This month’s episode of “News You Can Use” on HealthcareNOWRadio features news from the month of Aug 2023
I am joined by
The show that gives you a quick insight into the latest news, twists, turns, and debacles going on in healthcare with my friend and co-host Craig Joseph, MD (@CraigJoseph) Chief Medical Officer at Nordic Consulting Partners, and myself, where every diagnosis comes with a side of humor. We hope you stay curious, stay engaged, and keep seeking the truth in healthcare in a world that thrives on information.
Buckle up as we dive into the ER of excitement, the ICU of irrationality, and the waiting room of wacky wisdom in this month’s show that features a review of:
This episode includes a look at Artificial Intelligence and ChatGPT and their use in healthcare diagnostics. We highlight the case of ChatGPT providing valuable insights into a patient’s condition, ultimately leading to a diagnosis of tethered cord syndrome, something the many doctors that this mother and son had visited had not thought of. Although the patient’s mother, was not a medical professional, she used the AI to gather a list of potential diagnoses, leading to a successful outcome.
Even Robots Need a Wardrobe Change
With increasing usage of this tool and others like it, there are challenges and risks and we share some tips on improving the results by customizing instructions for the AI, allowing users to tailor AI responses to their specific needs. This customization can lead to more relevant and valuable information, and might even include asking the AI to “Take a chill pill”, and we look at a breakdown of where AI might be taking over activities and roles in healthcare, but looks like the role of “healthcare influencer” now needs to come with a health warning.
Listen in to hear our discussion on Drug Pricing and where it has gone mad and who you should be blaming with challenges thanks to the lack of transparency in healthcare costs. Patients often face varying prices for the same medication, and the blame for this issue seems to lie with the complex healthcare system itself, although there is plenty of blame to go around.
The Fountain of Youth or Just a Sweet Coincidence?
Lastly, we review the Epic Research published on Metformin, a diabetes medication and its potential for extending lifespan and improving various health outcomes. While more research is needed, the data suggests positive effects beyond diabetes management.
We hope you enjoy our take on the latest news and developments in healthcare and want to help you keep untangling the web of information, dodging the sensational pitfalls, and emerging victorious, albeit a little dizzy, on the other side. In the end, the stories we uncover, and the discussions we ignite, all shape the narrative of our shared future. We want to hear from you especially if you have topics covered or questions you’d like answered. You can reach out directly via the contact form on my website, or send a message on LinkedIn to Craig or me.
Until next week keep solving healthcare’s mysteries before they become your emergencies
Listen live at 4:00 AM, 12:00 Noon or 8:00 PM ET, Monday through Friday for the next week at HealthcareNOW Radio. After that, you can listen on demand (See podcast information below.) Join the conversation on Twitter at #TheIncrementalist.
Listen along on HealthcareNowRadio or on SoundCloud
Raw Transcript
Nick van Terheyden
So first off, Craig, let’s talk about the Tethered cord syndrome and some of the original work that’s being done for patients using chat GPT. As a diagnostic tool, we saw some amazing progress with this right?
Craig Joseph
Well, we have I think, AI seems to be a thing. That’s if you listen to us, you hear us talking about that. And this was recent, there was a recent discussion in a scholarly journal called The
Nick van Terheyden
Today Show today don’t come today show
Craig Joseph
from NBC where I get all of my medical information. I don’t really but they did. They did call out a this, this the story of a mom who, who had a child who was just always having rather disparate problems that that no one could, no one could, you know, look at 1.1 solution for this problem. And, you know, he was doing weird things he was, he was acting out from time to time, and he seemed to be getting better with with pain medicine, like Tylenol, or Motrin, and, and he had other kind of concerns that were maybe dental or maybe psychiatric, or maybe behavioral health or, and he had seen all of
Nick van Terheyden
every, every physicians, perfect patient that they look forward to seeing and, you know, in fairness, I don’t want to be mean, because obviously terrible for the child. But what a nightmare, right? I mean, they present and you just, you’ve got all of these things coming at you.
