Latest / Code & Cure / #46 - We Expect Patients To Learn Fast When They Feel Worst

Transcript

#46 - We Expect Patients To Learn Fast When They Feel Worst · Code & Cure

0:00
SPEAKER_02: Nobody's at their cognitive best after surgery or in the ER or after a new diagnosis, but that's exactly when we expect patients to absorb the most information.
0:21
SPEAKER_00: Hello and welcome to Coding Cure, where we discuss decoding health in the age of AI.
0:26
SPEAKER_00: My name is Vasant Sarathi.
0:27
SPEAKER_00: I'm a AI researcher and cognitive scientist.
0:30
SPEAKER_00: I'm here with Laura Hagopian.
0:32
SPEAKER_02: I'm an emergency medicine physician.
0:34
SPEAKER_02: And I I am completely guilty of discharging someone after giving them some verbal instructions with like a 15-page document of like here are all the things you need to remember when you go home.
0:43
SPEAKER_02: Right.
0:43
SPEAKER_02: Here are your medication changes.
0:45
SPEAKER_02: Here's 12 pages to read about your diagnosis.
0:47
SPEAKER_02: Have you ever received one of those things, Vassam?
0:50
SPEAKER_00: Absolutely.
0:50
SPEAKER_00: And and it's not just in those situations.
0:52
SPEAKER_00: I think when I get a flu shot, they give you a pamphlet to read over.
0:56
SPEAKER_00: When you pick up meds, they give you a pamphlet to read over.
0:59
SPEAKER_02: And have you ever actually read them?
1:01
SPEAKER_00: No.
1:01
SPEAKER_02: Just curious.
1:02
SPEAKER_00: Not really.
1:03
SPEAKER_00: I mean, I read the first paragraph and then I put it away with the hope that I will read them in the future, but I don't.
1:07
SPEAKER_02: Yeah.
1:08
SPEAKER_02: And I I think your experience is not uncommon.
1:11
SPEAKER_00: Yeah.
1:12
SPEAKER_02: And it's not uncommon for people to get these documents, intend to read them or want to read them, but put them away.
1:20
SPEAKER_02: And uh in fact, a lot of them are written to very high reading levels that we wouldn't expect everyone to necessarily grasp or understand.
1:29
SPEAKER_02: And of course, you can't ask questions of them.
1:31
SPEAKER_02: And so there are a lot of problems with the way that we provide information to patients.
1:36
SPEAKER_02: And I think this could be a really cool use case for AI to step in along with human oversight and add an extra layer here.
1:44
SPEAKER_02: I agree.
1:45
SPEAKER_00: Yes.
1:46
SPEAKER_02: Okay.
1:46
SPEAKER_02: So I think there are a few things that help create this issue of difficulty with patient education.
1:52
SPEAKER_02: One of them is that like you just don't have time.
1:55
SPEAKER_02: Like if I'm seeing 12 people in the ER at the same time, I may not have 10 minutes to spend counseling about a discharge diagnosis and Yeah, you have to see the next patient and you have lots of people lined up and you have other things going on.
2:08
SPEAKER_02: And that's true in primary care offices too.
2:10
SPEAKER_02: They get like 15 minutes to do like a full checkup on someone and counsel them and give them full discharge instruction.
2:15
SPEAKER_00: It's just like But how much are you trained also to do that, to do that effectively and do that fast and do that and communicate that?
2:21
SPEAKER_02: Well, I think everyone wishes they had more time to do that.
2:24
SPEAKER_02: Yeah.
2:24
SPEAKER_02: Right.
2:24
SPEAKER_02: It's like you really want that patient interaction, but you're also like, oh, I gotta move on.
2:29
SPEAKER_02: There's somebody waiting.
2:30
SPEAKER_02: There's seven people waiting, right?
2:32
SPEAKER_02: And so there's there's just like a lack of time.
2:35
SPEAKER_02: And another thing I think that's that we kind of touched on already, which is like, hey, um, as as providers, we use a lot of terminology, right?
2:46
SPEAKER_02: And so jargon.
2:47
SPEAKER_02: Jargon.
2:47
SPEAKER_02: Yeah.
2:48
SPEAKER_02: And it's not like we necessarily try to, but I think sometimes it's hard to say things in simple or lay terms.
2:54
SPEAKER_02: Um, and that's another place where AI could step in.
2:58
SPEAKER_02: Um, the other thing is that I don't know about you, but when I'm learning something new, I could I could hear it more than once.
3:07
SPEAKER_02: I'd like to hear it like multiple times for it to like really sink in.
