doctors may not needed in current numbers. One doctor with an AI system can work with many physician assistance compare to today 3to 4 assistance per doctor limit.
But only time will tell.
2
So we can't use private patient data to request that big corporations restrict the use of toxic chemicals that are killing us (thank you Scott Pruitt) but we should share private patient data with big corporations who want to monetize fake 'intelligence' technologies to tell us how, among other things, toxic chemicals are killing us. If the cognitive dissonance of this administration is making your head spin, this should make it explode!
3
But medical doctors are given five to ten minutes max to take history and do physicals on a patient. Physicians in the US are no longer in control but rather their corporate supervisors who are not subject to the same rules of ethics the doctors are
30
AI could be used to get better diagnoses and treatment plans for women and African Americans who are currently under treated for heart disease. Many women don’t get a correct diagnosis because their symptoms are different than those of men. Sometimes they are patronized and told they are just being neurotic when they attempt to find out the cause of their symptoms. If systems powered by AI can give accurate, unbiased expert diagnoses, healthcare would be improved for millions of people and lead to better outcomes. The benefits go far beyond cost savings and delivering medical care in isolated areas. Let’s keep working on this exciting new area.
12
“But even experts have difficulty understanding why such networks make particular decisions and how they teach themselves.”
There is no mystery - what these experts need to learn is a little bit of:
Linear algebra
Calculus
Probability
Statistics
In that order. Once they do, they’ll easily see how the neural network ‘learns’ from the medical data set.
11
@Luis
If it were merely a matter of those factors, we would be able to simply program a neural net, rather than training them. Always cracks me up when people think a little knowledge means they know more than experts.
5
The math mentioned does not explain how any particular decision is made, only how AI makes decisions in general.
5
@Luis, do you really think that experts in AI aren't familiar with these fields?
2
This should make a very big difference in the diagnosis of rare diseases. We have to rely on the knowledge and experience of a few doctors for our diagnosis but with a system like this that can be updated frequently as new treatments and medical discoveries become available throughout the world a million possible solutions can be narrowed down in seconds or minutes. I think this is the first step in a giant improvement in medical diagnosis.
11
@dc miller
Nope. Rare diseases are misdiagnosed because the doctors seeing the patients fail to ask the right questions or examine the body part, or order the right test BECAUSE they are unfamiliar with the rare disease. The doc at the mecca institution, who gets to see the rare cases because the common stuff has been weeded out, looks brilliant, but is also looking at “common” problems for their practice. The AI described here will get trash info at the local CVS urgent care clinic, and excellent info at the Center for Unusual Diseases of your local Prestigious Medical Center.
We believe that Dr. House is out there, waiting to see our kids for their colds, but also able to tell right away that the fluid leaking out the nose is actually a skull fracture, but that’s not human nature. You hear hoofbeats, you think TV Western movie, not zebras.
2
So despite the headlines and figure captions, AI does about average -- better than the least effective doctors and not quite as good as the best ones. With still no idea of the systematic biases in the evaluation of either. It is indeed good that medicine is a slow-moving discipline.
11
I love the idea of this, especially for radiology and pathology. These are 2 areas ripe for further development of these technologies. Path and radiology could be nearly entirely computer driven relatively soon.
Office visits and diagnostic schemes will be somewhat harder, but also could be excellent over time. Of course, much of office work has to do with the relationship and advice, and not just data mining and diagnosing. So AI will be somewhat limited. Would be great for ER, though, where there are limited relationship factors, and more one-time simple diagnostic and therapeutic templates being utilized.
Surgery is more difficult. That’s where you get complex: not just the procedure, but the judgment, counseling about options, weighing the pros and cons of options specifically for that patient personally. That is not likely to be mechanized any time soon.
8
A.I. shows promise as a physician assisting data interpretation that I can agree on. A.I.is as good as human programs it with massive digital data and allows the computer to analyze it in nano seconds. The physician will still make the final clinical decision.
The last time I check the practice of medicine is fifty percent science and fifty percent art. The treatment of patient is not like a commodity, but as a human being with caring, compassion and empathy. This is what the fifty percent art comes into play.
I hope A.I. will be useful to the health care profession as a tool to minimize medical error and not to make them into a robot.