Craig Joseph
And and as a pediatrician, it’s the same. The same frustration that the mom has, Why can’t someone just tell me what’s going on with my kid over, and this is not over a month, this is over the years. And ultimately, though, she she took all the all the lab tests and all the symptoms and just kind of threw them in as a story and to chat GPT and asked the AI, you know, what’s a list of differential diagnoses? What are some of the things that could cause this, and it came up with a diagnosis called tethered cord syndrome, tethered cord, so the spinal cord getting kind of stuck on something so that it can’t move. And as the child grows, it generally gets worse. And one thing that, that she said was cute for hers that he couldn’t sit with his legs crossed. And
Nick van Terheyden
I have that, but it’s a completely different problem.
Craig Joseph
And you’re a little you’re a little bit older, Dr. Nick and this patient. And so, you know, he couldn’t sit that way. And that was one of the one of the giveaways. And so it’s suggested this AI with no specific training, suggested that tethered cord could be a diagnosis. And And finally, she got to a pediatric neurosurgeon who ordered the appropriate MRI and said, There it is.
Nick van Terheyden
So I you know, that’s good news. I mean, clearly for the mom and the son who are experiencing all of this chronic pain. And since I mean, I can only imagine and, you know, my limits of pediatric was my memory of the pediatrics was limited to essentially one specific thing that I recalled always, if mom says X, X equals true, always, no matter what they said, they always knew that there was something wrong and that was about the limits of it. I gotta be honest, I had to look up this tethered cold thing because that was a newbie, but it wasn’t in your practice.
Craig Joseph
Know It, you know, tethered cord is something that I saw a lot of patients with as a as a pediatric resident at a at a children’s hospital or a regional Children’s Hospital. So certainly I saw that, but I have never seen a patient with tethered cord in practice, you know, just kind of walking in walking in the in the door. So it’s it’s not that uncommon. However, it’s not the first thing that you would think of right and, and I think one of the things that we’re often taught in medical school is that common things happen commonly. And so the first thing we’re thinking about are
Nick van Terheyden
we’re on to zebras again, I can feel it. We are
Craig Joseph
we are and so this was this is a zebra but I think you know, the good news is that the the AI given all this kind of disparate information was able to pull this out.
Nick van Terheyden
Yeah, yeah. So I just want to know why am I saying zebra instead of zebra, which is the right way to say it. But anyway, that’s a different story. So I you know, that’s good news. Who’s but aren’t clearly as you pointed out, untrained. So this is an AI system that doesn’t have medical training, although when it took the MCAT, and I think even some of the was it step one, step two, I forgot on the latest version of it, it was doing extraordinarily well. So it’s clearly got some capacity, but it doesn’t have intelligence. You know, there’s some risk with this, that, you know, maybe it comes up generate something that is entirely incorrect. How do we sort of reconcile that? How do we get to the sort of level of tuning this so that it becomes an appropriate tool for everybody to use? What’s the sort of pathway for this, this is difficult to manage? With a, an uncontrolled unit?
Craig Joseph
Well, it’s, it’s, you know, to your point. This is a tool that has to be used correctly, right? And so you’re absolutely right, I would expect that often you throw these, these kind of pieces of information in there, and you’re gonna get garbage, just kind of garbage back out. And, and the reason I think it worked well, and the reason that this story got on the today’s show, was that this mom was an expert. She She was an expert, and what her kid didn’t have, right, she knew that it wasn’t this and it wasn’t that because people have been working those things out those common answers. And so when she saw something that she wasn’t sure what it was, in fact, I think it says in the story that when it mentioned tethered cord, she found a Facebook support group for parents of kids with tethered cord, and started asking around there. And those parents were like, yes, yes. And yes. And and so, you know, then you walk in with a very focused question to your pediatrician, I need I need a referral to a pediatric neurosurgeon, because I’m afraid that my child might have tethered cord syndrome. You know, if you walk in like that, and and, and after you’ve looked on Dr. Google to find out what are the signs and symptoms of tethered cord syndrome, and they, they, they mimic what your what your child has, you’re gonna get that referral very quickly, as as she
Nick van Terheyden
did. So good story, good outcome. I think, you know, some lessons in there, I think one of the ones that I took away, and it was something that I came across and you know, I’ve been messing around with this stuff, and Chad GPT for a while I know you have and I think this was the discovery for you. And I tossed it over for you to play with. But Chad GPT actually comes with an option that I think almost nobody goes and spends any time on. And it’s called Custom instructions. And what it allows you to do is to tell chat GPT a couple of things are essentially two boxes, one says What would you like me chat GPT to know about you to provide better responses? And then the other one is, how would you like chat GPT to respond. And the the better responses already allowed it to reference me and know, some of my background, so It customizes so I’m getting actually better content coming back at me that is sort of tune to my clinical background, and some of the things that I know and do and understand and also some of my interests. But then what I really liked, really, really liked was the customization of what came back. And this was where I really went to town with some of the recommendations, I said, treat me as an expert, you know, don’t start providing, you know, content that tells me about moral choices and so forth. You know, there’s all these safety protocols in place that are actually distorting the content that’s coming back at us. And, you know, I’m not removing that, because I don’t think this is a bypass, that’s not my purpose. And to be clear, I actually don’t think that’s a good idea. I know that it’s sort of preventing some of the terrible use cases that we see. And of course, those are the ones that people pick out, but, you know, who’s deciding what’s appropriate, what’s not, and, you know, there’s, there’s all sorts of but it just, I got this, you know, much better content coming back. That gives me, you know, references even it puts the references and the links to where it got the information if it has links. I don’t need to be told each and every time that this is an artificial intelligence, it’s not by the way, it’s just a word generator in my experience, but you know, that’s another thing.
