Announcement

Collapse
No announcement yet.

Should doctors be afraid?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Tex, you would be right:

    www.enlitic.com

    Comment


    • #32




      Whenever I’ve mentioned on this board that a certain diagnostic specialty that does not require social perceptiveness, negotiation, or persuasion to process digital information to arrive at an answer is an appropriate target for automation (though not imminently at risk for many reasons), it hasn’t been well received.

       

      When this came out a few days ago I almost dug up those old threads for this quote just to mess with people  ? :
       “I think that if you work as a radiologist you are like Wile E. Coyote in the cartoon,” Hinton told me. “You’re already over the edge of the cliff, but you haven’t yet looked down. There’s no ground underneath… It’s just completely obvious that in five years deep learning is going to do better than radiologists,” he went on. “It might be ten years. I said this at a hospital. It did not go down too well.” Hinton’s actual words, in that hospital talk, were blunt: “They should stop training radiologists now.”

       

       
      Click to expand...


      Radiologists won't get replaced that easily. Computer vision is a tougher nut to crack then algorithmic patient moving. I can make an e-bet: ER physicians/"hey you got BP? try losartan" person get replaced before Radiologist.

      This whole argument about oh hey we add value etc etc - yes we do, but then why are NPs infiltrating clinical medicine? And since they can do certain tasks, why can't computer AI do their tasks?

      It all starts like this and reality sets in. Sorry but basic primary care level medicine is not super complicated. Radiology looks easy but at any given time, you appendix is anterior, lateral, retrocecal or deep in pelvis. Current computer vision and AI algorithms can't solve that, but they sure can understand natrual language (Watson et al) like a human. Permutations of which test to order is an easy next step. Only reason this hasn't happened is because medicine is slow to adapt. Just like Hospital admins...its coming.

       

      Comment


      • #33
        Again, I think the job will change, but we'll still need radiologists.  For example, I talk to the radiologist several times per shift.  Some of these conversations could be replaced by a text message from a robot--"Hi, the CT for Mr X shows a parietal mass, likely malignant, benign process not ruled out, infectious process unlikely but recommend clinical correlation.  MRI with and without may help delineate."  (Just kidding.)  But that weird belly case with the odd CT findings will require more than HAL to figure out, and Watson might be challenged to discuss the risk/benefit of thrombectomy in the 70 year old with the particular clot burden/lactic level/timing/etc.  Don't you think?  On the other hand, when I was playing with my 1200 baud modem I never could have imagined watching a movie on a phone that could fit in my pocket!

        Comment


        • #34




          Again, I think the job will change, but we’ll still need radiologists.  For example, I talk to the radiologist several times per shift.  Some of these conversations could be replaced by a text message from a robot–“Hi, the CT for Mr X shows a parietal mass, likely malignant, benign process not ruled out, infectious process unlikely but recommend clinical correlation.  MRI with and without may help delineate.”  (Just kidding.)  But that weird belly case with the odd CT findings will require more than HAL to figure out, and Watson might be challenged to discuss the risk/benefit of thrombectomy in the 70 year old with the particular clot burden/lactic level/timing/etc.  Don’t you think?  On the other hand, when I was playing with my 1200 baud modem I never could have imagined watching a movie on a phone that could fit in my pocket!
          Click to expand...


          Not really. Thats exactly where a Watson is already, today, light years ahead of humans. They dont have preferential recall or biases (outside of programmed ones) like we do. They can remember/compare everything. The more unique the case the more likely they pick up on something you wouldnt have and would tailor the treatment far more appropriately. They dont forget. Not saying they'll be able to read cts/etc...since thats above my pay grade, but thinking the tough diagnostic case is where you shine and they dont is exactly wrong. There was some interesting stuff out last year about that, its already there. Its coming folks.

          Extrapolating todays limitations and thresholds into tomorrow has been a persistent, albeit persistently wrong, bias. Have to think outside that. It wont happen all at once, not the least of which reasons will be risk management and figuring out who the heck is liable for mistakes. That will be the determining time frame issue. Everyone wants a piece of the money, no one wants the liability.

          Comment


          • #35
            wow- big computing must have censored this... Will post again later (I'm a radiologist and my post disappeared as I clicked on it - now I'm worried)

            Comment


            • #36
              Zaphod is correct. All I'm saying is that natural language processing is an easier problem to solve than complex computer vision problem. The former is what algorithmic medicine is, and will be first to go.

              Comment


              • #37
                All true with logic..as is Communism is superior to Capitalism in the textbook but we forget this one small thing---the human factor.   Something as simple as cholesterol medication and statins.  It's years of EBM boiled down into a simple single calculator and simple clear advise; yet the getting our patients to take the medication consistently and goes way beyond the simple >7.5% risk = take a statin = 25+% risk reduction.   A computer can recommend, graph, and nurse can spit all the recommendations out...yet more often than not, the patient waits until they see me to start the medication despite the clear recommendations---the human factor.

                This is the reason why physicians will have a place in medicine regardless what the future landscape holds.  We'll certainly be delivering it differently, but it won't be in the hands of a computer for the ultimate decision.  For you trekkers out there -- Booby Trap.  TNG.  

