Thank you, Dr. Topol, for yet another useful, information-filled article relating to AI. I was particularly struck by this:
“With the brief duration of a clinic visit, it is not surprising that there is little time to reflect, because it relies on System 1 thinking, which is automatic, near-instantaneous, reflexive, and intuitive. If physicians had more time to think, to do a search or review the literature, and analyze all of the patient’s data (System 2 thinking), it is possible that diagnostic errors could be reduced.”
Any way that AI can help with this will be to the good, as, from what I can see, the problems physicians face are not only not going away, but it seems are getting worse. Indeed, you link within this paragraph to Danielle’s 2019 NEJM article, where even the first paragraph speaks volumes on the issue: “In modern outpatient medicine, there’s no time to think. We race to cover the bare minimum, sprinting in subsistence-level intellectual mode because that’s all that’s sustainable. Such practice is a petri dish for medical error, patient harm, and physician burnout.”
I’m looking forward to having another tool to help with our practice, every Batman needs a Robin, I’m not sure which we would be in the long run but I do have some skepticism about the primary care study. I think the scenarios were pretty contrived and not like an actual primary care visit which is usually sprawling. Perhaps 15% of what we do is making new diagnoses. There is so much human level coordination of care, balancing conflicting treatments with multiple medical problems, tending to patient preferences about their medical care, patient education etc. etc. A lot of these headlines are sort of all or nothing dualities that lead to people thinking AI vs Doctor. And I certainly don’t want to be lead through diagnosis, treatment, and whatever comes next with cancer by a chat bot. Would love to see more studies where doctors are working with AI, as those results are probably the absolute best for now.
Not sure if you practice clinical medicine, but I’m trying to decide whether to use and train Dragon ambient AI for help with documentation of visits. I’m conflicted. The process of writing (dictating with Dragon) a SOAP note after each visit really helps me synthesize and organize my thoughts and plans for the patient, and then provides a narrative I can pick up with at their next visit. I think the clinician brain will atrophy in some respects by delegating too much of this to AI.
Nonetheless efficiency always wins, and so I guess I should try it out. This is the most concrete way AI is inserting itself into outpatient medicine that I can feel - LLM documentation.
I wondered what you might think about this, too, given our past exchanges. It’s very hard to assess this from my perspective as an “end-user,” as I only see what I (and friends) experience. From that vantage point, I am personally aware, in recent times, of three fine, empathic MDs--a PCP, an oncologist, and an endocrinologist--who appear to have burned out and retired early or otherwise just stopped practicing medicine. I also know Danielle Ofri, so when she speaks to this problem (as in the article Dr. Topol links), I sit up and take notice, too. With that preface, what I hope for with AI, as Dr. Topol describes it, is that it can be an aid to free up MD time to focus more fully on the patient in front of her/him. Related, did you happen to see the presentation by Bob Wachter related to this? I thought it spoke to the issues usefully, but would appreciate knowing your view, as always: https://youtu.be/2wPZ4kf1kLE?feature=shared
You are so invested in both sides of the patient doctor relationship, and I am so appreciative that people like you care about how it’s going on this end!
One of the best articles I’ve read in this burnout stuff in terms of family doctors was this from prepandemic times which have only become worse:
I’ll try to respond more specifically about AI but I think in general the potential benefits are a trade off, and dehumanization needs to be at the forefront of all AI considerations !
No need to respond, but I just wanted you to know I read, with extreme interest, the series of articles you noted. These are life lessons all of us can use--I particularly loved the work-home separation example of Mr. Rogers😎. So, from the patient POV, here are some individual level takeaways 1) do what I can as a patient to make my time with my PCP most efficient and effective for both of us; 2) be judicious and concise in use of the patient portal; 3) make sure and thank my providers/staff out loud, don’t just think it! (These are all things I do now, but could do better.)
The harder nut to crack is the systemic end of things, and harder still as one patient with such limited vision into the system. This article helps there, too, as I can frame requests and comments with the provider POV in mind. Beyond that, I do want to find a way to speak up constructively when I spot a possible “system” problem, that is, one that is beyond the control of the provider or individual staff to address. Where and how to do that effectively, including suggesting solutions, remains an open question.