Craig Joseph
Ouch.
Nick van Terheyden
Well, isn’t it
Craig Joseph
it is it is just a word generator, that that is accurate. At the at the at the bottom? Yeah. You know, I think this is all related to prompts, right. And so, essentially, as I understand it. And I had never heard of this before. You showed me an article and a couple of tweets about it. I think it just reads the it basically just runs these words that you’ve put in there. Before it answers every question, right? And so it’s essentially part of the prompt of, you know, hey, take listen to these, or read these words. And it’s not a lot, I think you get 1500 characters. Yeah,
Nick van Terheyden
it’s 1500 characters. It’s limited. Yeah.
Craig Joseph
And then it answers the question. And some of the things were pretty amazing. Like, hey, I’m into pop culture. So if you, if you find any pop culture references make those are also, if you get anything that seems like a strange Association, I’d love to hear about those things too. And so you’re able to kind of push it in a direction. And, and how it works, I don’t think anyone could tell you, but it does seem to, it does seem to work. And people are constantly going to be playing around with these and tweaking them. And then you and I could put the exact same prompt in and get and get ridiculously different responses.
Nick van Terheyden
And that’s a, you know, I think it’s a good thing, if you use it in the context of what it’s, I think designed or, you know, the, the general purpose, particularly with this general model that’s publicly available is, you know, go experiment, use it to sort of I use it to just sort of interact with and come up with ideas, it generates things that made me think and, you know, come up with some other things. I think the other thing that I saw at the same time was this whole take a breath. And that’s one of the instructions you can use to actually get it to come up with a different response. And that might sound really weird. But actually, the underlying reasoning is, you know, once again, it’s where it’s pulling all this information, it pulls it from the q&a forums, and if you’ve ever read those, and you see people getting really frustrated with technical challenges, or, you know, reasons for things not working, and, you know, one of the experts will jump in and say, let’s take a deep breath, you know, let’s go. And and it’s associating that with an actual answer and comes up with something that’s just a little bit more relevant. I think we could do that in healthcare as well.
Craig Joseph
Well, and this was specifically with I think it was a different AI. I think this was something that Google was putting out, but they literally added the words take a deep breath and work on this step by step and found that they ridiculously improved its ability, especially around math. So because it doesn’t really do math, like calculator does math. And so, you know, to go from, I think it was something like 30%, to that third is 34%. By adding those words, the accuracy went up to 72%. That’s, that’s amazing. That’s amazing. Right? Right, by just telling it hey, you know, work on this step by step. And somehow, magic happens, and no one really knows why. We, we have thoughts about it. And I was actually, as I was reading the article, I looked at some of the comments and the very first comment was prompt engineers a real job. Like, absolutely, and the next person responded. not to denigrate that work, but engineering makes it sound precise. It’s clear we barely understand the models right now. It’s more like prompt alchemy.
Nick van Terheyden
I like I want to be a prompt Alchemist. Definitely.
Craig Joseph
For those who aren’t prompt wizard,
Nick van Terheyden
I like it wizard.
Craig Joseph
It is there’s some there’s some magic to it. And, and and since we don’t really know why it works, we simply seem to find this association that makes it better. Alchemy seems like a much better, you know, prompt Alchemist or prompt magician or prompt wizard. Seems like a much better term.
