#68 - When AI Makes Care More Expensive

WEBVTT
00:00:00.239 --> 00:00:08.240
What if one of AI's biggest impacts on healthcare isn't better diagnoses or faster treatment, but higher bills?
00:00:08.480 --> 00:00:13.839
Today we're looking at how AI-driven documentation could increase the cost of care.
00:00:23.039 --> 00:00:28.239
Hello and welcome back to Code and Cure, where we discuss decoding health in the age of AI.
00:00:28.399 --> 00:00:30.160
My name is Vasant Sarathi.
00:00:30.320 --> 00:00:35.600
I'm an AI researcher and cognitive scientist, and I'm here as always with Laura Hagopian.
00:00:35.759 --> 00:00:38.880
I'm an emergency medicine physician and I work in digital health.
00:00:39.200 --> 00:00:50.320
Let's talk about what the world kind of was, kind of the world is right now, when you don't have AI in the mix, right?
00:00:50.399 --> 00:01:06.239
So can you, Laura, can you I mean we talked about this before the start of this, but like can you walk us through an example of what types of tasks um you might do and others in the hospital might do uh that maybe in the future and even in the process of do happening right now is getting automated, right?
00:01:06.319 --> 00:01:08.640
So like can you walk us through an example?
00:01:09.040 --> 00:01:09.760
Yeah, sure.
00:01:09.840 --> 00:01:20.799
I mean, I work shifts at urgent care right now, and um and one of the first things that I did was get access to the AI scribe for urgent care.
00:01:21.280 --> 00:01:31.680
And so um, when I go and see a patient, it used to be that I would, you know, listen, listen to what they had to say, ask questions, examine them, et cetera.
00:01:31.760 --> 00:01:38.000
And then I'd have to leave the room, go back to my computer and like either type up or usually dictate a note.
00:01:38.319 --> 00:01:39.040
What does that look like?
00:01:39.200 --> 00:01:44.400
And that looks like, hey, you know, this is a 35-year-old male with a history of hypertension.
00:01:44.560 --> 00:01:46.560
He came in today with sore throat.
00:01:46.640 --> 00:01:49.439
It's been going on for three days, it's getting worse.
00:01:49.599 --> 00:01:53.599
He has no associated fevers, his child is also sick, et cetera, et cetera.
00:01:53.680 --> 00:01:58.159
So I'm dictating all of that or typing all of that into a note.
00:01:58.319 --> 00:02:03.439
I'm clicking some boxes or using like a what we call like a macro, like in preload.
00:02:03.519 --> 00:02:06.560
Oh, this is what a normal sore throat exam looks like.
00:02:06.719 --> 00:02:16.560
Um, or I might have these things called dot phrases where I type like dot sore throat and and a lot of the normal sore throat questions will just pop up for me because I've programmed that in.
00:02:16.719 --> 00:02:17.039
Yes.
00:02:17.199 --> 00:02:20.560
Um personalized to me, right?
00:02:20.800 --> 00:02:26.560
But I do this, what we call the review of systems, which is like, hey, does anything is anything else bothering them?
00:02:26.719 --> 00:02:29.039
Do they have like a do they also have to pain?
00:02:29.120 --> 00:02:30.159
Do they have a runny nose?
00:02:30.240 --> 00:02:30.639
Whatever.
00:02:30.800 --> 00:02:42.400
And I go through my exam and then I do this sort of assessment and plan where I'm like, hey, you know, uh, based on the information that I've noted above, I don't think there's anything dangerous going on.
00:02:42.479 --> 00:02:47.520
I think this is probably a viral syndrome, or we're gonna do a strep swab to rule out strep throat.
00:02:47.599 --> 00:02:47.759
Yeah.
00:02:48.000 --> 00:02:49.919
And this is kind of the plan moving forward.
00:02:50.159 --> 00:02:50.960
And here's why.
00:02:51.039 --> 00:02:53.439
And this is the list of the diagnoses it could be.