3:11
SPEAKER_00: Yes, yes.
3:12
SPEAKER_02: And there's definitely no opportunity for that.
3:14
SPEAKER_00: Right, right.
3:16
SPEAKER_02: Um, and then maybe some of the questions that I might have might be clinical about, hey, what is this diagnosis or what what is this medication?
3:25
SPEAKER_02: And some of them might be more sort of like operational, like, oh, when, you know, can you book my next follow-up or can you help me with transportation here or whatever it is?
3:34
SPEAKER_02: And those are handled by different people in in a clinical location.
3:39
SPEAKER_00: Honestly, sometimes questions pop up later for me.
3:41
SPEAKER_00: Uh, it's, you know, I in the moment I don't have any questions.
3:43
SPEAKER_00: You always ask me, do you have any questions?
3:45
SPEAKER_00: And I don't have anything because I haven't had a chance to process it.
3:47
SPEAKER_02: Yeah, and I think that's really common.
3:49
SPEAKER_02: And it's not so easy to get a hold of people, right?
3:52
SPEAKER_02: A day or a week later.
3:53
SPEAKER_02: You can certainly send messages in the patient portal, you'll get a response.
3:57
SPEAKER_02: But I think that's a really common thing where people, you know, they have all this information, they go home, they have time to process and think about it, and they're like, oh, actually, I have seven questions that they didn't have in the moment.
4:10
SPEAKER_02: Um, and then there's this other piece that I think is sort of like infinite for AI.
4:16
SPEAKER_02: Like AI has infinite time to respond to you.
4:18
SPEAKER_02: You can ask it 17 million questions, but it also has this other component of it that it doesn't run out of, which is empathy, right?
4:25
SPEAKER_02: It's not really empathy, it's like Yeah, but it's a display of empathy.
4:29
SPEAKER_00: A display of empathy.
4:30
SPEAKER_00: That's enough for a lot of people to feel like they're being heard and they're they're they feel you know, there's a certain feeling that words have on you.
4:37
SPEAKER_00: So it no matter whether those are intentional or not, um, and and so that's maybe enough, and is what we're saying.
4:44
SPEAKER_00: Maybe I don't know.
4:44
SPEAKER_02: I don't know.
4:45
SPEAKER_02: I don't know if it's enough.
4:46
SPEAKER_02: But I think the current way that people like receive information about, you know, um uh patient education information about a new diagnosis or treatment or what to expect, it's like you know, there are limitations right now, and they're not surprising given that there's like limited time.
5:03
SPEAKER_02: Um, and oh sorry, one more thing I forgot to mention is is a language.
5:07
SPEAKER_02: So you know, if you have a uh if you speak a different language, sometimes discharge instructions would be available in that language and sometimes they would not be.
5:16
SPEAKER_02: And you're relying on like a translator uh and verbal instructions in that case.
5:21
SPEAKER_02: And so that could lead to a lot of confusion too, right?
5:24
SPEAKER_00: Right, right, right, right.
5:25
SPEAKER_00: So, so so I mean, in sum, it seems like what's happening is there's a whole bunch of information that either follows a visit or goes along with uh a drug that you've been prescribed or whatever it is, right?
5:38
SPEAKER_00: The flu shot you've been given, like a good example, right?
5:40
SPEAKER_02: It's like what do you actually need to know?
5:42
SPEAKER_00: Yes, there's so many, you know, so much information.
5:44
SPEAKER_00: And they try to make it as um compact and engaging as possible when they you know print these out, but it doesn't have the answers to all the questions, and also it is not personalized for the patient.
5:56
SPEAKER_00: And so as a result, you get, you know, potentially something that is you just throw out.
6:03
SPEAKER_02: Yeah.
6:04
SPEAKER_02: All right.
6:04
SPEAKER_02: So I think it's helpful to take a use case, which is what we did here.
6:07
SPEAKER_02: We took a use case of where they were like, hey, could this could this work?
6:12
SPEAKER_02: Could AI help supplement for patient education?
6:16
SPEAKER_02: And um, and in this pay, this paper was small, it was only like a handful of patients.
6:20
SPEAKER_02: They actually weren't patients, they were just like regular people that they recruited by the city.
6:24
SPEAKER_00: Yeah, to and and pretend to play patients.
6:27
SPEAKER_00: Like role playing patients.
6:27
SPEAKER_00: Yeah, role play patients, yeah.
6:29
SPEAKER_02: And they also asked the providers who could have been involved in care like this, like, oh, how did the AI do?
6:35
SPEAKER_02: Um, but they specifically decided to focus in on uh macular degeneration.