So, anyway, thank you so much for engaging. Your time is precious, too, and I have likely use up more than my share here!
Thanks so much for responding, and for the link! As for my investment, it’s surely within the interest of us as patients to help our health care providers succeed, if we can! As I and so many friends are getting older, this becomes even more important, as our medical needs inevitably grow.
I will read more thoroughly tomorrow, but here’s already something that leapt out at me: “With physician burnout at epidemic levels, we cannot assume that it is merely an individual physician's problem that needs to be remedied on an individual level. System-level changes are also required if we want to reverse this trend.”
Since I wrote earlier today, I have actually found an address at my AMC to which to send concerns about what I believe to be “system failures” of which I’ve become aware. What I find when I write to my electeds is I do best if I can also include sensible approaches to resolve problems I identify. On that score, it’s very hard to tell, from outside the hospital system, what would be most constructive to suggest and, most importantly, would not redound back on the provider or staff, as they are not at fault here. The type of thing I am talking about, as one example, were failures in the referral process for post-surgery PT that look to me as if they could be readily corrected on the admin side. (Whether AI is pertinent here, I don’t really know--I just got that idea from Dr. Wachter’s grand rounds.)
From my unlettered POV the misdiagnosis issue may (stress the word "may") be coming from several sources along with those already mentioned.
1. Diagnoses based on most-likely etiology. The long tails are seldom examined. Why..
2. Because both pathologists and PCPs are overworked, with huge case loads, especially as the Boomers (and I'm one) are aging and flooding the health system.
3. The insurers are influencing the system toward probabilistic diagnoses that are vectored toward 1. and away from more expensive treatment paths? Look at the recent cases of E multilocularis in Italy. Could our system supported such an investigation?
Anything AI can do to assist physicians and patients here will be literally life saving.
I will add that health systems, having lost $$ during the pandemic, are focused on doing anything to increase revenue. Meaning prividers need to see more patients forving them to spend even less clinic time with each than before. Additionally, understaffed ambulatory clinics increases stress on providers risking an increase in errors.
I haven’t scanned all the comments, but the elephant in the room is the “brief encounter”. Is the solution the introduction of yet MORE technology? EMRs were supposed to improve medical care. Instead they turned clinicians and nurses into data entry clerks.... so that burnout is huge and meaningful patient contact has been minimized.
The solution is to provide more time for patient care by reducing data entry and absurdly redundant checklists. Full stop.
You make a good point, and one that concerns me, too, from the perspective of the patient. I wonder what you might think of Bob Wachter’s presentation on AI as an aide to physicians, rather than just more trouble. He speaks to the issue of the EMR directly in his framing of where he thinks we are now: https://youtu.be/2wPZ4kf1kLE?feature=shared
Could we put H. Gilbert Welch on the case? I was profoundly affected by his 2011 book, Overdiagnosed. Its thesis reinforced a generational familial (in my DNA?) distrust of the medical profession and healthcare system that’s lasted a lifetime. (And don’t get me started on Big Pharma).
I never leave my skepticism at the door when entering a physician’s office.
Of course, the emergence of AI and LLM’s put an entirely new spin on it all. Should I add the growing reliance on diagnostic AI to the list?
Haven’t checked out what Dr. Welch has been up to recently but I’d welcome his perspective.
For the record, I do appreciate these Ground Truths. I first became familiar with your work in 2019 reading your seminal book, Deep Medicine.
Not that anyone will see this, but I have been reflecting on Dr. Topol’s excellent post, along with the comments in response, and thought I’d least record what I have been pondering since. Dr. Topol is rightly looking toward the future, and I am persuaded that, in good hands, AI can offer many benefits to the practice of medicine. Dr. Wachter is convincing (to me) on this as well, both as to the future and as to the present. But (yes there is a “but”):
That said, from the end-user (ie patient) point of view, along with a work life in healthcare-related fields, what I see is that health care providers and patients alike are increasingly caught in the cogs of an ever larger, completely impersonal, system the focus of which appears to have far too little to do with providing the best health care possible and far too much to do with increasing revenue to the system.
