Even though one can use AI to deal with lot of the paperwork, there still should be an effort to reduce the paperwork being asked for, down to what is actually needed. (It shouldn't just be, let's leave all that in there, since we are having a machine do it anyhow.) The examples here seemed hard to avoid and so a perfect use.
Dr. McBride's observation about the vignettes deserves to be pushed one layer further, because it explains more than she lets it.
Every study she cites that flatters the models handed them a case whose facts arrived complete, consistent, and true. That is what a vignette is. The Nature Medicine study she cites (Johri et al., 2025) found that accuracy fell as soon as the facts had to be gathered through a conversation instead. The same thing happens when the facts have to be gathered from the chart. Hager and colleagues (Nature Medicine, 2024) tested models on 2,400 real cases and found that giving a model all of the available information lowered its accuracy, and that reordering the same information moved the result by as much as 18 percentage points. Omar and colleagues (Communications Medicine, 2025) planted one false detail in the clinical material and found that six models repeated or elaborated it in 50 to 82 percent of outputs.
Generative AI succeeds wherever the world supplies a check on its output: code compiles or it does not; a cited paragraph can be opened. It fails wherever no such check exists. In medicine no such check exists, because the record itself is not a source of truth. It is a pile of assertions from systems that never agreed with one another: the diagnosis pasted forward for a decade, the medication list that describes a different year, the pathology report that sits in another institution's chart. A model reading that record is reading a vignette that nobody wrote and nobody verified. The failure at the bedside that Dr. McBride describes and the failure in the chart are the same failure. The premises were never established.
That is why her proposed division of labor worries me. "Give AI the gruntwork" hands the model the one task in which an unverified premise does the most durable harm: writing the record. Williams and colleagues (PLOS Digital Health, 2025) found that 42 percent of GPT-4 emergency department summaries contained hallucinations and 47 percent contained omissions. Topaz and colleagues (npj Digital Medicine, 2025) found that ambient scribes fabricate, omit, and misattribute speakers at measurable rates. A note the model wrote becomes the chart the next model reads.
The question beneath the question has a twin: the fact beneath the assertion. Until something establishes, patient by patient, which of the record's claims are true, with the evidence attached, the conversation Dr. McBride wants to protect will be happening over premises nobody has checked. That is not an argument against the models. It is an argument that the missing piece sits under them, not in front of them.
There are solutions that are emerging to fix this problem But they come from the Cognitive AI space, not the generative AI space. These are semantic issues, not fluency issues. They require deterministic solutions, not correlative engines. But until they are foundational, the proposed solution here will not succeed.
AI is a tool, and like a calculator, it can replace some skills while requiring us to develop others. Of course, it is much more powerful than a calculator, so if we do not work hard on the “soft” skills that so many of us have lost in Medicine 2.0, we will become obsolete. There are many things AI cannot do, but people will sacrifice those for the things it can do so much better and faster and reject the current practice of primary care if we do not adapt and adopt. I hope that we are flexible, smart, ingenious, and caring enough to do just that, but I guess time will tell.
The human interaction can be invaluable, but honestly, how many PCP's actually observe their patients with the intuition and empathy the author does? I use Grok, Gemini, Chat GPT, Claude, and Alter to get information about my medical issues. Increasingly they end their response with a list of questions I should ask my Dr.
I use multiple Ai's to suss out hallucinations as much as possible, also ask for, and read primary source material when talking about major issues such as surgery.
As a retired physician I appreciate the amazing breadth and depth AI brings to the search for information.
I agree AI is a wonderful and very powerful tool. I also agree that primary care is on life support such that most doctors don't have time to elicit information from patients that doesn't show up on a lab test.
But the solution, to me, isn't to leverage AI alone to reinvent our health care system; it's to resuscitate primary care by pointing AI at the administrative burdens that have hollowed out primary care (and medicine in general) in order to elevate, not eliminate, the patient-doctor relationship - and to scale (and pay for) a model of primary care where the commodity is conversation, not just transactions.