6:42
SPEAKER_02: What's that?
6:43
SPEAKER_02: And uh intravitrial therapy.
6:46
SPEAKER_02: What's that?
6:46
SPEAKER_02: Do you like these words?
6:48
SPEAKER_02: So age-related macular generation degeneration um affects the macula, which is like the central part of the retina in the back of the eye.
6:58
SPEAKER_02: Okay, and so it can cause blindness.
7:00
SPEAKER_02: Uh so the macula is responsible for like all of your sharp vision, um, reading, driving, all of those things.
7:08
SPEAKER_02: And wet age-related macular degeneration is actually what they were looking at, which is when you have like abnormal blood vessels that kind of grow there and they leak out.
7:18
SPEAKER_02: And so basically, this is a common cause of blindness.
7:22
SPEAKER_00: Okay.
7:22
SPEAKER_02: And so they decided to look at that because there are great therapies available, including intravitreal therapy, which I'll get into a second.
7:31
SPEAKER_02: But but adherence is really is actually pretty low.
7:35
SPEAKER_02: It's like 30 to 60 percent at one year, and so this is an area where it's like, well, hey, ultimately, uh you would think the goal is let's get people to adhere to this therapy more so that they don't they don't go blind, right?
7:49
SPEAKER_02: Right.
7:50
SPEAKER_02: And so the therapy, um intravitrial therapy, is when they like literally inject meds into the vitreous of your eye, which is a um like a gel-like substance in your eye.
8:02
SPEAKER_02: And what it does is it blocks those blood vessels from growing and leaking.
8:07
SPEAKER_00: Wow.
8:07
SPEAKER_00: Okay.
8:08
SPEAKER_02: And in order to do that, I know it sounds does that sound, does that give you like the heebie jeebies when I say that?
8:13
SPEAKER_00: I don't like I don't like anything with the eye.
8:15
SPEAKER_02: This is what I was like, needle in the eye.
8:17
SPEAKER_02: Yeah, exactly.
8:18
SPEAKER_02: These are like my these are actually some of my least favorite ER complaints too.
8:21
SPEAKER_02: Like someone comes in and they have a fish hook in their eye.
8:24
SPEAKER_02: I'm like, oh man, like that.
8:26
SPEAKER_02: That and like a cockroach in the ear or something.
8:28
SPEAKER_02: I just they just give me like the little heebie jeebies.
8:31
SPEAKER_00: Oh man.
8:31
SPEAKER_02: Um, but in this case, like the the therapy is pretty simple and they numb your eye before you do it.
8:38
SPEAKER_02: So they numb the eye, they inject your med, and and you get that med like often monthly at first, and then it spaces out to every two to four months.
8:45
SPEAKER_02: So you might have like seven or eight injections over the course of the first year.
8:50
SPEAKER_02: And and it continues chronically, and it really does work.
8:53
SPEAKER_02: In fact, it helps uh almost all people maintain their current vision, and some people even can see their vision improve.
9:01
SPEAKER_02: And if they don't get this treatment, they would lose their, they could lose their vision, right?
9:06
SPEAKER_01: Right.
9:06
SPEAKER_02: Right.
9:06
SPEAKER_02: So it's like, oh, that's vision's pretty important.
9:10
SPEAKER_02: There's a treatment available, it works for almost everybody to maintain their vision.
9:13
SPEAKER_02: And so it's something you would want people to do, right?
9:16
SPEAKER_01: Yes.
9:18
SPEAKER_02: Um, and so this was a case where they were like, hey, what if we mock up a scenario where we have a chat bot talk to a patient about age-related macular degeneration and what the intravitrial therapy is?
9:35
SPEAKER_02: And it wasn't just a chat bot, right?
9:37
SPEAKER_02: It was voice.
9:38
SPEAKER_02: Right.
9:38
SPEAKER_02: So that people could actually like have a conversation.
9:41
SPEAKER_00: Yeah, it's voice activated.
9:42
SPEAKER_02: Yes, yes, yes with this bot to learn more about their condition and more about their treatment.
9:48
SPEAKER_02: It's not a real condition, right?
9:50
SPEAKER_02: They were given like this hypothetical scenario.
9:52
SPEAKER_02: Right.
9:52
SPEAKER_02: And then afterwards they were asked, like, oh, did you know, was the information understandable?
9:57
SPEAKER_02: Do you feel like it like was good information, et cetera?
10:01
SPEAKER_00: It was a usability test.