00:02:53.599 --> 00:02:54.800
Here's why I think it's this one.
00:02:54.960 --> 00:02:59.199
Here's why, here's why I don't think it's anything dangerous, here's what I've counseled the patient on, etc.
00:02:59.680 --> 00:03:06.319
And then I'll go through and I'll put a billing code on it and I'll, you know, send send it away.
00:03:06.479 --> 00:03:12.879
Now I have to tell you, because I have the AI scribe, yeah, a lot of that is like a lot easier now.
00:03:13.199 --> 00:03:13.520
Sure.
00:03:13.680 --> 00:03:17.520
But okay, so now you've you've done that step.
00:03:17.919 --> 00:03:19.520
What happens in the billing?
00:03:19.680 --> 00:03:29.360
I guess I I it would be nice to kind of better understand what happens with respect to assigning billing codes and with respect to assigning just the coding in general, right?
00:03:29.599 --> 00:03:32.479
Yeah, it's funny because it's like not something that you get uh It's not a drop.
00:03:32.560 --> 00:03:33.120
Is it a drop down?
00:03:33.280 --> 00:03:40.159
Is it like in the note can you just like drop down and choose what quote unquote category of billing it should be, or no?
00:03:40.479 --> 00:03:47.680
Well, oftentimes you're assigning a diagnosis code, and then you're attempting, at least I'm attempting, to assign a billing code.
00:03:47.840 --> 00:03:53.360
When I worked in the ER, the um the coders would like check my chart and see if we could bill for anything else.
00:03:53.520 --> 00:03:57.439
Because it's not something you like learn in medical school how you could bill.
00:03:57.840 --> 00:03:58.719
You said the coders.
00:03:58.800 --> 00:04:07.599
So there's like a separate group of people who are whose job it is is to make sure that the correct um insurance billing codes are assigned.
00:04:08.240 --> 00:04:09.919
That is correct in some places.
00:04:10.159 --> 00:04:10.479
Okay.
00:04:10.639 --> 00:04:13.280
That's that actually doesn't happen at the urgent care where I work.
00:04:13.360 --> 00:04:14.319
I'm responsible for it.
00:04:14.560 --> 00:04:18.079
But but at a large hospital, that probably will be the case.
00:04:18.560 --> 00:04:24.959
And so these people they scour the charts and they're like, hey, did we bill for what we should have?
00:04:25.120 --> 00:04:25.360
Right?
00:04:25.600 --> 00:04:27.600
Like, oh, Laura did this procedure.
00:04:27.680 --> 00:04:29.680
Let's make sure we we bill for this procedure.
00:04:30.000 --> 00:04:30.319
Got it.
00:04:30.560 --> 00:04:38.480
Or uh Laura billed a level three, but it's more complex because of all the things she documented, and maybe we should bill it a level four.
00:04:38.720 --> 00:04:39.439
Yeah, yeah.
00:04:39.680 --> 00:04:47.040
Um, I don't spend a ton of time with the billing because it's, I don't know, I'm spending more of my time with the patient care, right?
00:04:47.199 --> 00:04:51.279
But when I when I do work my shifts at urgent care, I am putting in the billing code.
00:04:51.360 --> 00:04:54.240
Oh, this is a new patient versus an established patient.
00:04:54.480 --> 00:04:56.480
And then this is the complexity of their visit.
00:04:56.560 --> 00:05:08.399
It's more complex if I like prescribe them a medication, did uh an x-ray, um, you know, gave them a gave them a medication at urgent care, uh, did a flu or COVID test.
00:05:08.560 --> 00:05:12.399
Those are the things that make the visits more complex when I'm doing more for them.
00:05:12.560 --> 00:05:15.680
And then of course, there's a separate whole thing around procedures.
00:05:15.759 --> 00:05:20.480
Like if I did a laceration repair or I drained an abscess, then I'm billing for that as well.