12
Great idea, whose time has been here for a while. It's like a second opinion from the first moment a person sees a doctor. Let's keep working on this!
2
These systems will come in and will completely displace a lot of what doctors do. Health care has become so expensive, yet so desired by anyone who can access it, for the best technology to be ignored. It is what people in wealthy societies want to spend their money on, through taxes, insurance, and/or out-of-pocket. As others have suggested, the people delivering "hands-on" elements of medicine, ranging from most of what nurses do, to the parts of surgery and procedures that are not easy to do entirely by robots, will still be needed. Re-alignments of money and power will happen, and the fights won't be pretty.
2
It's not terribly important that these breakthroughs in artificial intelligence will happen. They've been available for many years, and only computer speed and network connectivity has held them back.
What's more important is who will teach these systems to think, and more importantly what is right and wrong.
4
Human guinea pigs who lack informed consent or any legal recourse. Property of the state as a function of being born into it.
The article concentrates on one problem -- building a system that avoids false negatives. But there is another one -- gaming the diagnostic system to generate false positives in order to increase reimbursements to the providers. And then gaming the followup treatment system to generate enough ambiguity in the success of the treatment that the patient has to continue with it. And anyway, Medicare is paying, or so the provider will say, so why worry ?
2
@JM
Hmmmm.... curious point about false positives. Patients are as guilty as providers of gaming things (histories), with complicated incentives, many times leading to over testing and over treatment. I'm curious how AI will handle the narcissists, the borderlines, but most especially the neurotics and histrionics. It makes up 30% of my day but 100% of my stress. Playing Peter and the wolf all day can be exhausting.
8
A neural network is a vast array of computers and databases that are all interlinked. High speed computers can analyze untold numbers of combinations of events and draw conclusions and make predictions. This is the future. It will drive your car and as illustrated here, become your doctor. I welcome it.
With the minimal amount of time physicians spend with people, the reliance of protocol medicine to order tests, the neural network may become the first best option. It also has the potential to greatly reduce costs. The doctor's is now a terminal that feeds data into the network. The computing power will not be resident in the office. It will be like cloud computing.
We now have the drug store chains getting into clinical practice. They are buying insurance companies. They have walk in clinics and dispense drugs. I envision neighborhood drug stores becoming the front line of healthcare. You walk in, are taken care of by a medical technician who interfaces you with the network. The network then orders any necessary tests and makes an evaluation. If a specialist is needed, you get referred. Otherwise, the terminal spits out a prescription with any lifestyle instructions and off you go.
This opens the door to foreign competition. If a human needs to evaluate the results, that human could be located anywhere in the world. Hands on will only be required for procedures and surgeries. Amazon will treat your sore tummy. All for a monthly membership fee.
8
@Bruce Rozenblit
Bruce, it's clear that you don't know much about neural networks. You are confusing a physical network such as the Internet with the mathematical, conceptual network of an multilayer array of simulated neurons. For many years, it was thought that these simulated neurons were of limited power. Researchers, primarily in Canada, discovered fairly recently that by making the "neural nets" very large and by layering them, they could be trained to make very complex discriminations. It is these "deep networks" that are now revolutionizing artificial intelligence.
2
This article while interesting from the standpoint of futuristic simplicity is flat out misleading and misguided. The use of Physician Assistant in the article's title suggests that this writer has not the first clue about healthcare delivery systems in the US and therefore belies most of the following information that follows in the body of the article. Physician Assistants are an integral part of the healthcare team and not relegated to assisting the physician. PAs provide way more than a diagnosis. Their skill and compassion care are a central part of the patients healing and wellness. It would be far more advantageous if AI could be used to figure out ways to help patients by reduce barriers to good health such improve exercise rates, reduce obesity, and uncover methods for improving mental health and outlook rather than sussing out a diagnosis from a collection of symptoms. That would be real progress and something that would assist this this PA with her practice.
22
If ai ever becomes capable of that, we better get real good at diagnosing wounds caused by plasma rifles in the 40 watt range
Silver lining for the humans: Recent research has shown that radiologists and New AI work synergistically, supporting a long-term added value of human cognition even in predominantly non-patient facing specialties.
The advancing power of neural networks and big data are undeniable. As with everything in life, they have pluses and minuses. From a privacy standpoint, the originating data points should be both qualified and ananymous. But incorporating human observations and comments as part of the data set can only be to our benefit as patients...