10:02
SPEAKER_00: I mean, they they this is pretty common in in product testing where they build the product and they try to test in the user and say what kind of, you know, whether that user, whether the user liked it, whether it helped them accomplish their task, um, what kind of friction they experienced and you know, that sort of thing.
10:18
SPEAKER_02: Yeah.
10:18
SPEAKER_02: And was that information they got, did it make sense?
10:21
SPEAKER_02: Was it like easy to follow?
10:24
SPEAKER_02: Um, you can't really ask the patient, pretend patient if it's accurate, right?
10:28
SPEAKER_02: But they did go back to providers, healthcare professionals, and say, hey, was it good?
10:32
SPEAKER_02: Was it accurate?
10:34
SPEAKER_02: Um, and was it easy to interact with?
10:37
SPEAKER_02: Was it easy to understand?
10:38
SPEAKER_02: Was it easy to follow?
10:40
SPEAKER_02: Was it organized?
10:40
SPEAKER_02: And they asked all of those questions.
10:42
SPEAKER_02: And for the most part, people were like, Yeah, this was this was something that was helpful.
10:47
SPEAKER_01: Yeah.
10:48
SPEAKER_02: But it was like this very mocked up scenario.
10:51
SPEAKER_02: So I'd love to like dive in a little bit more to this mocked up scenario.
10:57
SPEAKER_00: And they actually provided um an example of like Yes, they had a transcript there of uh of a patient interacting with the bot.
11:05
SPEAKER_00: Um, and you know, one thing is when you read the transcript, it's very clear, very easy to tell which is the patient and which is the bot, even without knowing that they they they give it names, right?
11:14
SPEAKER_00: But uh this, I think Rachel is the is the bot and Sam is the patient.
11:18
SPEAKER_00: And Rachel, you can tell immediately that it's the bot because it talks like all the chat GPTs that you know about right now, which is in this very um happy kind of you're always right.
11:28
SPEAKER_00: Oh, yeah, that's a good point kind of kind of tone.
11:31
SPEAKER_00: Um and you know, so that you can try right away.
11:36
SPEAKER_00: Yes, yes, thing that we talked about before in other other episodes.
11:40
SPEAKER_00: But yeah, I mean, uh you know, I I think uh it's worthwhile to pause for a second and think about what an interaction should look like, right?
11:52
SPEAKER_00: Because uh ultimately we're saying, hey, here we like the idea of a tool, an AI tool, that will help a human after they visited the doctor or whatever to get a better understanding and to take the next steps and overall kind of just like improve their adherence and keep them engaged, right?
12:09
SPEAKER_00: So these are all very uh this there's a range of goals here.
12:12
SPEAKER_00: And this is very interesting because if you think about your use of Chat GPT um and how you use you, you jump on it and you ask it a question, right?
12:21
SPEAKER_00: It's like sitting there waiting to answer your questions.
12:23
SPEAKER_00: And that sort of interaction is very different from say uh a bot that is used for making restaurant reservations or airline, airline ticket reservations.
12:33
SPEAKER_00: So I'll let's talk about that for a second.
12:35
SPEAKER_02: So Yeah, I was gonna say to walk, walk like I get what you're saying, but I'm like, walk me through it.
12:39
SPEAKER_00: So when you make and restaurant reservation, uh the restaurant needs to put you in their system.
12:47
SPEAKER_00: And in order to put you in their system and reserve the table, they need certain information from you.
12:52
SPEAKER_00: They need questions like, how many people are there in your party?
12:55
SPEAKER_00: What time are you thinking of making a reservation?
12:58
SPEAKER_00: And there might be some back and forth, as in, I don't know, there's no availability at 5.30, but there's an availability at 6.30, whatever, right?
13:04
SPEAKER_00: But the point is they need that information about timing, the size of the party.
13:10
SPEAKER_00: Um, any they might have any information about allergies and things like that to inform the chef in advance.
13:15
SPEAKER_02: Um It's almost like a form that you're filling out.
13:18
SPEAKER_02: That's what you're telling.
13:19
SPEAKER_02: It's like logistical stuff, like here is called task-oriented dialogue.
13:23
SPEAKER_02: Like you already know it.
13:24
SPEAKER_02: It's kind of like you could like march it out.
13:26
SPEAKER_00: So you basically imagine the bot having a form that it needs to fill out.
13:30
SPEAKER_00: And if the human doesn't provide it, then the bot has to interject and say, Oh, yeah, no, this is great.
13:36
SPEAKER_00: I think we can accommodate you, but uh what you know, how many people did you want in, you know, or or you know, fill in the things that the human maybe didn't provide voluntarily, right?
13:44
SPEAKER_00: You might have to follow up and ask the question.