Nick van Terheyden
Well, I was all about grabbing usernames. I got Obi Wan Kenobi a long time back before it was cool. Just to be clear when he wasn’t the young Obi Wan. Now I’m wishing I had Gandalf because I could be Gandalf the prompt generator or something. So
Craig Joseph
yeah, no, thank you for for telling us that. That wasn’t cool at some point. But now those
Nick van Terheyden
those characters are you needed to know that right?
Craig Joseph
I was not aware that they were made cool. But if you say so I’m not
Nick van Terheyden
sure. I just in my own mind, you know, I reject your reality. Substitute my own or that’s fair. Yeah. So we saw some other interesting news. I thought jobs to be done in healthcare by AI. You saw that right. There’s, you know, things that I think we’re going to see changes and I want to say adjustments, at least at a minimum thoughts.
Craig Joseph
Well, they, I think this article does a good job of kind of breaking these down into clinical parts and non clonal cohorts and then kind of going from consumer or patient facing to professional or, you know, clinician facing and, you know, right in the center are things that affect both parties, like, you know, patient intake, why am I? Why is a human necessarily involved in asking questions can the, you know, why are you here? Probably can collect that. But it gets much more complicated when you talk about clinical decision support. So, suggestions that physicians might want to follow. That’s, that’s a lot different, right? I think for things that just seem to be rote, and that there’s not a lot of clinical experience or clinical thinking involved. AI is generally going to be the way to go, hey, when can I get that? When are we going to schedule that next visit? How do I get that bill out to the to the insurance companies as quickly and accurately as possible? How can I not how can a physician not be involved in in pre authorization of, you know, getting an MRI?
Nick van Terheyden
Because I’ve already chicken of touching? You
Craig Joseph
know, I know, I know. But but you know AI is can probably handle the vast majority of those things as well or better than humans, either from the insurance company side or from the the doctor or hospital side. So why are we involved? Why can’t the AI just answer most of them. And then for the ones where it’s a little iffy, shoot that out and let a human actually deal with that.
Nick van Terheyden
I think the capitalist nature of healthcare at the moment is that both sides are trying to build AI tools that essentially are competing, it’s a bit like the chess competitions. And, you know, it’s a nuclear arms race as to who can develop the fastest and smartest as quickly as possible. And, you know, the better solution, as you suggest, is work together to find a reasonable approach to this that everybody goes, Yeah, that makes sense. But I think sometimes these are economic decisions or not actual clinical decisions, but opportunities for discussion at another time, I think,
Craig Joseph
absolutely. I just have one one quick story. And you may not know this that I actually, at one time, when I was practicing, I had a day off. And I spent some of that time as a as a advisor to our local our state Medicaid system. And as such, as a practicing physician, I was asked to, to authorize surgeries that needed to be authorized that the nurses couldn’t Authorize. So the nurses that worked for the same for this quality organization had, you know, some standards that they could look through, and most of the time 80 or 90% of the time they could approve the surgery. Well, if they couldn’t, it would come over to a doctor. And so I was asked to approve or disapprove of a of a patient to get a hysterectomy, which is awkward as a pediatrician. And I looked through all of the medical records and couldn’t find a good reason. So I called the physician. And this is where AI will fall down. I called the physician and I said, Can you explain to me why you want to do this surgery because I’m looking through all the records, and I don’t find any of the major criteria. And, and he said, Well, I don’t I don’t want to do the surgery. And there was a pregnant pause there. And I’m like, I’m sorry, I’m confused. We got a request from your office to have the surgery. And but you don’t want to have it isn’t like No, no, no, I don’t want to have it. But I can’t convince that the patient that she doesn’t need it. So I said I would do it. And if you want to if you want to say no, that would be great. And I Yeah, yeah.
Nick van Terheyden
So talk about abdicating responsibility. Holy smokes,
Craig Joseph
you know, an AI is not going to be able to handle that situation.
Nick van Terheyden
Just like can you guess which movies I was watching over the weekend? When I say Merlin’s beard.
Craig Joseph
It was not a James Bond.