                Comment


                • #38
                  Natural language processing depends on the computer being able to understand the language.  "I hit my head on the refrigerator and now it hurts" is a slam dunk for a computer already.  What is extraordinarily difficult to write NLP for is the idiosyncrasies of human language.  A computer can't tell that the patient is drunk, or from a foreign country, or is starting the story where THEY think it matters, or is distracted, or the patient misunderstands the question. Heck, I had a patient who came for a lung mass spend 10 minutes talking about their foot pain first because that's what made them decide to see a doctor in the first place.  Show me a computer that doesn't order a foot X-ray and I'll concede defeat, but I just don't see it.  Fail to ask the right questions and and it's GIGO.  A computer can potentially be helpful in decision support, but even then is fundamentally limited to the "right" answer.  How can a computer calm irrational fears, or willingly accept the second-best recommendation that is the best one the patient will comply with, or discuss limitations on care?  How will computers decide the best course of action in the face of conflicting data?  I don't lose any sleep at night that I will be replaced by a computer in my career lifetime.  If I do, at least I found WCI early enough in my career that I'll be able to afford to retire ?

                   

                  Comment


                  • #39




                    Natural language processing depends on the computer being able to understand the language.  “I hit my head on the refrigerator and now it hurts” is a slam dunk for a computer already.  What is extraordinarily difficult to write NLP for is the idiosyncrasies of human language.  A computer can’t tell that the patient is drunk, or from a foreign country, or is starting the story where THEY think it matters, or is distracted, or the patient misunderstands the question. Heck, I had a patient who came for a lung mass spend 10 minutes talking about their foot pain first because that’s what made them decide to see a doctor in the first place.  Show me a computer that doesn’t order a foot X-ray and I’ll concede defeat, but I just don’t see it.  Fail to ask the right questions and and it’s GIGO.  A computer can potentially be helpful in decision support, but even then is fundamentally limited to the “right” answer.  How can a computer calm irrational fears, or willingly accept the second-best recommendation that is the best one the patient will comply with, or discuss limitations on care?  How will computers decide the best course of action in the face of conflicting data?  I don’t lose any sleep at night that I will be replaced by a computer in my career lifetime.  If I do, at least I found WCI early enough in my career that I’ll be able to afford to retire ?

                     
                    Click to expand...


                    Who said about losing sleep? Doctors wont be completely replaced - duh we need the human interaction. You gonna be served by a robot regarding your life? not very appealing. People arguing this point are saying - hey 2+2 =4. Yes. Got it. Not what I am or Zaphod saying. What this will do is severely curtail need for the number of doctors. I mean, a PA/NP can just be the human face while AI does the heavy lifting of actually thinking. Why need an expensive MD?

                    Do not discount NLP that easily. It is rapidly improving and Watson got all the proverbs, innuendos etc that Jeapardy is famous for...that was 4 years ago.

                    Also you didn't order a foot xray for a patient that came in with foot problem? Really? ok. I am willing to bet that lung cancer or no lung cancer, many physicians would order foot xray to "rule pathology out", just cuz the patient complained.

                    Comment


                    • #40
                      Yeah, I say nothing to worry about as well.  Has anyone on here ever met an "obsolete" doctor?  There's always work for doctors.  All you need is a medical degree, state license, and DEA registration.  You might not be getting paid boat loads for doing a 10 minute specialty procedure or whatever, but you'll be able to work and earn a good living.

                      Plus, I feel like the medical field actually changes too slowly for any doctor to suddenly find him/herself obsolete.  You'd have plenty of time to adapt to something new if you found your specialty at risk.  For example, most doctors still use pagers even though we're all carrying smart phones around in our pockets  I always find that hilarious.  My group just will not let our pagers go away despite the fact that we are all now required to have smart phones to run an app called doc halo that gives us hipaa compliant texting.  Our pagers have been obsolete for years but we still use them!  They're truly vintage technology now.  I'm on call right now and I have to carry a pager for cross coverage, a personal pager for my patients admitted to me, my smart phone, also for cross coverage and patients admitted to me, and a hospital cellular phone for ER admissions.  Hold habits die hard I guess.

                      Comment


                      • #41
                        I have found pagers to stand up better against the new and emerging technologies of today.  The point of carrying something is that you can be reached when needed.  Vocera is not only annoying as ************************ but it doesn't work outside the hospital.  Smart phones often have service gaps in hospitals, making them imperfect for satisfying the above criteria.  And this can be very service-provider dependent.  I get zero reception in my hospital.  Zero.  The other problem with that is privacy.  Some people don't want their phone numbers given out.  When calling a page back you can block your number if you want.  Getting a paging system in our hospital would be a godsend.  Just because newer technology exists doesn't mean they satisfy the basic communication criteria of being able to be reached (alpha-numeric being superior here because it can provide important info too).