13:46
SPEAKER_00: So there's a little bit of asking for the most part, but otherwise, the human is just kind of providing the information that's needed for the bot to fill out the form.
13:54
SPEAKER_00: This is sort of true in restaurant reservations.
13:56
SPEAKER_00: You can think of airline ticket stuff.
13:58
SPEAKER_00: Um It's not true with age related macular degeneration and intra vitreous therapy, though.
14:05
SPEAKER_00: Right.
14:05
SPEAKER_00: But also, if you think about your use of Chat GPT, it's more varied than that.
14:09
SPEAKER_00: You're not just having restaurants reservations filled, you're asking kind of open-ended questions.
14:14
SPEAKER_00: Maybe you're working on a writing project and you want to explore a couple of ideas, so you might suggest a few things and it provides you with recommendations.
14:22
SPEAKER_00: But even there, the um approach is kind of one-sided.
14:28
SPEAKER_00: One person controls the conversation.
14:30
SPEAKER_00: In the task-oriented dialogue, it was the bot that's controlling the conversation and getting the information it needs to make the reservation.
14:37
SPEAKER_02: Yeah, so like 5:30 p.m., two people on this date, whatever.
14:41
SPEAKER_02: Okay.
14:42
SPEAKER_00: It's who's controlling the conversation.
14:43
SPEAKER_02: At ChatGPT, it's like the opposite.
14:46
SPEAKER_02: I'm going in and being like, hey, tell, tell me, give me a summary of everything you know about cheetahs or whatever, and it will then do what I ask it to do.
14:54
SPEAKER_00: Yes.
14:55
SPEAKER_00: In um in an educational setting, the interaction is potentially different.
15:00
SPEAKER_00: Because there, imagine a teacher and a student sitting next to each other, and they're the teacher is teaching the student, and this just imagine human students for in a second here, and human teachers, but they're they're not aliens, not or robots or chatbots.
15:14
SPEAKER_00: Um but you know, so the the teacher is asking, um, maybe maybe the teacher has a goal.
15:20
SPEAKER_00: The teacher's goal is for for the student to understand the best they can uh a certain topic.
15:26
SPEAKER_00: And the student's goal might be also that, uh, but might also be things like I want to do well on the exam or whatever, right?
15:31
SPEAKER_00: Um and so the teacher is saying things to the student to explain a concept, right?
15:38
SPEAKER_00: So the teacher's taking control of the conversation and saying, here is the concept, blah, blah, blah.
15:42
SPEAKER_00: Then passes the control over to the student to ask questions.
15:46
SPEAKER_00: The student asks a question, and then the the teacher themselves might answer the question, but then in answering the question, realize that the student's question actually exposes a gap in their understanding and not only answers the question, but and then takes control back and actually says, no, no, no, actually, your question suggests that you don't understand this concept really well.
16:06
SPEAKER_00: Maybe let's focus on that for a second, and turns the conversation, changes the topic, maybe focuses on a specific aspect of it and really gets into that.
16:15
SPEAKER_00: So there's a back and forth in the conversation.
16:18
SPEAKER_00: And people have used in the conversation literature, people have used ideas of playing ping pong or dancing kind of as a back and forth between the two two parties who are conversing.
16:28
SPEAKER_02: Yeah.
16:28
SPEAKER_02: And you can see that happening in medicine too when you have a conversation like this.
16:32
SPEAKER_02: So please explain, you know, you explain the diagnosis, you explain what the treatment, you know, is or that what the treatment options are.
16:40
SPEAKER_02: And then there's this back and forth where people are asking questions, trying to figure out, you know, maybe they're weighing one treatment against another, or trying to decide like um if they understand X, Y, or Z.
16:53
SPEAKER_02: And so you you can ex you would see that same sort of dance in this clinical scenario.
16:58
SPEAKER_02: Absolutely, right?
16:59
SPEAKER_00: Absolutely.
16:59
SPEAKER_00: And we just talked, and we talked about how a chat bot like this could be used for uh different purposes for education, but also for like maybe scheduling the next appointment or something, right?
17:09
SPEAKER_00: So those are like different threads that when we humans talk to each other, we jump back and forth easily between the different threads.
17:16
SPEAKER_00: We might be talking about something and be like, oh, you know what?
17:18
SPEAKER_00: Can we also just make a quick uh schedule appointment for this?
17:21
SPEAKER_00: Oh, yeah.
17:21
SPEAKER_00: I mean, you jump into that thread and then you do the thing and then you jump back out.
17:25
SPEAKER_00: So when you're in that thread, can we schedule an appointment?