Nick van Terheyden
No, it was not. There was Harry Potter. I was living through that dream again. All right, so we’re talking healthcare influences. I just a delightful piece of news to finally find some of these folks who are pushing out this junk information. And boy, there’s a big list of folks that could potentially go on this but actually got prosecuted for pushing out just unfounded scientific, and I didn’t I hadn’t even heard about this the same as I hadn’t heard Have you been asked whether you think about the Roman Empire every day? Because I got asked that just recently and I was
Craig Joseph
well, I a lot of men say yes to that question. I’m not what
Nick van Terheyden
JCP de superyacht. Oh, go. Yes. Go look that up.
Craig Joseph
Did you just swear at me in Latin?
Nick van Terheyden
Well, I didn’t. My last it wasn’t ever that Last, but that no so this this guy is recommending a raw meat based diet I add, dear lord i Even if that’s your thing. Anyway, he’s and he was really if you saw the pictures he was totally bulked. It’s the best part of it. He’s he got there with I can’t even believe this bill $11,000 of steroids per month. How do you pay that much for steroids? Expensive? Did I miss something?
Craig Joseph
Did he forget to mention that on his influencing it was just he forgot that part. Yeah. I just I’m shocked. In general, you should not choose your medical professionals, based on a follower count. For you. Yes. For you. Yes. So me? Yes. But for most people don’t go just by that there’s probably other criteria you want to look at that are far more predictive of a good answer.
Nick van Terheyden
shocking to think that that’s actually a thing. But then when you see some of these people that were promoting, just go and get your How much was it? Was it a $4,000 MRI every quarter or something as one particular several a celebrity was promoting that? I wonder if they’ll go and prosecute there as well? What do you think?
Craig Joseph
You know, takes time takes time to catch up to people. So I don’t know,
Nick van Terheyden
Boy, I sure hope so. So, alright, so a little bit of time left. And we’ve got a couple of things left, the blame game on drug pricing. Essentially, everybody’s pointing fingers in the high cost of drugs, massive differences, and we come out on top.
Craig Joseph
Yeah, well, I know what comes out on top. But I know who comes out on the bottom that much I can answer. It’s the patient. This one, this one study looked at the exact same medicine from for the exact same kind of patient, you know, with with the exact same kind of insurance actually. And on one day that that same medicine, one person would get charged $9, and another person would get charged $96. And your your insurance card will look identical to that person’s and one of you was paying nine and one of you was paying 96. And there’s there doesn’t seem to be any rhyme or reason. And I think the article was looking for who’s to blame? And the answer was all of us that put up with a system like this. Because there’s no there’s no logic. So it kind of throws a whole monkey wrench into the idea of that. Patients will want to save money if you make healthcare transparent, or at least healthcare data transparency a priority, because then I can go and oh, I know if I get this medicine at this place. It’s going to cost me this much. Nope, you don’t know that.
Nick van Terheyden
It’s still still, it’s still shocked me It shouldn’t but it does. Alright, we’ve just got time to squeeze in one last thing and of course favorite topic of mine longevity. And, you know, what can you do? There’s lots of noise, certainly in the space talking about Metformin, which is drug use for diabetes. It’s based on goat’s Rue, I think that was the original source of it. It’s been used for hundreds of years at this point in various forms. And one of the most interesting intriguing facts was that you look at diabetics and their life expectancy and cause of disease. And if they were on Metformin for diabetes type two, we saw improved outcomes. And the epic research which is a favorite of yours showed or confirmed that that was the case. Now, correlation causation to be fair, no recommendations there is clearly lots of research to do the problem is doing research is very difficult. Because there’s no money in it because it’s a very cheap drug. So we’re going to start taking Metformin now.
Craig Joseph
I’m not going to start taking it yet. However, it does seem you know, based on these data from from Epic, a big Electronic Health Record system that that if you’re on Metformin for diabetes type two, it’s it’s likely to help your your your sugars, but it’s also likely to help other things like Alzheimer’s and other kinds of cancer and heart disease. I don’t I am worried about the the correlation and causation question. So is it just that Wow, your your your glucose is much better controlled. So now all these other things that are related directly or directly have also been improved. I don’t know it probably doesn’t matter what the cause is, if it continues to be associated, it’s a good thing. And Metformin seems to be a med that I would jump on if I were if I were diagnosed with diabetes type two,
Nick van Terheyden
I think fair comment, but we find ourselves at the end of another episode, exploring Healthcare’s mysteries before they become your emergencies. Until next time, I’m Dr. Nick.
Craig Joseph
And I’m Dr. Craig