Given that psychosocial and emotional factors are a substantial and growing driver of primary care, I wonder if the majority of humans would be on antidepressants by year-end if AI was writing prescriptions. Seems like it’s a trade of factory primary care being quick to write scripts versus an AI bot writing scripts for every solution. Thanks for the article. Primary care is much bigger than simply writing a script for every problem, but rather is an adventure of discovery.
Amen 100%. As I wrote to a reader above, primary care is on life support, such that most doctors don't have time to elicit information from patients that doesn't show up on a lab test.
The solution, to me, isn't to leverage AI alone to reinvent our health care system; it's to resuscitate primary care by pointing AI at the administrative burdens that have hollowed out primary care (and medicine in general) in order to elevate, not eliminate, the patient-doctor relationship - and to scale (and pay for) a model of primary care where the commodity is conversation, not just transactions.
Definitely. I think AI in the future could also be like having a consultant in the room, giving the doctor prompts to consider a broader differential diagnosis or an alternative therapy. The value of the doctor is their breadth of experience, the human connection, and the ability to identify when to reassure a patient and when to dig in until we find an answer. Say I have two patients with chest pain. They have mostly identical symptoms. Maybe they would even give the same anxiety score on the GAD7 questionnaire. However, because of relationship, I know one of these patients tends to underreport emotional symptoms and is going through a tough job transition. That is an important data point that I doubt AI could ever pick up.
Totally wrong. How do you account for the burgeoning popularity of AI girl/boyfriends - 1/3 of US adults and growing? Or the 30-50% using AI for counseling or psych therapy? With each iteration these percents will increase. People bond just fine with LLMs, and have transparent, constructive conversations with them. I am not a primary care doc (I'm a periodontist) so count me as somewhat clueless. But my interactions with Claude, Chat, Gemini, etc. have been overwhelmingly constructive wrt medical, diagnostic, surgical, and relational advice. I filter its output through my PCP and specialists, but I usually benefit more from AI than my "real" docs. Given the ridiculous time pressure on most primary docs these days, that is no surprise. And as Dr. Cifu and his colleagues on Sensible Medicine, etc., acknowledge, the average MD bases little of his/her practice on rigorous understanding and application of clinical research - a foible not shared by AI (although it is not flawless in its understanding of the science). I routinely recommend that my patients check whatever I tell them using AI when I hit a diagnostic challenge. (Still not on the BFF or shrink page though ;)
IMO AI is an existential threat to primary care medicine, sadly for the profession. Whether that will be a good thing for humanity remains to be seen. I'm betting yes.
We can agree to disagree about the importance of human-to-human care. I think it's essential - it's just in short supply and difficult to access. Machines are better than humans in myriad ways. ChatGPT can answer the question you ask it faster, more accurately, and more empathetically than most doctors! But most medical issues in primary care require understanding another human -- i.e., consciousness -- and the ability to build trust. Sure, many (most?) Americans already trust AI more than they trust their doctor - and that, to me, isn't a reason to remove humans from the most human endeavor there is - providing care. Let's use AI to rebuild primary care, so that every American has unfettered access to a primary care medical home - where AI is doing everything BUT the conversation.
Thanks for the conversation. I was hoping to drop off since I am getting tiresome. But you seem wildly idealistic. "Let's use AI to rebuild primary care". Sorry but NFW. There are no incentives or interest groups who are motivated and/or capable of this. All the energy is going the other way. Increased impersonalization, corporatization, time constraints, "efficiency (thus not enough beds during the Covid crisis)", GDMT (which essentially never argues for less treatment or god forbid watchful waiting), and record-keeping mandates (AI for EMR is one area where you and I totally agree, and it is happening). Not to mention burnout, staff shortages, and a coming crisis of insufficient supply of MD/NP/PA's. Plus, AI is just plain smarter, better informed, more empathic, and more rational than many docs (not mine - I love my PCP, even if he occasionally contradicts Mandrola, Cifu, Prasad, Foy, Welch...). And at the behest of Google, Anthropic, Openai, etc., it is getting more attractive all the time, as human doctors degrade. But my opinion and yours mean nothing. The population is already voting with its feet. So long doc ;) At least AI might mitigate the doctor shortage.