00:05:20.720 --> 00:05:26.160
And so I'm putting in those codes and I'm making an assessment of the complexity.
00:05:26.560 --> 00:05:33.920
But one of the things that I think people learned along the way, and this is pre-AI, was hey, providers aren't the best at doing this.
00:05:34.079 --> 00:05:41.839
We could make more money if we have coders come in who like really understand all of these criteria because there's so many codes out there.
00:05:41.920 --> 00:05:56.319
There's like the ICD 10 codes, there's the ENM codes, there's CPT codes, hic pix codes, and like there are people whose entire job it is andor was to like kind of go through the charts and be like, hey, can we can we get more money for can we get more money for this?
00:05:56.399 --> 00:06:00.720
Can we bill for this better based on what was actually done during the visit?
00:06:01.120 --> 00:06:01.920
Right, right, right.
00:06:02.000 --> 00:06:17.519
And again, that job is pretty complex because you end up having to look at charts and and notes and labs and all these other things to kind of figure out what the right um, you know, what are all the possible codes that need to be uh uh you know recorded.
00:06:17.839 --> 00:06:18.079
Right.
00:06:18.160 --> 00:06:24.160
So they have to synthesize a lot of information, which could be in disparate data sources, right?
00:06:24.319 --> 00:06:24.560
Yeah.
00:06:24.720 --> 00:06:30.480
And then they have to make an assessment, and a lot of that is kind of rules-based of what codes can be applied.
00:06:30.879 --> 00:06:46.480
Yeah, and this this is I think a good kind of context because now, more recently over the last few years, as AI has made its way into various aspects of the uh the medical practice and the healthcare system, um, you start to see AI uses.
00:06:46.560 --> 00:06:53.600
And and one, two particular uses I think we want to focus on today is on um these AI-enabled scribes, right?
00:06:53.759 --> 00:07:09.439
Where uh or ambient AI, we've talked about this in prior uh prior episodes, but um and it's a pretty cool technology because that kind of all it does is is listens to the your your your conversation with the patient and is able to put together that note that you described um in uh automatically.
00:07:09.680 --> 00:07:11.839
Saves saves doctors a ton of time, right?
00:07:11.920 --> 00:07:13.759
You don't have to like write the stuff later.
00:07:14.079 --> 00:07:14.319
Right.
00:07:14.399 --> 00:07:18.959
It listens to what I'm doing in the room and then turns that into a note.
00:07:19.040 --> 00:07:28.800
And I find it particularly helpful in what I call the HPI, the history of present illness section, where I where I'm asking, hey, how long's that sore throat been going on for?
00:07:28.959 --> 00:07:30.639
And is it getting worse or better?
00:07:30.720 --> 00:07:37.120
And have you had any fevers with it and any trouble opening your jaw and um anybody else been sick at home?
00:07:37.279 --> 00:07:41.199
Because those answers may be different for every every person who comes in with a sore throat.
00:07:41.600 --> 00:07:43.519
So it's able to personalize it.
00:07:43.680 --> 00:07:48.079
And then the other piece that I really like it for is that sort of assessment and plan.
00:07:48.160 --> 00:07:50.240
Like, here's what I think is going on.
00:07:50.480 --> 00:07:51.759
Here's what we're going to do.
00:07:51.839 --> 00:07:56.720
Because I'm when I counsel the patient about that, now it's gonna automatically appear in the chart.
00:07:56.800 --> 00:08:05.759
And again, in a personalized way, like, oh, you have strep throat, we're gonna treat you with a course of penicillin, or for the person who's penallergic, we're gonna treat you with a course of zithromycin.
00:08:05.920 --> 00:08:09.279
Or for the person who doesn't have strep throat, hey, I think this is a viral infection.
00:08:09.360 --> 00:08:16.399
And so whatever I'm counseling the patient about in the room, it's a personalized version of that that appears in the chart for me.
00:08:16.480 --> 00:08:22.639
And whatever questions they have, whatever return precautions that I give them, those automatically get entered in.