1
Diagnosis is working through a large decision tree. Given a set of inputs from tests, imaging, physician exams, it's going to excell over what a human Doctor can do.
It can't touch a patient. So a human Doctor can have inputs a computer can't have. Given enough data it can ask for more data, but it's slow at "improvising".
Medicine is based on research, statistics, patient populations so computers start almost on "level ground" with Doctors. A Doctor reads the same journals, has the same data, and typically does that at night before a late bedtime, after a long day at work.
The project at Google that I am familiar with is laser scanning of retinas for pathology. They work with the top doctors at Stanford and UCSF who provide laser scans from their patients and review the results of the Google AI diagnosis.
The doctor I know works on this project in the hopes that the laser equipment can be made cheap enough to operate anywhere, and the results can be analyzed for a 100 million people. Not just the Tech engineers in Silicon Valley that can afford state of the art medical treatment.
9
Sounds like a technology, data, and expertise transfer program disguised as human services research. Cherry-picking good results for certain diagnostically "trivial" or simple diagnoses will not show how an integrated medical system can function in an ambiguous or fuzzy world--do we really want robot doctors? The Chinese might I suppose....for their low social credit scorers?
2
If the algorithm diagnoses you with something, who or what prescribed the medicine or treatment or operation? Who or what is sued if there is a mistake?
4
It is isn't a question of if, but simply when.
Humans can no longer compete with AI in complex games such as chess and GO. Medicine is certainly a complex game but diagnosis of illness is certainly something susceptible to machine learning.
What is far more susceptible to machine learning and AI is the interpretation of genetics and the role it plays in disease and development. This area is likely to be all AI fairly soon. Diagnosis in the future may be primarily made electronically without much patient interaction at all.
What may be lost is the healing touch which has served physicians (no matter what they were called) for millennia. Even when a doctor's bag contained just a few herbs and perhaps some shamanistic paraphernalia, being a healer was a noble calling.
We advance, an in advancing, we lose touch with why we are here.
20
Medicine is not a game like chess. Medicine is a discipline to try to help other human beings heal. There are a lot of things that should not be regarded as games. And that probably should not be industry, either.
2
"Physicians try to be systematic when identifying illness and disease, but bias creeps in . Alternatives are overlooked".
The statement in the article shows the possibility for the persisting uncertainty in diagnosis could impact the output of the diagnosis by AI. I would like to bring the two examples
- One is the process sequence and the other is the medical diagnosis of a cancer patient, I came across in the past.
1) Equipment,materials and ingredients for the product are same but in the process engineering of the technical know how offered by two different nations differed. The ingredient 'A" that is fed first in the process technology of one nation has become 'B",the second ingredient in the technology offered by the other nation.Both nations have their own justification of the sequence of the ingredients addition into the process that determines the quality of the final product.
2) A team of physicians had a debate on the type of treatment given to a cancer patient who had been diagnosed for 'tumor on femur head". Whether to grind the tip of the femur head or to treat it with chemotherapy was the subject caused the difference of opinion among the doctors.
My suggestion is to use the output of the AI as a tool to aid the diagnosis and not as the inference to make the final decision to go ahead.
We shouldn't mortgage the 'human intelligence' into the data bank of 'Artificial Intelligence" in the name of scientific revolution.
2
We tend to forget that much of this rests on information that is entered into records. Just because information is in a system doesn't mean that it is good information - garbage in, garbage out. Anyone who has looked at the abomination that is EMR in this country will recognize that what is in records is often garbage (and, as one poster has mentioned, because in our current factory medicine and procedures-based system, quality time interacting with patients, getting a good H+P, and having the time to properly enter it into a record is not given priority. Often the EMR even offers text to populate fields => AI scouring artificially generated text to diagnose people...).
Not that this is unpromising, people just need to take it with a big grain of salt, and a lot of caution.
22
@pdrothstein
The accuracy of computer records would also include the manner in which they are stored. This usually implies that there is a human being who can 'raise their right hand' in a Court proceeding to swear about the accuracy of a record. This can be very difficult to do in the digital world. In the U.S. we assume that a medical record was produced and stored under the HIPPA regulations, but a lot of data pertaining to a medical diagnosis will come from unregulated sources. To illustrate the point, consider the types of questions that a doctor might ask about a medical problem. Most of your answers will be from unregulated data.