We will have to disagree that "There are no incentives or interest groups who are motivated and/or capable of this." Tech will reinvent healthcare. It's already happening. The tech companies that are allowing physicians to guide the AI model specifically not to replace the human elements of care are the ones I am watching closely and aligning myself with. I believe my opinion matters, but we can disagree on that too.
Said every author and bookseller who got in bed with Amazon.
One last jab: "Why would anyone want to ride in a smoky mechanical box instead of a horse?" "Why would anyone prefer industrial bread and milk when they're getting a cold bottle and a warm loaf delivered to their house in the morning?" "Why would anyone want to travel five miles up in an aluminum can that will kill them when it crashes, when they can get there in a train?" "Why would anyone want to buy a book without leafing through it in a store?" Why would anyone want to have their clothing delivered to their house without trying them on in a store? Ad nauseum.
You and Dr. Cifu conflate your gratification seeing patients with their gratification seeing you. No one wants to go to the doctor (take it from me, a dentist). Sorry. If they did, they would ask you to lunch. They go for 1. Prevention and 2. Diagnosis. They appreciate your hand-holding and the placebo effect when they need you. Otherwise, no.
BTW even the placebo effect will be solved by AI, as it has already solved the empathy/human contact challenge. What a world.
You nailed it. When a patient learns how to use AI efficiently, it will beat a 20 minute primary care visit 9 times out of 10. A part of the reason is that EMRs they use, like Epic, are are hopelessly antiquated and cumbersome.. Just upload your symptoms, hx, labs and imaging to Chat GPT and it will put it all together for you. Also, you get real time answers without having to wait 3 months to schedule a appointment
It already is just a "set of transactions" in most cases: 2 -3 months to get an appointment with your PCP; and messages sent to them in between visits are "screened' by non-clinical staff (often never seen by your own doctor). Then you get referred to another doctor who can see you right away -- but they know nothing about you or your history. No continuity of care = transactional healthcare. You and your colleagues allowed this to happen.
Interesting perspective. Yes healthcare has been reduced to a set of transactions. But who is responsible? I believe it's the massive profits being made by insurance companies and the pharmaceutical industry - plus policy that has decided to spend on 5 c on the dollar on primary care - that has done this. Not doctors. We are trying to provide care in an impossible system with ass-backwards incentives - one that rewards volume of patients seen over quality of care/outcomes. Hence the burnout, the mass exodus, and the moral injury experienced by primary care docs. We went into medicine to treat people, not diagnosis codes.
great for the 0.1% of people in America like you that have all their medical history, every x-ray and ultrasound and MRI and all their blood and urine tests neatly filed on their laptop to share with Chat GPT. Cheers
No. I access the patient portal for my physicians the same way anyone can. Perhaps you were not aware that, legally, all electronic health information must be accessible to patients on line. So not 0 1%, closer to 100% if the person has access to the internet.
I can definitely tell you are not a primary care doctor or you wouldn't respond to such a brilliantly and accurately formed writing the way you just did. How many people have you sat next to in a room like the patient she just described in the article? Or the woman who walked in to find her mother had attempted suicide and came directly to my office? Or the people who come to you just their normal annual visits but want to hug you and just discuss their normal chronic illnesses? You miss the point of the complexity of the human experience. ChatGPT can't hold you hand during the hospice discussion. It makes me concerned that you have so little insight in to the human condition and the role it plays in all of healthcare.
While I have long agreed with you and try hard to build a human connection and community with my patients, the harsh truth is that doctors like you and the amazing author of this piece are a vanishingly small minority.
The past few years have made it clear that the majority of doctors have become almost soulless robots. You mention hospice - in Canada now doctors give patients assisted suicide pamphlets before they even meet with an oncologist, they kill the elderly against their will, they have disposed of tens of thousands through MAID.
In my practice, I’m full of patients who got fired by their PCPs for not strictly following the AAP vax schedule, including COVID boosters. So if you try to ask for DTaP and MMR but are hesitant about Covid or Hep B, the doctor doesn’t hug you and share your worries- he fires you!
And don’t even get me started on the masking…
Unfortunately, doctors acted with tremendously inhuman arrogance for years because they thought they were irreplaceable, but by making themselves so inhuman it makes it all the easier to replace them with AI!