00:08:22.800 --> 00:08:26.560
And so I find it really simplifies that piece of charting.
00:08:26.879 --> 00:08:30.800
Yeah, and you know, obviously there's issues of like it, there are issues, right?
00:08:30.879 --> 00:08:36.159
There's some there are there are issues of how accurately it captures everything or it misses important things or it captures other things.
00:08:36.240 --> 00:08:38.639
I mean, that's we're we're not talking about that in today's episode.
00:08:38.799 --> 00:08:41.279
But that is one use case of in this system.
00:08:41.440 --> 00:08:54.720
And the other big use case is um AI assisted coding, uh, which we talked about, you know, as being this separate task where you have uh somebody looking through the labs and charts and stuff and identifying the medical codes to bill.
00:08:55.039 --> 00:09:18.399
And again, that's another one would imagine, at least as an AI researcher, that comes for me to me across as an exact exact great example of an AI use case where you have all of these charts and you're putting together something, a system where it just identifies discrete set of codes, one or more discrete set of codes that captures the complexity uh and the you know what was done uh on the patient.
00:09:18.639 --> 00:09:25.039
And so I think to me that is a great use of AI and it's it's being used a lot more now.
00:09:25.360 --> 00:09:26.480
Absolutely it is, yeah.
00:09:26.879 --> 00:09:38.000
I think this is uh where for us at least, uh this all these papers that we're gonna uh talk about today, kind of collectively, uh kind of flips that on the head a little bit.
00:09:38.080 --> 00:09:44.960
It sort of says, wait a minute, what is the con I mean, sure, it feels like it's faster and more efficient and better for the system, but is it?
00:09:45.600 --> 00:09:48.480
Well, it's interesting because it like depends who you ask, right?
00:09:48.559 --> 00:09:55.440
Like some of the papers that we have today are written by Blue Cross Blue Shield, and they're like, hey, for the insurance companies, this isn't so great.
00:09:55.600 --> 00:10:00.080
You're you know, everything's getting billed for higher amounts, and it's giving us more money.
00:10:00.399 --> 00:10:01.120
So let's talk about that.
00:10:01.200 --> 00:10:02.399
Like, why does that happen, right?
00:10:02.559 --> 00:10:10.879
What we're trying to say is that um with AI enabled scribes and AI assisted codes, somehow the healthcare price goes up.
00:10:11.200 --> 00:10:11.360
Right.
00:10:11.440 --> 00:10:16.480
And uh, that's not surprising to me at all because one, you have more complete clinical documentation.
00:10:16.639 --> 00:10:20.320
Like I can tell you that the AI scribe writes way more in the chart than I ever would.
00:10:20.799 --> 00:10:21.200
That's right.
00:10:21.440 --> 00:10:27.440
So now, now when it when the when these agents go in to look and see, hey, what can we code for?
00:10:27.519 --> 00:10:32.720
There's way more information there than there was when I was just like typing it out or dictating it.
00:10:32.960 --> 00:10:37.759
And two, is that the coding practices are now uh quote unquote optimized.
00:10:37.919 --> 00:10:39.759
I mean, who knows what optimized really means?
00:10:39.840 --> 00:10:42.080
But like I'm not an expert coder.
00:10:42.240 --> 00:10:44.159
Um, and I code things all the time.
00:10:44.320 --> 00:10:48.000
So I'm uh what if I'm not coding to maximize reimbursement?
00:10:48.159 --> 00:10:50.080
What is what is my healthcare system?
00:10:50.240 --> 00:10:51.840
What does any healthcare system want to do?
00:10:51.919 --> 00:10:53.600
At the end of the day, they need to bring in money.
00:10:53.759 --> 00:11:00.480
So they should be coding to the best of their ability for what was medically justifiable and happened in that situation.
00:11:00.720 --> 00:11:03.919
And that's a lot of that is rules-based, what you can bill for.