                        Comment


                        • #42







                          Natural language processing depends on the computer being able to understand the language.  “I hit my head on the refrigerator and now it hurts” is a slam dunk for a computer already.  What is extraordinarily difficult to write NLP for is the idiosyncrasies of human language.  A computer can’t tell that the patient is drunk, or from a foreign country, or is starting the story where THEY think it matters, or is distracted, or the patient misunderstands the question. Heck, I had a patient who came for a lung mass spend 10 minutes talking about their foot pain first because that’s what made them decide to see a doctor in the first place.  Show me a computer that doesn’t order a foot X-ray and I’ll concede defeat, but I just don’t see it.  Fail to ask the right questions and and it’s GIGO.  A computer can potentially be helpful in decision support, but even then is fundamentally limited to the “right” answer.  How can a computer calm irrational fears, or willingly accept the second-best recommendation that is the best one the patient will comply with, or discuss limitations on care?  How will computers decide the best course of action in the face of conflicting data?  I don’t lose any sleep at night that I will be replaced by a computer in my career lifetime.  If I do, at least I found WCI early enough in my career that I’ll be able to afford to retire ?

                           
                          Click to expand…


                          Who said about losing sleep? Doctors wont be completely replaced – duh we need the human interaction. You gonna be served by a robot regarding your life? not very appealing. People arguing this point are saying – hey 2+2 =4. Yes. Got it. Not what I am or Zaphod saying. What this will do is severely curtail need for the number of doctors. I mean, a PA/NP can just be the human face while AI does the heavy lifting of actually thinking. Why need an expensive MD?

                          Do not discount NLP that easily. It is rapidly improving and Watson got all the proverbs, innuendos etc that Jeapardy is famous for…that was 4 years ago.

                          Also you didn’t order a foot xray for a patient that came in with foot problem? Really? ok. I am willing to bet that lung cancer or no lung cancer, many physicians would order foot xray to “rule pathology out”, just cuz the patient complained.
                          Click to expand...


                          Exactly the point is how many will be needed, and at what level. What will effectively happen is a doctors inability to scale will go away and your presence will be magnified by these augmentations of slower less efficient versions of things you use to do. There will also be a lot more midlevels in place of doctors. Remember how expensive physicians are, and all these things that are said computers cant do, a cheaper version of you can, and a simple algorithm of known AI/cpu difficult pts will be reserved for compete MD interaction. How many midlevels could be hired for the price of one doc in this instance? They are also easier to employ and fall into corporate tactics with large downward pressure on wages once a certain size.

                          I know we like to think since we've had so much education and time we're basically super amazing, but in reality being a doctor is very much dependent on an apprenticeship model and hands on rote learning by doing until you've done enough its automatic. You may have to extend a residency a year, but no reason PAs/NPs cant jump into a model like that

                          I doubt that comes with a pay raise. Doctors wont disappear, thats a bit absurd, but there wont be a pressing need for as many as there are today on a per capita basis. One doc will be able to easily do the work of 2, then 4, etc...Thats the issue. I dont think this generation has too much to worry about, but you can see it will eventually come. Primary care docs are taking on all kinds of future risk, competition risk, and high cost/low pay risk right now.

                          I would have loved to see the car production plant forums back in the day the first time someone thought robots posed a job risk to them. Pretty sure it would have been brutal and all kinds of mockery, that you need more people to fix the mistakes, etc...yet now those guys dont really exist anymore. US manufacturing continues to hit all time highs while employment in that sector hits all time lows.

                          Comment


                          • #43
                            I was going to give the EXACT example in the last response, but eh too much effort posting.

                            Hit nail on the head.

                            Comment


                            • #44
                              Being nearly twenty years out and as a general internist, have seen the landscape change in primary care quite a bit.  If people don't believe there's a shortage of quality primary care out there, give your local large group a call and see how fast one can get a new patient appointment.   Industry standards are <30 days; new value based systems are looking to push for 14 days or even 7 days.    They are accomplishing this mainly through mid-levels for access purposes and that has been a major change.

                              Though the vast majority of "need appointment NOW" are the unanticipated crashes of new cancer diagnosis or uncontrolled diabetes crashing down after years of deferred care or new stroke or uncontrolled sleep apnea --- all needing a good primary care doctor to oversee over multiple episodes of care.   That, is not for the mid-levels.  I would challenge any of us here to put our lives in watson and mid-levels hands in any one of those cases.

                              We will certainly change the way we deal with low-hanging fruit visits -- common cold, bronchitis, medication titrations, refillls, and healthy wellness visits all come to mind for mid-levels and watson to assist upon.   CDS/Watson will help efficiency at the high level too, but wouldn't supplant.

                              For the production model-- the highly skilled worker still remains.  The work is different, but still highly sought after.  GE CEO piece ran on CNN over the weekend to this very issue of delivery.  That is what we'll see in our service sector in years to come.....but afraid?  No.  Adapt.

                              Comment


                              • #45
                                As a radiologist I'm not too worried about AI taking my job anytime soon. I will be long retired. Look at how long we have had CAD (computer aided diagnosis) for mammography, and how mediocre it is. As AI develops and the computers become better I think they will be used to make us more efficient. They will 'read' the study and flag areas of concern, perhaps even trying to make a a diagnosis. Maybe they can clear areas that are normal so they can be safely ignored. But I think there will still be a human looking at the images. There are much easier jobs to replace by AI than physician. We will not be the canary in the coal mine.

                                Comment

                                Working...
                                X