Thanks. I get your point, I just totally disagree. I am >70 and have seen it all, including your (and the author's) scenarios, with friends, family, and patients. Trauma, cancer, suicides, death...Not as an MD. If you are concerned for me, it is bc you don't know me, even a little bit. Just to keep piling on, Claude knows me a lot better than you. And is much more empathic.
Let's come back in 3-4 years. I'll buy you dinner if AI has not taken over much of primary care, including the holding hands part.
Sensible Medicine has been great at educating readers on critical appraisal concepts lacking in most physicians and “providers”. So, consider the way-to-common 80 y/o patient complaining about the $500/month prescription she received (and can’t afford) from the …ologist who e-scribed the drug while proudly documenting compliance with “GDMT” (guideline directed medical therapy). Of course, the drug only offers a 1.5% absolute risk reduction instead of the 82% relative risk reduction touted by the drug company and neither the patient or the “provider” are aware of, or understand, the difference. Furthermore, there were no multiple co-morbidity patients over age 60 in the research trial that was used to develop the “evidence” that became the “guideline” being followed by the “provider”. But, why couldn’t AI be embedded in the EMR to offer up a brief and honest critical appraisal of the “evidence” cited in support of the treatment being prescribed to the specific patient? Unfortunately, the challenge would be to prevent big pharma and others from corrupting the AI.
"I cannot imagine a time when a person, a patient, will be satisfied making a complicated decision about their health with only a computer consultant."
I already made decisions like that regarding prostate cancer and atrial fibrillation treatment using ChatGPT. It could read my entire medical history and a huge number of trials quicker than my regular doctor could log into Epic.
What do I make of the Pangram opinion that 8% of the text of the article was AI generated? Perhaps simply that editing is one of the tasks that can be off-loaded to AI.
How to tell if a current EHR using doc is still alive….. is his index finger still tapping as if clicking on the mouse … We are advancing at warp speed on fancy stuff but still don’t have common stops for more common situations like having to enter every first visit of the year that the amputated leg remains, absent. That the pt had bilateral mastectomies so mammogram not indicated. That the pt in a coma is not well developed, well nourished , in no acute distress and alert and oriented x 3. Or asking a Caucasian pt on preop anesthesia screen , whether they have sickle cell. Need to CTL-ALT-DEL and focus on some basic issues that we encounter daily. We are becoming templaters and box checkers.
AI isn’t making anyone smarter. It brings laziness, loss of critical or imaginative thinking, and is bringing on a new generation of people watching a screen or phone. I loved the article and it’s 100% true. Great job.
I sure appreciate this article. I am doing my best not to become that crotchety jaded old man with AI’s miraculous help for humans - although I AM jaded by the propaga-I mean media-that says AI will kill us all by next week. Besides, humans have been using AI to kill humans for awhile now. I digress, er, sorry. I have watched what I call clicker medicine replace critical thinking, a very real politicization of medical care, and wonder why anyone would even WANT to be in primary care. We need as many Dr. McBrides as possible, and if my own senses notice anything other than human in my care, I’ll find another PCP while I still have an option.
Wow after reading some of the comments, I wonder if I would have survived at all had AI been around in the year I was born. As it was, euthanasia was suggested. Yet here I am 70+ surgeries later, in my 60’s, having had two absolutely incredible careers, an amazing bride, two daughters, 4 grands, and more love than anyone could ever ask for. Yes ma’am, I’ll take that human touch every time. And yes, I educate the heck outta my decisions but seriously trusting AI alone is like AI music to me. It ain’t real and it sounds like doo doo. I’ll take the heart and soul of a human every single time.
Thank you for this.
Even though one can use AI to deal with lot of the paperwork, there still should be an effort to reduce the paperwork being asked for, down to what is actually needed. (It shouldn't just be, let's leave all that in there, since we are having a machine do it anyhow.) The examples here seemed hard to avoid and so a perfect use.
Dr. McBride's observation about the vignettes deserves to be pushed one layer further, because it explains more than she lets it.