For me, as a patient and out of my health care-related work, I see the pivotal person in the health care system as the primary care provider. Accordingly, I would like to see our health care systems prioritize investment so that primary care providers have the time and space they need to think beyond Stage 1, see the whole person in front of them, and be enabled, from a holistic perspective, both to treat them directly and also coordinate their care.
Right now, from my vantage point, the administrative and managerial “systems” in which PCPs operate and which patients must navigate are well and truly broken. There is not even anyone, in my own AMC, designated to receive patient concerns about problems in this area, let alone intelligent use of tools, like AI, that might ameliorate this for all concerned. Dr. Wachter, in his grand round presentation, speaks of use of AI to address the admin problems as “low hanging fruit.” What I would like to know is why AMCs and hospital systems generally, are failing so utterly to pick that fruit.
I can well understand the lack of confidence health care providers have in embracing AI tools, as there is little to point to that is positive from the past, notably the severe unintended consequences for health care providers associated with implementing the EMR. Dr. Wachter addresses this in his grand rounds directly, and he believes that, going forward with AI can be different and better, in part because of lessons learned from the problems attendant to implementation of the EMR.
We do not need our health care providers to be glorified data entry clerks. We need to free them to do the job for which they trained, which to my mind is patient care.
I remember seeing similar articles 20 years ago with similar orders of magnitude... and I found them simply unbelievable. I still do.
Why?
There are just over 1,000,000 physicians in the United States.
This study implies that about one LETHAL or permanently disabling diagnostic error is made by every year PER PHYSICIAN. (Yes, of course I know there are PA’s and NP’s, but you get the point.)
Now I don’t consider myself to be a stellar diagnostician. But I knew virtually all the pediatricians in my communities AND heard the scuttlebutt.
In my particular sphere I heard of or was aware of ONE lethal diagnostic error in practice and ONE in training by docs and ONE by a nurse. Each time we were devastated. Far from covering up, we tried to understand and improve.
The magnitude of lethal error implied by such studies - which invariably have assumptions that can skew the results - is not credible.
Thank you, Dr. Topol, for yet another useful, information-filled article relating to AI. I was particularly struck by this:
“With the brief duration of a clinic visit, it is not surprising that there is little time to reflect, because it relies on System 1 thinking, which is automatic, near-instantaneous, reflexive, and intuitive. If physicians had more time to think, to do a search or review the literature, and analyze all of the patient’s data (System 2 thinking), it is possible that diagnostic errors could be reduced.”
Any way that AI can help with this will be to the good, as, from what I can see, the problems physicians face are not only not going away, but it seems are getting worse. Indeed, you link within this paragraph to Danielle’s 2019 NEJM article, where even the first paragraph speaks volumes on the issue: “In modern outpatient medicine, there’s no time to think. We race to cover the bare minimum, sprinting in subsistence-level intellectual mode because that’s all that’s sustainable. Such practice is a petri dish for medical error, patient harm, and physician burnout.”
I’m looking forward to having another tool to help with our practice, every Batman needs a Robin, I’m not sure which we would be in the long run but I do have some skepticism about the primary care study. I think the scenarios were pretty contrived and not like an actual primary care visit which is usually sprawling. Perhaps 15% of what we do is making new diagnoses. There is so much human level coordination of care, balancing conflicting treatments with multiple medical problems, tending to patient preferences about their medical care, patient education etc. etc. A lot of these headlines are sort of all or nothing dualities that lead to people thinking AI vs Doctor. And I certainly don’t want to be lead through diagnosis, treatment, and whatever comes next with cancer by a chat bot. Would love to see more studies where doctors are working with AI, as those results are probably the absolute best for now.
You can count on lots more studies to come. Agree with the contrived nature of the studies performed o date.
Not sure if you practice clinical medicine, but I’m trying to decide whether to use and train Dragon ambient AI for help with documentation of visits. I’m conflicted. The process of writing (dictating with Dragon) a SOAP note after each visit really helps me synthesize and organize my thoughts and plans for the patient, and then provides a narrative I can pick up with at their next visit. I think the clinician brain will atrophy in some respects by delegating too much of this to AI.
Nonetheless efficiency always wins, and so I guess I should try it out. This is the most concrete way AI is inserting itself into outpatient medicine that I can feel - LLM documentation.