17:27
SPEAKER_00: It's more like the restaurant reservation because at that moment, the uh the AI system should recognize that what it needs is actually to get the information from the human to make the scheduling stuff work.
17:35
SPEAKER_02: Right.
17:36
SPEAKER_00: So then it changes the role into that mode and comes back.
17:39
SPEAKER_00: This whole thing about passing control um back and forth between the speakers uh is what people call initiative.
17:45
SPEAKER_00: And so when you have two interlock uh two TP interlocutors, that's a jargon word for conversation people, uh, is uh you have this notion of mixed initiative that is people go back and forth, take control, pass on, change topics.
17:58
SPEAKER_00: That's very normal in human conversation.
18:01
SPEAKER_00: And they take turns and they they they pass turns, and we're we know when to jump in, when to jump out, you know, and I'm coming in right now.
18:08
SPEAKER_02: Yeah, that's it, right?
18:09
SPEAKER_02: Though it's it's like that very human type of interaction.
18:15
SPEAKER_02: It's not like I ask a question, you answer it.
18:18
SPEAKER_02: I ask another question, you answer it.
18:20
SPEAKER_02: I ask a third question, you answer it.
18:22
SPEAKER_02: No, there's like a back and forth of giving and receiving information.
18:26
SPEAKER_02: And then on top of that, there's like uh there's this deeper piece that you you talked about where it maybe I ask you a question, you realize I don't understand the like the underlying concept.
18:39
SPEAKER_02: And so when you give me the information back, you can answer the specific question, but then you can speak more broadly about the concept as well.
18:45
SPEAKER_00: Right.
18:46
SPEAKER_00: That and so people who are wondering why do we need this kind of sophistication for something that's just answering your questions or providing you with information, that's why.
18:54
SPEAKER_00: Because sometimes you yourself don't know what you don't know.
18:57
SPEAKER_00: And when you ask a question, a person who is an expert in something knows what you're missing right away.
19:03
SPEAKER_00: And they're in the best position to provide you with that information.
19:06
SPEAKER_00: And they're, but they have to recognize that.
19:08
SPEAKER_00: They have to recognize that gap that is there in your understanding, right?
19:11
SPEAKER_00: Or there's some misunderstanding between the people, and they have to immediately capture that gap.
19:16
SPEAKER_00: And that's why you need a bot that's able to do this kind of sophisticated interaction that we expect from other humans.
19:22
SPEAKER_02: And this happens all the time in clinical medicine, right?
19:24
SPEAKER_02: I mean, you might have someone where you've ordered a colonoscopy on them, and they might say, Okay, like I'm gonna do the prep.
19:32
SPEAKER_02: I understand, I need to drink all this stuff, and then I'm gonna go in for the procedure and um and I can drive myself home afterwards, right?
19:40
SPEAKER_02: And so you're like, well, no, you can't drive yourself home afterwards, but then you have to back up and be like, and here's why.
19:47
SPEAKER_02: Right.
19:47
SPEAKER_02: Like you are going to be getting an Ivy, you're going to be receiving sedating medication.
19:52
SPEAKER_02: They're gonna help you sort of breathe when that happens, but you're not actually not gonna be able to drive and you're gonna be out of commission kind of for the rest of the day.
20:00
SPEAKER_02: You're not gonna want to work, you're not gonna be able to drive, et cetera.
20:03
SPEAKER_02: And so not only have you answered the specific question, no, you cannot drive, but you've backed up and explained why that is the case and also given them more bit a better understanding of what the test actually is.
20:16
SPEAKER_00: Yes, exactly.
20:17
SPEAKER_00: And you know, that's I think one of the uh the in this paper, they they don't they don't go into any of that.
20:22
SPEAKER_00: Uh the examples that they provide are uh the one transcript that they provided was, you know, fairly one-directional and didn't have that kind of nuance.
20:30
SPEAKER_00: And so you're not able to evaluate, in fact, if the patient is actually understanding a concept better after this interaction, is going to adhere to it more after this conversation.
20:40
SPEAKER_02: Yeah, this is this is like an interesting piece to me because a lot of the um the patient, the pretend patients interact, and they just say, like, yes, yes, please, my name is Sam.
20:50
SPEAKER_02: Um, they don't really go into a lot of detail, they don't ask a lot of questions, they kind of respond to what the chatbot is saying.
21:00
SPEAKER_02: And ultimately, the goal of doing something like this is to improve what a patient understands in order to improve their adherence to the treatment.
21:10
SPEAKER_02: Yes, and that's not the paper didn't actually look at that.