Every study she cites that flatters the models handed them a case whose facts arrived complete, consistent, and true. That is what a vignette is. The Nature Medicine study she cites (Johri et al., 2025) found that accuracy fell as soon as the facts had to be gathered through a conversation instead. The same thing happens when the facts have to be gathered from the chart. Hager and colleagues (Nature Medicine, 2024) tested models on 2,400 real cases and found that giving a model all of the available information lowered its accuracy, and that reordering the same information moved the result by as much as 18 percentage points. Omar and colleagues (Communications Medicine, 2025) planted one false detail in the clinical material and found that six models repeated or elaborated it in 50 to 82 percent of outputs.
Generative AI succeeds wherever the world supplies a check on its output: code compiles or it does not; a cited paragraph can be opened. It fails wherever no such check exists. In medicine no such check exists, because the record itself is not a source of truth. It is a pile of assertions from systems that never agreed with one another: the diagnosis pasted forward for a decade, the medication list that describes a different year, the pathology report that sits in another institution's chart. A model reading that record is reading a vignette that nobody wrote and nobody verified. The failure at the bedside that Dr. McBride describes and the failure in the chart are the same failure. The premises were never established.
That is why her proposed division of labor worries me. "Give AI the gruntwork" hands the model the one task in which an unverified premise does the most durable harm: writing the record. Williams and colleagues (PLOS Digital Health, 2025) found that 42 percent of GPT-4 emergency department summaries contained hallucinations and 47 percent contained omissions. Topaz and colleagues (npj Digital Medicine, 2025) found that ambient scribes fabricate, omit, and misattribute speakers at measurable rates. A note the model wrote becomes the chart the next model reads.
The question beneath the question has a twin: the fact beneath the assertion. Until something establishes, patient by patient, which of the record's claims are true, with the evidence attached, the conversation Dr. McBride wants to protect will be happening over premises nobody has checked. That is not an argument against the models. It is an argument that the missing piece sits under them, not in front of them.
There are solutions that are emerging to fix this problem But they come from the Cognitive AI space, not the generative AI space. These are semantic issues, not fluency issues. They require deterministic solutions, not correlative engines. But until they are foundational, the proposed solution here will not succeed.
YOU SAID IT!
AI is a tool, and like a calculator, it can replace some skills while requiring us to develop others. Of course, it is much more powerful than a calculator, so if we do not work hard on the “soft” skills that so many of us have lost in Medicine 2.0, we will become obsolete. There are many things AI cannot do, but people will sacrifice those for the things it can do so much better and faster and reject the current practice of primary care if we do not adapt and adopt. I hope that we are flexible, smart, ingenious, and caring enough to do just that, but I guess time will tell.
Agree. Finding that balance is difficult but necessary work.
The human interaction can be invaluable, but honestly, how many PCP's actually observe their patients with the intuition and empathy the author does? I use Grok, Gemini, Chat GPT, Claude, and Alter to get information about my medical issues. Increasingly they end their response with a list of questions I should ask my Dr.
I use multiple Ai's to suss out hallucinations as much as possible, also ask for, and read primary source material when talking about major issues such as surgery.
As a retired physician I appreciate the amazing breadth and depth AI brings to the search for information.
I agree AI is a wonderful and very powerful tool. I also agree that primary care is on life support such that most doctors don't have time to elicit information from patients that doesn't show up on a lab test.
But the solution, to me, isn't to leverage AI alone to reinvent our health care system; it's to resuscitate primary care by pointing AI at the administrative burdens that have hollowed out primary care (and medicine in general) in order to elevate, not eliminate, the patient-doctor relationship - and to scale (and pay for) a model of primary care where the commodity is conversation, not just transactions.
Given that psychosocial and emotional factors are a substantial and growing driver of primary care, I wonder if the majority of humans would be on antidepressants by year-end if AI was writing prescriptions. Seems like it’s a trade of factory primary care being quick to write scripts versus an AI bot writing scripts for every solution. Thanks for the article. Primary care is much bigger than simply writing a script for every problem, but rather is an adventure of discovery.
Amen 100%. As I wrote to a reader above, primary care is on life support, such that most doctors don't have time to elicit information from patients that doesn't show up on a lab test.