An analogous problem comes in the Law Courts. I've recently noted that the U.S. Supreme Court, when there is a discrepancy between electronic and written case law, the bound (paper) copy will take precedence over the electronic one. I assume that this is because the provenance of the written data can be established, whereas the electronic records might not be. According to the U.S. Supreme Court: " Only the printed bound volumes of the United States Reports contain the final, official opinions of the Supreme Court of the United States. In case of discrepancies between a bound volume and the materials included here--or any other version of the same materials, whether print or electronic, official or unofficial--the printed bound volume controls."
Cite:
www.supremecourt.gov/opinions/boundvolumes.aspx
2
@W
"...when there is a discrepancy between electronic and written case law, the bound (paper) copy will take precedence over the electronic one".
That is so interesting!!
It speaks to a recent conversation had by a colleague and myself. We were talking about the variety of electronic calendars and devices available to us to help us stay on top of all the things we need to do and deadlines we need to meet at work.
But as we talked, each of us pulled out our bound (paper) day planners to share and compare, and both of us came to the end of the conversation with a laugh saying, "Nothing like paper!!".
4
@W
That is great news. Thanks for sharing opinion synopsis and cite.
More and more legal material is available only online, and courts have gone to requiring electronic filing. I understand the reduction in paper, but I still keep paper records of what is important to me as I find them easier to work with, especially when I have to deal with numerous records. I also vastly prefer reading cases in hard copy.
1
This year ummm errrrr ???
INTERNIST-I mid 1970s
CADUCEUS late 1970s
MYCIN 1970
Followed by EMYCIN and TMYCIN
The fact is that application based on symbolic reasoning have been providing medical diagnostics since the 1960s and unlike Machine Learning and Deep Learning application they have always been able to provide an audit of their reasoning.
1
Do you know that Physician Assistant is an actual designation? Not just as assistant to a Physician right? They write prescriptions and first assist in operations, run remote clinics. Did you mean to target P.A.'s. ?
42
We need better privacy laws - ones that work for the real world.
I don't need my past records kept private from my new doctor, nor from my pharmacist. Aggregate health data that does not have my name on it should be freely available for research. And - if I choose a weak stupid password for my medical account, it should be my choice!
HIPAA is often overly aggressive, and some of it seems oriented around the few people who have extraordinary levels of attention - celebrities and politicians and such. For most of us, it is only a hassle that we have to keep granting rights to every doctor we consult, that medical records are such a pain to pass around as we choose.
4
Good! We are on our way to physician replacement. That will drop the price of medicine way down.
2
@Keith Dow
Physician reimbursement is 10% of the spend. The AI will cost at least that much.
1
@Keith Dow - Not if insurance companies are involved.
2
This fine accomplishment will benefit all of us, and soon.
Like 55% of all peer-reviewed STEM papers we publish, it is a collaboration between Chinese and US scientists.
More than half the scientific papers published by American institutions have at least one Chinese co-author.
6
As a medical professional, there is no way around this. We are already trained to use myriad algorithms to diagnose and treat illnesses. We use robotics in surgery. But the more you remove the "touch" and "feel" and art of medicine - of knowing and reading your patients - the more it will continue to be reduced to a set of symptoms, a diagnosis, and the correct treatment. Presto! Medi-mat!
And why not? We already rely on myriad medical apps and on-line subscriptions to the professional medical literature to check & double-check our litigious livelihood with its massive & ever-expanding base of knowledge - but it's also going to put a lot of professionals out of jobs. Nothing new there, and it's no shock that trump signed an executive order to develop and regulate AI. After all - he loves comic-book science (Space Force!!).
Hopefully there will still be a role for providers like me to provide oversight; oversight informed by the nuance and through the lens of body & soul. It might even allow those of us who like to practice medicine in that way to return to the kind of practice for which we hunger most.
5
@Kitt Richards
May I commend you for your deeply thoughtful remarks.
You have something to say, and you know how to say it.
1
I would be curious about how the algorithms perform on high stakes diagnoses? The ones where misdiagnosis can lead to death in the near term.