21:12
SPEAKER_02: No, it's and that's okay, but it I think it was more of like, hey, could could this work?
21:20
SPEAKER_02: And I'm not sure they answered that question.
21:22
SPEAKER_02: It it could work as in there are patient education tools out there, but really what we want is what what counts as working is people actually going and getting their intraventrial therapy.
21:34
SPEAKER_02: And they didn't look at that.
21:35
SPEAKER_00: Yeah, and you know, just asking the patient if they like the tool is not enough, right?
21:39
SPEAKER_00: Because they're in a study and they're gonna either like it or not like it for whatever reason, and that may not translate to the real world when after something they're actually going and using the tool, right?
21:47
SPEAKER_00: They still have to click on something and log in and type a question in or or log in and ask a question to get the conversation started.
21:53
SPEAKER_02: And liking something doesn't mean that it's actually improving clinical outcomes.
21:56
SPEAKER_02: Oh, that's true.
21:57
SPEAKER_02: We know that from like Pres Gaini scores.
21:59
SPEAKER_02: It's like, well.
22:00
SPEAKER_02: Just because you like your doctor doesn't mean that you are having a better, better health outcome at the end of the day.
22:06
SPEAKER_00: Isn't it actually an inverse relationship?
22:07
SPEAKER_00: I don't know if that's true.
22:09
SPEAKER_02: In some studies, they've it's it's definitely been um not one-to-one.
22:15
SPEAKER_02: There's debate about how useful these things are, for sure.
22:18
SPEAKER_02: And so I think here it is what they showed in the transcript is a one-sided conversation, which is not particularly surprising because they're having not they don't have real patients.
22:29
SPEAKER_02: Right.
22:30
SPEAKER_02: So these these people who are playing patients who don't actually have the may not actually have the condition, they're not asking real questions because they don't really have any.
22:38
SPEAKER_02: I mean, they could, but a lot of times it was like, oh, like, here, let me give you some education about what age-related macular degeneration is.
22:47
SPEAKER_02: Does that make sense?
22:49
SPEAKER_02: Yes.
22:49
SPEAKER_02: Like, that's the response.
22:51
SPEAKER_02: The response from the pretend patient is just like, yes, that makes sense.
22:54
SPEAKER_02: But there's nothing there that like actually checks.
22:57
SPEAKER_02: Like, did we actually improve this person's health literacy?
22:59
SPEAKER_02: Right.
23:00
SPEAKER_02: Or is this person actually, of course, because it's a fake patient, it can't be, but like, can we track like were they actually more adherent to their intra vitriol therapy than someone who didn't receive this education?
23:14
SPEAKER_00: And that would be a stronger test for sure.
23:16
SPEAKER_00: Um, I also want to note that you know, one of the challenges with this is drawing the boundary of what it can and can't do.
23:22
SPEAKER_00: And they sort of talk about it not being used for diagnosis, but how do you control for that?
23:27
SPEAKER_00: What if in the middle of the conversation the patient starts to ask questions about the diagnosis or starts asking the the whether the drugs are the right drugs that they were prescribed, or whatever else, right?
23:37
SPEAKER_00: So there's like nuance there, which is another thing that they did not cover, which I think is very important still, especially given all the stuff we've talked about with the AI systems, not um being easy to control, especially control their output.
23:51
SPEAKER_02: Yeah, you don't want this thing going rogue.
23:52
SPEAKER_02: Like you want to make sure, and in this specific use case, the providers came in and said, hey, it's giving accurate information.
24:00
SPEAKER_02: That's great.
24:00
SPEAKER_02: But this is all like exactly what you would expect it to be saying.
24:05
SPEAKER_02: There wasn't a question that would like make it go kind of off script.
24:08
SPEAKER_02: Right, right.
24:09
SPEAKER_02: Right?
24:09
SPEAKER_02: There wasn't, there wasn't anything complicated.
24:12
SPEAKER_02: A lot of the stuff from the fake patient was like, yes, I'm on insurance Medicare.
24:17
SPEAKER_02: Uh, no, I don't have any issues with transportation.
24:21
SPEAKER_02: Let's confirm the date of my appointment, whatever.
24:24
SPEAKER_02: Uh it doesn't have anything that would put it off target.
24:28
SPEAKER_02: It doesn't have anything that's more complex, right?
24:30
SPEAKER_00: Yeah, and there's the right, and there's nothing stopping it from being being being steered that way.
24:36
SPEAKER_02: Right.
24:37
SPEAKER_02: And that's that's not what this patient interaction was, but uh you could imagine in a real world scenario, you could get a patient asking something kind of obscure or weird or off topic.