I wondered what you might think about this, too, given our past exchanges. It’s very hard to assess this from my perspective as an “end-user,” as I only see what I (and friends) experience. From that vantage point, I am personally aware, in recent times, of three fine, empathic MDs--a PCP, an oncologist, and an endocrinologist--who appear to have burned out and retired early or otherwise just stopped practicing medicine. I also know Danielle Ofri, so when she speaks to this problem (as in the article Dr. Topol links), I sit up and take notice, too. With that preface, what I hope for with AI, as Dr. Topol describes it, is that it can be an aid to free up MD time to focus more fully on the patient in front of her/him. Related, did you happen to see the presentation by Bob Wachter related to this? I thought it spoke to the issues usefully, but would appreciate knowing your view, as always: https://youtu.be/2wPZ4kf1kLE?feature=shared
You are so invested in both sides of the patient doctor relationship, and I am so appreciative that people like you care about how it’s going on this end!
One of the best articles I’ve read in this burnout stuff in terms of family doctors was this from prepandemic times which have only become worse:
https://www.aafp.org/pubs/fpm/issues/2015/0900/p42.html
I’ll try to respond more specifically about AI but I think in general the potential benefits are a trade off, and dehumanization needs to be at the forefront of all AI considerations !
No need to respond, but I just wanted you to know I read, with extreme interest, the series of articles you noted. These are life lessons all of us can use--I particularly loved the work-home separation example of Mr. Rogers😎. So, from the patient POV, here are some individual level takeaways 1) do what I can as a patient to make my time with my PCP most efficient and effective for both of us; 2) be judicious and concise in use of the patient portal; 3) make sure and thank my providers/staff out loud, don’t just think it! (These are all things I do now, but could do better.)
The harder nut to crack is the systemic end of things, and harder still as one patient with such limited vision into the system. This article helps there, too, as I can frame requests and comments with the provider POV in mind. Beyond that, I do want to find a way to speak up constructively when I spot a possible “system” problem, that is, one that is beyond the control of the provider or individual staff to address. Where and how to do that effectively, including suggesting solutions, remains an open question.
So, anyway, thank you so much for engaging. Your time is precious, too, and I have likely use up more than my share here!
You are the best! Thanks for being a force to make the world better, in multiple realms I'm sure!
Thanks so much for responding, and for the link! As for my investment, it’s surely within the interest of us as patients to help our health care providers succeed, if we can! As I and so many friends are getting older, this becomes even more important, as our medical needs inevitably grow.
I will read more thoroughly tomorrow, but here’s already something that leapt out at me: “With physician burnout at epidemic levels, we cannot assume that it is merely an individual physician's problem that needs to be remedied on an individual level. System-level changes are also required if we want to reverse this trend.”
Since I wrote earlier today, I have actually found an address at my AMC to which to send concerns about what I believe to be “system failures” of which I’ve become aware. What I find when I write to my electeds is I do best if I can also include sensible approaches to resolve problems I identify. On that score, it’s very hard to tell, from outside the hospital system, what would be most constructive to suggest and, most importantly, would not redound back on the provider or staff, as they are not at fault here. The type of thing I am talking about, as one example, were failures in the referral process for post-surgery PT that look to me as if they could be readily corrected on the admin side. (Whether AI is pertinent here, I don’t really know--I just got that idea from Dr. Wachter’s grand rounds.)
Another great report!
From my unlettered POV the misdiagnosis issue may (stress the word "may") be coming from several sources along with those already mentioned.
1. Diagnoses based on most-likely etiology. The long tails are seldom examined. Why..
2. Because both pathologists and PCPs are overworked, with huge case loads, especially as the Boomers (and I'm one) are aging and flooding the health system.
3. The insurers are influencing the system toward probabilistic diagnoses that are vectored toward 1. and away from more expensive treatment paths? Look at the recent cases of E multilocularis in Italy. Could our system supported such an investigation?
Anything AI can do to assist physicians and patients here will be literally life saving.
thanks and I'm in full agreement
I will add that health systems, having lost $$ during the pandemic, are focused on doing anything to increase revenue. Meaning prividers need to see more patients forving them to spend even less clinic time with each than before. Additionally, understaffed ambulatory clinics increases stress on providers risking an increase in errors.