The solution, to me, isn't to leverage AI alone to reinvent our health care system; it's to resuscitate primary care by pointing AI at the administrative burdens that have hollowed out primary care (and medicine in general) in order to elevate, not eliminate, the patient-doctor relationship - and to scale (and pay for) a model of primary care where the commodity is conversation, not just transactions.
Definitely. I think AI in the future could also be like having a consultant in the room, giving the doctor prompts to consider a broader differential diagnosis or an alternative therapy. The value of the doctor is their breadth of experience, the human connection, and the ability to identify when to reassure a patient and when to dig in until we find an answer. Say I have two patients with chest pain. They have mostly identical symptoms. Maybe they would even give the same anxiety score on the GAD7 questionnaire. However, because of relationship, I know one of these patients tends to underreport emotional symptoms and is going through a tough job transition. That is an important data point that I doubt AI could ever pick up.
Huzzah!!!
Totally wrong. How do you account for the burgeoning popularity of AI girl/boyfriends - 1/3 of US adults and growing? Or the 30-50% using AI for counseling or psych therapy? With each iteration these percents will increase. People bond just fine with LLMs, and have transparent, constructive conversations with them. I am not a primary care doc (I'm a periodontist) so count me as somewhat clueless. But my interactions with Claude, Chat, Gemini, etc. have been overwhelmingly constructive wrt medical, diagnostic, surgical, and relational advice. I filter its output through my PCP and specialists, but I usually benefit more from AI than my "real" docs. Given the ridiculous time pressure on most primary docs these days, that is no surprise. And as Dr. Cifu and his colleagues on Sensible Medicine, etc., acknowledge, the average MD bases little of his/her practice on rigorous understanding and application of clinical research - a foible not shared by AI (although it is not flawless in its understanding of the science). I routinely recommend that my patients check whatever I tell them using AI when I hit a diagnostic challenge. (Still not on the BFF or shrink page though ;)
IMO AI is an existential threat to primary care medicine, sadly for the profession. Whether that will be a good thing for humanity remains to be seen. I'm betting yes.
We can agree to disagree about the importance of human-to-human care. I think it's essential - it's just in short supply and difficult to access. Machines are better than humans in myriad ways. ChatGPT can answer the question you ask it faster, more accurately, and more empathetically than most doctors! But most medical issues in primary care require understanding another human -- i.e., consciousness -- and the ability to build trust. Sure, many (most?) Americans already trust AI more than they trust their doctor - and that, to me, isn't a reason to remove humans from the most human endeavor there is - providing care. Let's use AI to rebuild primary care, so that every American has unfettered access to a primary care medical home - where AI is doing everything BUT the conversation.
Thanks for the conversation. I was hoping to drop off since I am getting tiresome. But you seem wildly idealistic. "Let's use AI to rebuild primary care". Sorry but NFW. There are no incentives or interest groups who are motivated and/or capable of this. All the energy is going the other way. Increased impersonalization, corporatization, time constraints, "efficiency (thus not enough beds during the Covid crisis)", GDMT (which essentially never argues for less treatment or god forbid watchful waiting), and record-keeping mandates (AI for EMR is one area where you and I totally agree, and it is happening). Not to mention burnout, staff shortages, and a coming crisis of insufficient supply of MD/NP/PA's. Plus, AI is just plain smarter, better informed, more empathic, and more rational than many docs (not mine - I love my PCP, even if he occasionally contradicts Mandrola, Cifu, Prasad, Foy, Welch...). And at the behest of Google, Anthropic, Openai, etc., it is getting more attractive all the time, as human doctors degrade. But my opinion and yours mean nothing. The population is already voting with its feet. So long doc ;) At least AI might mitigate the doctor shortage.
We will have to disagree that "There are no incentives or interest groups who are motivated and/or capable of this." Tech will reinvent healthcare. It's already happening. The tech companies that are allowing physicians to guide the AI model specifically not to replace the human elements of care are the ones I am watching closely and aligning myself with. I believe my opinion matters, but we can disagree on that too.
Said every author and bookseller who got in bed with Amazon.