3
What was the comparison data by which the percentages were found? We have the doctor who also put in the diagnosis and they compared this to what other diagnosis.
Surely you could have come up with better terminology - a Physician Assistant is a mid-level medical provider. I know this because I have been in this profession for nearly 35 years. I graduated from a medical based school of allied health and spent an additional year in a post-graduate program for surgery. I am still in a clinical practice at a VA hospital, where I have a great deal of autonomy in treating patients.
Please be more careful how you word your articles as it can cause distortion for those who are not as informed.
54
@Debra Mandelblatt
Amen, sister
Signed,
Another PA
12
I, too, am a PA. Most of the professionals I know don’t use the term mid-level provider anymore. It’s demeaning. We use the term advanced practice provider or advanced practice clinician.
10
@Hunter Stidham. Wow—the PC world has reached medical titles. Mid-level is exactly accurate in describing PAs and NPs. It’s more training that a nurse, but less than a physician. It is literally mid-level. It’s a pragmatic description. There is nothing demeaning about it at all.
16
We need fewer regulations and less privacy hysteria.
Doctors have always protected their patients' privacy. HIPAA just adds paperwork and makes it difficult to do research. Repeal it.
4
I disagree. I have worked at a county’s medical records storage facility. I was hired off of craigslist and not subjected to a background check. Hard copy records dating back nearly 30 years filled a warehouse. While most records were in folders, there were individual X-rays scattered all over the floor and piled up in trash cans inside the facility. It was broken into once during my time there. Just because privacy law means something and is respected by most, you really don’t have any guarantee of it. It depends on others respecting it to meaningfully exist.
1
"Physician Assistant" is a mid-level certified professional role in the U.S. It is misleading to use that term for clinical decision support systems which have been developed and deployed in practice for years, long before AI became the technology du jour.
Further, while today's AI excels at image-recognition and the like where trained, existing decision support systems always mix some explicit medical rules with some interpretation of semi-structured data like case reports; the only truly AI part is the data interpretation.
Researchers always want freer access to data; the people who have it and have been using it for decades, like the VA and Kaiser Health, have made good improvements, but it turns out the real advances in health care are coming from simple things like checklists and taking time to listen to patients.
For my money, instead of putting more data in the hands of researchers (who use it for insurance companies to decide what to cover), we should be putting more explanations in the hearts of patients, so they can more effectively advocate for their own health.
58
@Wes
I didn't even notice "Physician Assistant" in the headline - I was busy reading the article!
Thank you for bringing out and underscoring that very important point.
2
Mid-level is an outdated, demeaning term. Preferred terms: advanced practice clinician or advanced practice provider.
5
@Hunter Stidham I prefer to be called Thor. I am frequently disappointed.
6
American Big Medicine salivates at how they can profit from this, while Pelosi is all ears to learn from them and associated donors what to say to Americans so that they don't start getting their hopes up about there ever being some low cost, actually available version.
She knows already an affordable version would be socialism, and she is devoted to protecting us from the moral hazard of expecting good health care without a huge struggle ie destroying our entrepreneurial drive. But she will still need to coordinate the talking points, cross the dollar signs and dot the i's.
3
Interesting article - although my comment is in regards to the title. Physician Assistant, or PA, is a profession in the medical community. PAs are licensed healthcare providers who can perform a history and physical, diagnose an illness, and create a treatment plan for a patient. They work in all different medical fields and would therefore be impacted by advances in A.I. - but hopefully not the sole target as is lead to imply.
29
You mean someday a middle-aged or older woman can have her symptoms addressed seriously, without a doctor assuming she’s just unhappy/menopausal/a hypochondriac/prone to symptom shop on Google...or just nuts? That would be nice. Give me an appointment with a computer!
There is a more basic failing in medicine here in the US, though, and it may be at least partially due to our stringent HIPAA requirements. A relative of mine died at a too-young age because she was misdiagnosed, and that mistake was made worse because she saw different doctors (not in the same medical group) for different problems, and those doctors did not consult with each other. So she was diagnosed with “asthma” that was actually fluid in her lungs caused by coronary heart disease, and another doctor diagnosed her with edema unrelated to an heart problem. She was on conflicting medications. She died horribly. If her medical record had been on a central database that these various caregivers could access, she might not have died. You can’t rely completely on a patient to provide all medical records or an oral history, especially if they aren’t feeling well.
