This month’s episode of “News you can Use” on HealthcareNOWRadio features news from the month of February 2023
As I did last month I am talking to Craig Joseph, MD (@CraigJoseph) Chief Medical Officer at Nordic Consulting Partners.
What Can ChatGPT Do for Healthcare
With the continued media coverage and focus on ChatGPT we took a deep dive into the technology, the latest news and use cases being explored, and where this might bring value in healthcare.
Chat GPT (Generative Pre-trained Transformer) is using technology that has been in use in Speech Recognition for many years. Currently on GPT v3 (Large Language model of 175 Billion parameters). It is a generative AI technology that has been used mainly by high school and college students to write essays. As Craig details the technology is increasingly being explored in healthcare, where it can be used to translate doctor talk into layman’s terms and create patient-specific responses to lab or imaging results. The technology is not yet mainstream, and doctors need to validate its output. Nonetheless, it has the potential to automate the creation of messages that go back and forth between patients and doctors, and it could be incorporated into the electronic health record in the future.
Essentially ChatGPT is a word predictor machine learning tool that has made natural language processing more accessible and acceptable. Chat GPT predicts the most likely next piece of data to send out by pulling from data sets, but errors have occurred as the machine can make things up. We are both excited but cautious, as there has been some overextension of this technology. Despite some concerns, chat GPT has opened doors for accessibility and created many opportunities for exploration.
Listen in to hear about the rise of the “Prompt Engineer” – the salary expectations will surprise you.
We discuss Mastodon the new social media app that is seen as a competitor to Twitter but operates as a distributed network with no central authority. Craig shares details of a new app called Ivory, that allows for direct posting to Mastodon
We review the paper on Direct to Consumer Drug Advertising: Association Between Drug Characteristics and Manufacturer Spending on Direct-to-Consumer Advertising, lamenting that the US and New Zealand remain the only 2 countries allowing this activity
And listen in to hear our alternative views on hot or cold tea or coffee and if you really are borrowing from your sleep bank when you drink these caffeinated beverages.
You can read more about the series here and the concept of keeping up with innovation in healthcare. Please send me your suggestions on topics you’d like to see covered. You can reach out direct via the contact form on my website, send me a message on LinkedIn or on my Facebook page (DrNickvT), or on Twitter by tagging me (@DrNic1) and #TheIncrementalist or you can click this link to generate a ready-made tweet to fill in:
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
And today as I am each and every month, I’m delighted to be joined by Dr. Craig Joseph. He’s the Chief Medical Officer of Nordic consulting. Craig, thanks for joining me.
Craig Joseph
I just like with you, it is a pleasure and something I look forward to every month.
Nick van Terheyden
I doubt that completely. But okay, I’ll believe it just this once.
Craig Joseph
Realistically, it’s the $100,000 a month that you pay me, but we’ll just go with, it’s an honor and privilege.
Nick van Terheyden
I only wish that were true, because that would make me a rich man, too. So. So we’re looking at news from February of this year, and it would be impossible to consider the world without chat GPT In fact, I think chat GPT has taken over everything is producing all content. And if you haven’t heard of it, you must have been hiding underneath a mushroom somewhere.