One last jab: "Why would anyone want to ride in a smoky mechanical box instead of a horse?" "Why would anyone prefer industrial bread and milk when they're getting a cold bottle and a warm loaf delivered to their house in the morning?" "Why would anyone want to travel five miles up in an aluminum can that will kill them when it crashes, when they can get there in a train?" "Why would anyone want to buy a book without leafing through it in a store?" Why would anyone want to have their clothing delivered to their house without trying them on in a store? Ad nauseum.
You and Dr. Cifu conflate your gratification seeing patients with their gratification seeing you. No one wants to go to the doctor (take it from me, a dentist). Sorry. If they did, they would ask you to lunch. They go for 1. Prevention and 2. Diagnosis. They appreciate your hand-holding and the placebo effect when they need you. Otherwise, no.
BTW even the placebo effect will be solved by AI, as it has already solved the empathy/human contact challenge. What a world.
https://dl.acm.org/doi/10.1145/3529225
https://pmc.ncbi.nlm.nih.gov/articles/PMC12356606/
You nailed it. When a patient learns how to use AI efficiently, it will beat a 20 minute primary care visit 9 times out of 10. A part of the reason is that EMRs they use, like Epic, are are hopelessly antiquated and cumbersome.. Just upload your symptoms, hx, labs and imaging to Chat GPT and it will put it all together for you. Also, you get real time answers without having to wait 3 months to schedule a appointment
I think we conceptualize health in a very different way :)
But that's okay - for some people it already is - and always will be - a set of transactions
It already is just a "set of transactions" in most cases: 2 -3 months to get an appointment with your PCP; and messages sent to them in between visits are "screened' by non-clinical staff (often never seen by your own doctor). Then you get referred to another doctor who can see you right away -- but they know nothing about you or your history. No continuity of care = transactional healthcare. You and your colleagues allowed this to happen.
Interesting perspective. Yes healthcare has been reduced to a set of transactions. But who is responsible? I believe it's the massive profits being made by insurance companies and the pharmaceutical industry - plus policy that has decided to spend on 5 c on the dollar on primary care - that has done this. Not doctors. We are trying to provide care in an impossible system with ass-backwards incentives - one that rewards volume of patients seen over quality of care/outcomes. Hence the burnout, the mass exodus, and the moral injury experienced by primary care docs. We went into medicine to treat people, not diagnosis codes.
great for the 0.1% of people in America like you that have all their medical history, every x-ray and ultrasound and MRI and all their blood and urine tests neatly filed on their laptop to share with Chat GPT. Cheers
No. I access the patient portal for my physicians the same way anyone can. Perhaps you were not aware that, legally, all electronic health information must be accessible to patients on line. So not 0 1%, closer to 100% if the person has access to the internet.
I can definitely tell you are not a primary care doctor or you wouldn't respond to such a brilliantly and accurately formed writing the way you just did. How many people have you sat next to in a room like the patient she just described in the article? Or the woman who walked in to find her mother had attempted suicide and came directly to my office? Or the people who come to you just their normal annual visits but want to hug you and just discuss their normal chronic illnesses? You miss the point of the complexity of the human experience. ChatGPT can't hold you hand during the hospice discussion. It makes me concerned that you have so little insight in to the human condition and the role it plays in all of healthcare.
Amen!
Nope, Shawn. AI beats a 20 minute Primary Care visit 9 times out of 10. Plus,you don't have to wait 3 months to get an appointment.
While I have long agreed with you and try hard to build a human connection and community with my patients, the harsh truth is that doctors like you and the amazing author of this piece are a vanishingly small minority.
The past few years have made it clear that the majority of doctors have become almost soulless robots. You mention hospice - in Canada now doctors give patients assisted suicide pamphlets before they even meet with an oncologist, they kill the elderly against their will, they have disposed of tens of thousands through MAID.
In my practice, I’m full of patients who got fired by their PCPs for not strictly following the AAP vax schedule, including COVID boosters. So if you try to ask for DTaP and MMR but are hesitant about Covid or Hep B, the doctor doesn’t hug you and share your worries- he fires you!