I haven’t scanned all the comments, but the elephant in the room is the “brief encounter”. Is the solution the introduction of yet MORE technology? EMRs were supposed to improve medical care. Instead they turned clinicians and nurses into data entry clerks.... so that burnout is huge and meaningful patient contact has been minimized.
The solution is to provide more time for patient care by reducing data entry and absurdly redundant checklists. Full stop.
You make a good point, and one that concerns me, too, from the perspective of the patient. I wonder what you might think of Bob Wachter’s presentation on AI as an aide to physicians, rather than just more trouble. He speaks to the issue of the EMR directly in his framing of where he thinks we are now: https://youtu.be/2wPZ4kf1kLE?feature=shared
Could we put H. Gilbert Welch on the case? I was profoundly affected by his 2011 book, Overdiagnosed. Its thesis reinforced a generational familial (in my DNA?) distrust of the medical profession and healthcare system that’s lasted a lifetime. (And don’t get me started on Big Pharma).
I never leave my skepticism at the door when entering a physician’s office.
Of course, the emergence of AI and LLM’s put an entirely new spin on it all. Should I add the growing reliance on diagnostic AI to the list?
Haven’t checked out what Dr. Welch has been up to recently but I’d welcome his perspective.
For the record, I do appreciate these Ground Truths. I first became familiar with your work in 2019 reading your seminal book, Deep Medicine.
Pamela
Not that anyone will see this, but I have been reflecting on Dr. Topol’s excellent post, along with the comments in response, and thought I’d least record what I have been pondering since. Dr. Topol is rightly looking toward the future, and I am persuaded that, in good hands, AI can offer many benefits to the practice of medicine. Dr. Wachter is convincing (to me) on this as well, both as to the future and as to the present. But (yes there is a “but”):
That said, from the end-user (ie patient) point of view, along with a work life in healthcare-related fields, what I see is that health care providers and patients alike are increasingly caught in the cogs of an ever larger, completely impersonal, system the focus of which appears to have far too little to do with providing the best health care possible and far too much to do with increasing revenue to the system.
For me, as a patient and out of my health care-related work, I see the pivotal person in the health care system as the primary care provider. Accordingly, I would like to see our health care systems prioritize investment so that primary care providers have the time and space they need to think beyond Stage 1, see the whole person in front of them, and be enabled, from a holistic perspective, both to treat them directly and also coordinate their care.
Right now, from my vantage point, the administrative and managerial “systems” in which PCPs operate and which patients must navigate are well and truly broken. There is not even anyone, in my own AMC, designated to receive patient concerns about problems in this area, let alone intelligent use of tools, like AI, that might ameliorate this for all concerned. Dr. Wachter, in his grand round presentation, speaks of use of AI to address the admin problems as “low hanging fruit.” What I would like to know is why AMCs and hospital systems generally, are failing so utterly to pick that fruit.
I can well understand the lack of confidence health care providers have in embracing AI tools, as there is little to point to that is positive from the past, notably the severe unintended consequences for health care providers associated with implementing the EMR. Dr. Wachter addresses this in his grand rounds directly, and he believes that, going forward with AI can be different and better, in part because of lessons learned from the problems attendant to implementation of the EMR.
We do not need our health care providers to be glorified data entry clerks. We need to free them to do the job for which they trained, which to my mind is patient care.
I welcome any thoughts on this anyone may have.
I feel the need to make one other comment.
I remember seeing similar articles 20 years ago with similar orders of magnitude... and I found them simply unbelievable. I still do.
Why?
There are just over 1,000,000 physicians in the United States.
This study implies that about one LETHAL or permanently disabling diagnostic error is made by every year PER PHYSICIAN. (Yes, of course I know there are PA’s and NP’s, but you get the point.)
Now I don’t consider myself to be a stellar diagnostician. But I knew virtually all the pediatricians in my communities AND heard the scuttlebutt.
In my particular sphere I heard of or was aware of ONE lethal diagnostic error in practice and ONE in training by docs and ONE by a nurse. Each time we were devastated. Far from covering up, we tried to understand and improve.
The magnitude of lethal error implied by such studies - which invariably have assumptions that can skew the results - is not credible.
Somebody’s gotta look at the forest, not just the trees.