Craig Joseph
I think I previously said that I was actually chat GPU did. But and in in the month or two since I said that it’s I’d become more of Chet GPT The thing is really cool. And and this generative AI, as it’s called, I don’t understand how it works. But I do. And we are seeing more and more actual real use cases for it. Besides the high school or college student who has an essay do certainly it’s it’s pretty good for them. Although a significant chunk of what it says is just garbage. So that’s a problem. Because I think the teachers and professors might might see through that. But there have been some exciting uses, or use cases probably is better, better term for Chet GPT, and healthcare that people are exploring. None of these are mainstream, in my opinion, yet, but they seem exceedingly promising. Even today. So let me just give you one example that kind of shocked me taking lab results are very specific kinds of imaging or results. And, and having chat and GPT kind of translated from Doctor talk into regular person talk. And it’s, it did a really good job. So essentially, you give it up a prompt that the input that you give, and say, you know, create a message for a patient who has a potassium of x with congestive heart failure, you know, and then recommend that they continue their dose of diuretic or whatever it is that they whatever it is that you want to say. And it does a really good job of of writing it. And
Nick van Terheyden
it’s probably I this is this is really interesting, but I think it’s important to clarify, and I want to make sure that this is true, which is why I’m interrupting you, did you validate the content, because that’s one of the things that we see repeatedly. So this was, when you say it was really, really good, it was really good, really good, clinically, it was accurate. And
Craig Joseph
it was at Yes. Now this is a pretty simplistic example that I gave you. And this was the kind of of response that I saw. And this wasn’t I didn’t come up with this, let me be clear that it I’ve just been kind of poking around the interwebs. And seeing what people are doing. This was one of them. So basically, the use cases, pouring or putting significant lab results or imaging results in Doctor talk and then having it give it give you a return to a patient specific or a more kind of patient centric version of that, of course, 100%, the doctor needs to validate it. And that’s why I was saying like, Hey, these are just people playing around at this point. However, one could imagine within a few years, this being an automatic tool that’s incorporated into the electronic health record that doctors and nurses are using, where whereby it can kind of automate a lot of this. A lot of the messages that go back and forth. I don’t know how long until we’re comfortable with doing it without validation, if ever, I’m not sure that I would ever be but everything’s changing so quickly. I wouldn’t I would never say never at this point. But it’s a it’s a use case that I never really thought of before. Or it’s seems seems pretty good at also giving patient advice. So again, if a patient rights to their physician via the patient portal and says, you know, hey, I’ve got I’m having side effects from this medicine and I’m not sure if I should stay on it. Typically li the patient, the physician is going to have to write up a pretty nice response to that. Because that’s not a yes or no question that that’s going to take, you know some time. And to be able to put into this, this generative AI and say, Hey, write a response to a patient is concerned about a side effect from this medicine, make sure they realize that research shows that there’s no need for them to change. And this will probably go away in one to two weeks and tell them about the warning signs to call me back if it gets worse, right. And what I just said was probably much blurrier than it needed to be. And then the thing spits out like three, three paragraphs that are typically data. And because again, none of this is really rocket science. And for, you know, these are not complicated questions that were posing. I just thought, wow, that’s really great. And that’s that could be used today. Today, if the physician does a good job of reviewing it very carefully before it goes out. And we’re talking about two or three paragraphs. So I think it’s, it’s a great, those are some things that I had never, never seen coming like I didn’t, I didn’t think of that, if you would have told me that four or five months ago, that there’ll be a tool that could do this pretty well today. I would, I would have not believed you. And yet, there it is. And, and this is GPT, three, again, just one vendor that’s doing this right now. And we know that Google and others are out there, lurking and potentially being even further ahead, but not going public yet. And my understanding is chatting GP, or GPT for is, is likely to come out by the end of the year, and will be even more like exponentially better. And I’m already kind of stupefied by how cool this is at this point. So just some of the uses in healthcare specifically I I’d never anticipated and think there’s more common everyday people are trying to use it. Again, you can you can run into trouble when you think seemed physicians use it to generate appeal letters to insurance companies. And new specific and they’ve specifically said cite the literature. And it cites made up articles. So that’s not cool.
Nick van Terheyden
Yeah, so I think you hit for me, the key point, and that is that there are so many other more opportunities to be explored. And what what I think chat GPT has done has opened the door in the same way that we saw that with some other technology made it more accessible, more acceptable in many instances, and, you know, has essentially allowed people to go, wow, this is really cool and interesting. But it comes with a bunch of caveats. And you know, that was why I interrupted because I’ve seen a number of instances and you even highlighted it where, you know, gets it wrong. Not even not even just wrong, but actually make stuff up which let’s be clear, people do that as well.
Craig Joseph
Most of what I say Nick is made up, I think
Nick van Terheyden
our chat GPT. So we expect that, but it’s
Craig Joseph
okay, if I make it up. It’s not okay, if a machine mix it up. I think that’s what you’re saying.