And don’t even get me started on the masking…
Unfortunately, doctors acted with tremendously inhuman arrogance for years because they thought they were irreplaceable, but by making themselves so inhuman it makes it all the easier to replace them with AI!
Much more here:
https://gaty.substack.com/p/the-three-wise-men-walk-into-a-doctors
Thanks. I get your point, I just totally disagree. I am >70 and have seen it all, including your (and the author's) scenarios, with friends, family, and patients. Trauma, cancer, suicides, death...Not as an MD. If you are concerned for me, it is bc you don't know me, even a little bit. Just to keep piling on, Claude knows me a lot better than you. And is much more empathic.
Let's come back in 3-4 years. I'll buy you dinner if AI has not taken over much of primary care, including the holding hands part.
(Or maybe it will subjugate or kill us all.)
Sensible Medicine has been great at educating readers on critical appraisal concepts lacking in most physicians and “providers”. So, consider the way-to-common 80 y/o patient complaining about the $500/month prescription she received (and can’t afford) from the …ologist who e-scribed the drug while proudly documenting compliance with “GDMT” (guideline directed medical therapy). Of course, the drug only offers a 1.5% absolute risk reduction instead of the 82% relative risk reduction touted by the drug company and neither the patient or the “provider” are aware of, or understand, the difference. Furthermore, there were no multiple co-morbidity patients over age 60 in the research trial that was used to develop the “evidence” that became the “guideline” being followed by the “provider”. But, why couldn’t AI be embedded in the EMR to offer up a brief and honest critical appraisal of the “evidence” cited in support of the treatment being prescribed to the specific patient? Unfortunately, the challenge would be to prevent big pharma and others from corrupting the AI.
"I cannot imagine a time when a person, a patient, will be satisfied making a complicated decision about their health with only a computer consultant."
I already made decisions like that regarding prostate cancer and atrial fibrillation treatment using ChatGPT. It could read my entire medical history and a huge number of trials quicker than my regular doctor could log into Epic.
💯 Richard!
Exactly.
What do I make of the Pangram opinion that 8% of the text of the article was AI generated? Perhaps simply that editing is one of the tasks that can be off-loaded to AI.
I did appreciate the article.
You can make of it what you will! 100% human written - I use Claude for final edits and flow. :)
How to tell if a current EHR using doc is still alive….. is his index finger still tapping as if clicking on the mouse … We are advancing at warp speed on fancy stuff but still don’t have common stops for more common situations like having to enter every first visit of the year that the amputated leg remains, absent. That the pt had bilateral mastectomies so mammogram not indicated. That the pt in a coma is not well developed, well nourished , in no acute distress and alert and oriented x 3. Or asking a Caucasian pt on preop anesthesia screen , whether they have sickle cell. Need to CTL-ALT-DEL and focus on some basic issues that we encounter daily. We are becoming templaters and box checkers.
AI isn’t making anyone smarter. It brings laziness, loss of critical or imaginative thinking, and is bringing on a new generation of people watching a screen or phone. I loved the article and it’s 100% true. Great job.
Thanks, Gene.
I sure appreciate this article. I am doing my best not to become that crotchety jaded old man with AI’s miraculous help for humans - although I AM jaded by the propaga-I mean media-that says AI will kill us all by next week. Besides, humans have been using AI to kill humans for awhile now. I digress, er, sorry. I have watched what I call clicker medicine replace critical thinking, a very real politicization of medical care, and wonder why anyone would even WANT to be in primary care. We need as many Dr. McBrides as possible, and if my own senses notice anything other than human in my care, I’ll find another PCP while I still have an option.
I hear you!
Wow after reading some of the comments, I wonder if I would have survived at all had AI been around in the year I was born. As it was, euthanasia was suggested. Yet here I am 70+ surgeries later, in my 60’s, having had two absolutely incredible careers, an amazing bride, two daughters, 4 grands, and more love than anyone could ever ask for. Yes ma’am, I’ll take that human touch every time. And yes, I educate the heck outta my decisions but seriously trusting AI alone is like AI music to me. It ain’t real and it sounds like doo doo. I’ll take the heart and soul of a human every single time.
Ditto. 👍