24:48
SPEAKER_02: Right.
24:48
SPEAKER_02: And what's gonna happen in that situation?
24:50
SPEAKER_02: I don't know.
24:51
SPEAKER_02: I don't know what's gonna happen in that situation.
24:52
SPEAKER_02: Anything could happen in that situation.
24:54
SPEAKER_00: Yeah, yeah, yeah, yeah.
24:55
SPEAKER_00: No, exactly, exactly.
24:57
SPEAKER_00: Yeah, so I and I think there's a host of interesting this is a fantastic use case.
25:03
SPEAKER_00: Absolutely.
25:04
SPEAKER_00: And and I think that it is pot has a lot of potential.
25:07
SPEAKER_00: I hope people are building these kinds of tools.
25:09
SPEAKER_00: Um at the same time, I hope that they're also taking into account the intricacies and nuance of human communication, which is not just text-based, you know, question answering, it is a lot more.
25:21
SPEAKER_00: It's mixed initiative dialogue that requires um proactive people, proactive speakers.
25:29
SPEAKER_00: It requires a shared building a shared understanding of something that you're talking about.
25:35
SPEAKER_00: It requires um understanding what you can and cannot say, um, and it requires making sure that your goals and the uh the other person's goals are kind of met, right?
25:46
SPEAKER_00: So there's like a lot of lot that goes on in communication.
25:49
SPEAKER_00: And of course, then there's the nuts and bolts of communication.
25:51
SPEAKER_00: This is a voice-activated device, and we know that the research isn't quite there.
25:55
SPEAKER_00: They don't, you know, there's like challenges with when to um when to um jump into conversation when not to.
26:04
SPEAKER_00: Turntaking is a very complex area in and of itself.
26:07
SPEAKER_02: I know I like to interrupt you all the time.
26:08
SPEAKER_02: It's not so easy with a bot.
26:10
SPEAKER_00: But even you timed your interruption, right?
26:11
SPEAKER_00: You didn't just jump in while I was saying a word.
26:14
SPEAKER_00: You jump, you jumped in when there was a moment, maybe because of a pause, but we know that I can go down a rabbit hole here because there's a lot of research about it.
26:20
SPEAKER_02: I mean, this is fascinating.
26:22
SPEAKER_02: Look, I just I just jumped in in the middle of a sentence and I didn't even think of it.
26:25
SPEAKER_00: But that's fine.
26:26
SPEAKER_00: I mean, it was it was normal in the conversation, right?
26:28
SPEAKER_00: Yeah.
26:28
SPEAKER_00: And there are rude interruptions and there are fine interruptions, right?
26:31
SPEAKER_00: And so there's that, there's that distinction.
26:32
SPEAKER_00: So there's a lot of nuance there.
26:34
SPEAKER_00: And all of that matters.
26:35
SPEAKER_00: And people might say, oh, well, that doesn't really matter.
26:37
SPEAKER_00: Well, it matters because um that changes the tone of the interaction and that changes the trust you place on the person you're speaking with and the comfort level you have.
26:47
SPEAKER_00: And that speaks to things like empathy and other all those other pieces, which are super important in this medical context because it's a patient that is trying to understand their situation, their plight.
26:57
SPEAKER_00: And in that case, all of the little nuts and bolts matter too.
27:01
SPEAKER_02: Yeah, I can totally see that.
27:03
SPEAKER_02: And I think this is something that could probably only be like a complement or a supplement to a human-provided education.
27:11
SPEAKER_02: There's definitely a use case for it because there is limited time in the office.
27:15
SPEAKER_02: There are questions that come up afterwards, uh, there's language uh barriers, um, there's there's so much going on.
27:23
SPEAKER_02: And the empathy piece is like or the display of empathy piece is something that's really important for people.
27:30
SPEAKER_02: But there's a lot that still needs to be worked out.
27:32
SPEAKER_02: And I honestly, until we did this episode, I hadn't even thought about the way that people that humans talk and sort of the one-sided nature of a lot of the interactions that we have with AI on a regular basis and why having this like two-sided give and take is really important and really hard to reproduce.
27:50
SPEAKER_00: Yeah, exactly.
27:53
SPEAKER_02: So the future hopefully will have some of this, but I think there's a lot of work that needs to be done first in order to get that double sided give and take interaction uh into patient education.
28:07
SPEAKER_02: Agreed.
28:08
SPEAKER_02: All right, I think we can end here.
28:10
SPEAKER_02: We'll see you next time on Coding Cure.
28:12
SPEAKER_00: Thank you for joining us.