Nick van Terheyden
Right? Exactly. Machines? Absolutely not so. So let’s just for a second for the benefit of people to understand. So you know, chat GPT. So, you know, what is it well, generative pre trained transformer is what GPT stands for which put it another way as a word predictor. That’s essentially what it does. And, you know, you talked about GPT, three to four, what was the difference between one two and three? Well, it was a number of parameters or data. And one, I think had about 100 or so million. Two was up in the 1.5 billion, so consuming much more, and three was up to 175. And then ostensibly people here that go, alright, four is going to be out there. And yeah, they’re going to add a bit more, but actually, more is not always last, we’ve seen others trying to think I want to say it was IBM, I forget which one had somewhere of the order of 530 billion parameters, and it wasn’t any more anyway, better. It’s more about fine tuning. But that word predictor is important in all of this because essentially what it’s doing is pulling from all that data and saying what do I think is the most likely next piece of data to send out and, you know, I can’t be certain of this but you know, the errors the made up, I’m willing to bet has appeared somewhere in its dataset. That’s why it’s got that I don’t know if that’s true or not. deeply, sort of invested in that. But I’m with you, I’m excited. But I’m also a little bit cautious because I’ve seen you know, some of this overextension. I’ve used it, I’ll say non healthcare, I was at a, an event, it was for a bit of fun, actually. But you know, how they always say, All right, everybody introduce themselves. And you’re sort of what do I say? So I asked chat, GPT, to give me an intro, it was actually pretty cool. And I did declare it and you know, this, this intro was, and it was, it was funny, but it was actually a pretty good use case. And it had data about me, which was also kind of interesting. But it’s, it’s still limited. However, here’s the one thing that I feel is important to sort of emphasize and all of this, and that is, you know, there’s all this concern, you know, I use, and I follow two different people, one who says, I’m actually declaring that I never use chat GPT or human generated content and somebody else that actually has a, you know, generated with the help of, and I actually think both extremes Well, actually, the other is not an extreme. But the point that I think that was made to me that I think is entirely irrelevant here. Nobody puts checked by spellcheck before they send out stuff, or in my case, I use Grammarly all the time, thank God because my typing is awful. I’m pretty sure that my grammar, I don’t know if my spelling is any more because I write nothing. And my typing is so bad, it always produces lots of errors. But I don’t put a little subtext that says, hey, check by, but that’s essentially doing similar kinds of things. So I think we’re gonna move past that. And I’m incredibly excited by this, mainly because it’s opened the door in a way that has made this accessible. If you aren’t trying it, you’re missing out, you’ve you’re just missing out on the fun as much as anything else. I’ve had it do all sorts of crazy things. And it’s been quite entertaining. In fact, I’m even working on a presentation and I used it to sort of come up with an outline the whole thing, and it’s pretty cool at that. So I’m excited about it. But I think we have to sort of temper it. But you know, the enthusiasm shouldn’t be killed, along with Oh, my God, it made this terrible error. We should, you know, I saw somebody saying you should control it. And you know, limit AI use? Well, no, I mean, let’s not have it decide on medication therapy, perhaps. But open this up. This is fantastic.
Craig Joseph
Yeah, I agree. I have to say like, the last time I was excited about a technology it was it was Google. I mean, to me in terms of kind of transforming, you know, Google changed for, for I think all of us. And I guess I shouldn’t say Google, I should say search, but let’s be honest, it’s Google. You know, we don’t argue very much anymore about facts. Well, that’s funny that I just said that. Things are knowable. Okay, what was that TV show back in the 70s? Like, hey, who was you know, when did that test come out? Like, these are things that we now can find out because of search in the in the internet. And this seems like the next the next frontier, again, emphasize what I said earlier, this is one company. And and, and they’re not even charging money for this yet. I think they’re pretty close. I see now that they you can pay them $20 A month and get expedited service and not get kicked out. There’s been a couple of times where I’ve tried to go in and play around with it. And it’s like, Hey, we’re too busy. Call us later. So so for a little bit of money, you can you can bypass that. But I am with you. And I again, I think as long as we’re we’re all on the same page that we probably shouldn’t be deciding what surgery you should have. Because chat GPT we put your symptoms in there and Jeff GPT came out with some suggestions. Although might be a good starting point. I don’t know. But certainly, it should be not the ending point, right.
Nick van Terheyden
I wonder what would have happened if if Gregory House had had access to chat GBT during that TV series.