99
@Passion for Peaches I'm excited about getting an appointment with a computer too.
As a black woman I've been misdiadnosed more times than I care to remember, including misdiagnosis for preeclampsia and endometriosis.
I've also had friends pass away or get horrible medical care from bad and biased diagnoses, so I hear your pain. I'm so sorry for your loss. It's sad, because I wish the system and the people who work on it could evolve, but I feel getting AI into hospitals will be a boon for women of all backgrounds.
34
@Passion for Peaches Patients usually don't posses a copy of their records, and cannot provide them to a new doctor.
9
@Passion for Peaches
I'm a PA (not a robot, as the headline implies, but a human, licensed medical professional) - but I'm also a middle-aged woman! And as both a provider & a patient I agree with you, Peaches - I wouldn't mind there being an objective data-collector and processor in medical encounters, overseen by wise & compassionate human oversight.
I went into medicine bc I wanted to care for patients and to make a difference in their lives, and to engage my mind with the marvels & the mysteries of human physiology. I have been so discouraged in my attempts to practice medicine by the Greed Machine (hospital profits and insurance companies) that now calls all the shots.
An awful lot can be gained by sitting in quiet conversation with a patient, making full eye contact, sensing their emotions, and observing body language. These things have been forced out of patient encounters as providers like me are lashed to see more and more patients, all for the glorification of the bottom line.
51
First humans made machines to do their physical work for them; now we're making machines to do our thinking for us. I, for one, don't want to know what's next.
2
@RR Medicine is incredibly complex. Every doctor has something like EPocrates on their phone, to tell them about each drug and its interactions with other drugs. Every doctor checks Uptodate for the latest research and guidance on a particular condition. Without these systems, doctors could not keep up with the evidence-based standard of care, and people would die as a result.
I for one want them to have all the support they need!
17
@RR, all I want is a machine that can clean my house thoroughly.
8
We don't need a tech solution. We need to do better histories and physicals, and payment should be for performing good H and Ps and not for doing procedures.
AI will make a few prevailed people in Silicon Valley rich, and will not help;p patients or doctors. Looks at the catastrophic failure of Watson.
One of the less insightful assumptions is that everything has a tech fix. Not everything c an be fixed by tech. Somethings it make much worse.
Any clinician who has to use the abomination called the electronic health record knows that.
16
@JB
Did you note that the machines did BETTER than the doctors at diagnosing the patients?
And this was doctors at their best, in a competition, not doctors in their offices, with distractions and interruptions.
Not everything has a tech fix, but it's long been shown that an AI can do better than a doctor at diagnostics. It's well worth adding it to the set of tools doctors have.
37
Unfortunately many people can’t see past their prejudices, like “mid aged woman = menopausal/has hysteria”, “young woman has stress (hysteria)”. Of course, they call it “somatization” or “anxiety” now.
2
@susan
Actually, the machine seems to have done about average in the ranges that the report cites for human docs.
The question one has to ask is who/what will gather the information the AI machine uses to base its diagnoses. You know the old saying, garbage in, garbage out. So far, it’s been humans who input the data, ask the questions and examine the patients. Until we get those little scanners they used on Star Trek, diagnosis will depend on the examination skills of humans.
One fact that the article ignores, is that Chinese medicine has relatively few midlevel practitioners (and many, many doctors) compared to the US (see comments from American PAs), so that most of the data this study used most likely came from trained pediatricians, who given the sheer number of patients they brag about in Chinese pediatric clinics (or data, as referred to in the article) may be too overwhelmed with work to make enough good decisions to beat the machine. Which prejudices one to pick/use the machine to check on the docs’ accuracy. But don’t expect a doctor serving as a data entry clerk to stay happy and recruit more smart motivated young people into the field.
4
It could actually replace doctors in many areas, when say the flu is ranging you really don't need a doctor's appointment to know what drugs you need.
3
@vulcanalex, Kaiser allows online consultations for things like that. It keeps all those contagious people out of the Urgent Care waiting room.
1
@vulcanalex
True, but I wouldn't want to rely on it if I had undiagnosed heart failure and the treatment prescribed for the flu killed me as a result.
It's an important tool (and we are already utilizing lots of like technology to practice medicine).
3