Craig Joseph
And one thing just I know we keep kind of beaten this, this topic, but it seems now that a lot of it is the prompt, right? I remember the first the weekend this this came out that I released the first week and I heard about it. Someone wrote something about a new job called a prompt engineer, a prompt engineer. They said that’s going to be the new thing. And that people will be paid because they have expertise in knowing how to tame the AI So they get what you want out, they can put the words in to get what you want out. And I kind of rolled my eyes at that. And then a couple things happen. One this weekend, I read in the newspaper that there is now an advertisement from some company for a prompt engineer making, not making this up $250,000 a year. And so I thought, that’s pretty cool for a job that I don’t think existed six months ago. And secondly, I was telling another physician colleague of mine about some, there was a blog post that I was thinking of writing. And again, as we explore this, I was like, Hey, let me see if I can get the chat GPT to write this blog post for me, and I put in the idea, and it came out with just garbage. It was just horrible, because it didn’t see what I was trying to connect. And I mentioned this to a colleague of mine, and he’s like, Yeah, you didn’t put it in the right prompt. And he seems to have some kind of natural born ability. Because he showed me some some of the things that he’s he’s come up with. And so I do think there’s that as well. So it’s, it’s, you know, putting in the the correct prompt, or at least putting in the information that the AI needs to direct it to where you want to go is not is not obvious. So there are some of us that have this ability. I’m not one of them, apparently. But I do think it’s, it’s interesting, and, and again, just kind of this idea that I could, you know, take, I could take an app note, an operative note and just say, Hey, make this into one paragraph with seven sentences. Explain to the patient what we just did. Boy, wow.
Nick van Terheyden
So for those of you just joining on Dr. Nick the incrementalist today, I’m talking to Dr. Craig Joseph, he’s the Chief Medical Officer at Nordic consulting where news you can use for February and we were just talking about chat GPT. As Craig described, the prompt, engineer, I would say the prompt whisper or let’s say the chat GPT whisper is the new. I’m the chief Japp chat GPT whisper. And I’m actually going to say that I was pretty good at some of this stuff. Certainly my results. And there are some super experts, as we saw with the launch of Microsoft’s version of it, which which took a nosedive a lot. And I’m pretty sure I think it was yo or ye or whatever. It was cool that that was theirs as well. So it’s not quite as easy as people make out in all of this. So anyway, great conversation. Let’s move on to a couple of other things whilst we still got time. Mastodon still a thing. But Is anybody getting anything out? There? Turns out there’s some new ways to do so. Right?
Craig Joseph
Yeah. So Mastodon is another social media app. Really kind of showing up as a pseudo competitor of Twitter. I say sudo. Because it’s not a company. It’s a it’s a network, a distributed network. And don’t ask me to tell you anything more, because I don’t really understand. But basically, as I understand it, no one’s in charge of mastodon. And so there’s, there’s no risk of a company selling advertising or doing anything else. Yeah. So there’s a new app called ivory, get it mastered on ivory. That it’s out in beta. I know. It’s pretty cool. And so I’m starting to explore that. But you know, Twitter is still there. And LinkedIn is still there. And Facebook is still there. Did they?
Nick van Terheyden
Did they consult the Craig Joseph marketing company for that naming?
Craig Joseph
If they had, I wouldn’t need the $100,000 that you’re paying me to appear on this podcast. But no, they, they sadly did not. And also, sadly, I’ve never gotten any of the money you promised me. So I have to keep coming to work every
Nick van Terheyden
day. GPT where I got my money’s worth date. Yeah,
Craig Joseph
I think so. I think that’s a much better, much better use. So yeah, Mastodon, it’s interesting. I’m not sure where it’s gonna go. You know, Twitter is Twitter’s a whole thing. We’ll see. I think the jury’s still out on Twitter. But Mastodon seems to be making a surprising jump. And I think what’s interesting is actually it’s different groups like there’s no there’s something we still I think we still call med Twitter. You know, and those are healthcare professionals that that are active on Twitter and ask questions and interact and it is another one where there’s a lot of technology folks that are used to be on Twitter are moving and mass to Mastodon as I understand it. So it’ll be interesting to see if these certain sub populations kind of bail and go from one place to another place. Mostly it just forces us to to post our things on multiple social media, we want to, we want to interact with everyone. That’s
Nick van Terheyden
been my experience. I gotta say, but I’m there, but I’m not really there. It’s, I think they, somebody had a term post in Danville. So anyway, it’s it’s, I’m, I’ve done a little bit, but it’s really sort of it’s out there. So I got to ask you, I’m going to, we’ve talked about this I know before, but direct to consumer advertising of drugs occurs in two countries. One is the United States, the other is New Zealand still don’t understand why New Zealand is there, but the US huge market, really interesting athlet article in JAMA that looked at the association of the drug characteristics and the amount of money that was spent on it. And worryingly, not surprisingly, in my view, the drug companies spend more money advertising if the drug showed less clinical benefit. Why, Craig, Joseph, does the United States still allow direct to consumer advertising of drugs?
Craig Joseph
Well, it’s a it’s a great question. I don’t think it’s a steal. You know, the question I would have in my mind is why did we start in the first place?
Nick van Terheyden
That’s a good question, but I don’t know the answer to it.
Craig Joseph
Yeah. Back in the day, admitted, you know, pharmacies, pharmacies, drug companies could sell or advertise prescription medications only in journals or other tools that were focused on physicians. Because they’re the ones that control it’s a prescription medication that we’re talking about. And I have to say, I’m not aware of why that change, but boy, it did. And it’s, I think everyone is that’s not under a rock knows, because you can’t turn on a television, or watch the show without kind of like, Hey, do you have these symptoms? Or do you have this disease? So my question is not why, why are we still doing this? Why did we start in the first place? I the only rationale could be that there are patients out there who have diseases or symptoms for which there are medications that might help them prescription medications, and they were not aware. So hence, they did not ask their physician. I am sure that that’s the rationale that was explained to our representatives and senators who made this happen. I suspect there’s not data to support that and and this articles and is another piece in the puzzle to say that that’s probably not the case. But I’m not sure that that genies out of the bottle, I’m not sure how you’re gonna get that genie back in.
Nick van Terheyden
Yeah, interesting. Interesting. You say piece in the puzzle, not nail in the coffin, which is what I’d like it to be, but I’m pretty sure you’re correct. It’s, there ain’t no changing that that particular process, unfortunately, it’s it, I gotta say, it’s really troubling. I’m all over patient awareness. And, you know, making sure that people understand. But the context of all of this, especially with these fractional improvements that you see with some drugs, where, you know, they change just a little bit, and then, you know, suddenly you got to get the latest and greatest, it’s off pattern. It’s very troubling. So we’ve got a little bit of time left. You and I think are both big coffee drinkers in some version. In the remaining time. I saw a piece that talked about coffee, essentially borrowing from your sleep bank, and it won’t give you the energy, you essentially you have to pay it back at some point. I don’t buy that I drink gallons of the stuff albeit up until a certain point in the day, and then I just I stopped wanting it quite frankly, but I really do drink a lot of coffee. Do you ever have trouble with your caffeine addiction? I’m saying you’re an addict. I don’t know if that’s true, but I’m an addict, and I’m proud of it of caffeine anyway.
Craig Joseph
I mean, Nick, I haven’t slept since 1990. So I don’t I don’t really know what you’re talking about. I do drink coffee, generally of the iced variety and like you I don’t seem to notice sleeping differences between when I do drink coffee and when I don’t drink coffee. That’s not to say that someone with a all kinds of wires connected to my head I might notice but they might notice but I don’t seem to notice one but that’s why we do studies and that’s why we do research and and and of course like you anything that does that I disagree with. Based on my personal experience. I just dismiss like you just did so because you don’t like it. So then you’re just gonna say it’s not real and that’s fine, Nick. That’s not very scientists. I
Nick van Terheyden
actually I wasn’t going to say that it’s not real I’m just going to say what is wrong with you Americans and ice in these are ice for crying out loud there is nothing right about that. And there’s nothing right about ice coffee either. Unfortunately crate there is no time for you to give any funny reports. We have run out of time as we have each and every week. Thank you for joining me Drake.
Craig Joseph
Thank you, Nick and this is the last episode that will ever occur because of you saying bad things about iced coffee.