00:11:04.080 --> 00:11:15.759
And so if they can ink, if if these AI coding agents can show a higher complexity, then that means that the healthcare system can bill for something that's more money.
00:11:16.080 --> 00:11:22.080
And those AI agents who are doing the coding, uh, let's call them AI coders and AI scribers, right?
00:11:22.240 --> 00:11:22.639
Scribes.
00:11:22.720 --> 00:11:22.960
Yeah.
00:11:23.200 --> 00:11:28.399
So um, so the AI coder coders um are going to are incentive.
00:11:28.559 --> 00:11:30.960
I mean, maybe they're incentivized, maybe they're trained like that or not.
00:11:31.039 --> 00:11:31.759
It doesn't matter.
00:11:31.919 --> 00:11:37.279
Are that's totally how they're advertised when people sell these AI coders, like it's gonna make you more money.
00:11:37.600 --> 00:11:38.480
Oh, interesting.
00:11:38.639 --> 00:11:38.879
Yeah.
00:11:39.039 --> 00:11:45.200
So it is designed to sort of squeeze out all the possible codes from a certain situation.
00:11:45.360 --> 00:11:54.080
And when you combine that with the AI scribes who are providing more and more information, it's easier for the AI coders to squeeze out more information.
00:11:54.240 --> 00:12:10.559
And together, the combined effect is quite interesting because what ends up happening is you have way more complexity emerging from the particular case, even if the exact case that you just described, you know, um, is the same.
00:12:10.960 --> 00:12:16.480
I think that's the thing, is like what you're not changing in this system is the patient care, right?
00:12:17.120 --> 00:12:18.879
The patient still has the same exact problem.
00:12:19.120 --> 00:12:19.279
Right.
00:12:19.440 --> 00:12:21.919
And and and you're as a doc as a doctor helping them with that.
00:12:22.159 --> 00:12:25.679
And I'm doing the as a provider, I'm doing the exact same thing.
00:12:25.919 --> 00:12:29.279
And now the healthcare system is billing for it at a higher complexity.
00:12:29.360 --> 00:12:31.679
It's getting reimbursed at a higher complexity.
00:12:31.919 --> 00:12:36.720
And so because it was medically justifiable, and so now the healthcare costs are going up.
00:12:37.120 --> 00:12:48.159
Now let's let's just take a pause for a second because the some of the papers we've identified here are written by the insurance companies and they're upset because now they're uh because the the cases are uh over time have gone up in complexity.
00:12:48.320 --> 00:12:58.480
Now, they could have gone up in complexity because people are just sicker, but they've been able to identify that that may not be the case, that in fact that's not a reason uh for for that for that increased complexity.
00:12:58.639 --> 00:13:05.919
And let's just even assume that the AI systems are the ones, AI scribes and coders are the ones that are actually increasing the complexity.
00:13:06.000 --> 00:13:07.360
Let's just assume that for a second.
00:13:07.600 --> 00:13:14.080
It is also possible that the hospitals were getting underpaid before in the first place, right?
00:13:14.159 --> 00:13:20.559
Like maybe, maybe we didn't, I mean, I can tell you like I was not a great cod, I was never a great coder.
00:13:20.720 --> 00:13:23.840
It wasn't really like I was always focused more on the patient.
00:13:24.080 --> 00:13:30.399
Um, and so like I I'm sure that I missed certain things when I was told to code.
00:13:30.559 --> 00:13:30.799
Yeah.
00:13:30.960 --> 00:13:31.200
Right.
00:13:31.360 --> 00:13:35.120
I'm just I'm like making sure I stick in a code so I can complete my chart.
00:13:35.679 --> 00:13:39.840
Not trying to necessarily make the most money off the code.
00:13:39.919 --> 00:13:42.960
I'm more concentrated on like, oh, I'm doing CPR on that patient.
00:13:43.039 --> 00:13:44.159
I'm intubating them.
00:13:44.320 --> 00:13:45.679
I need to do this procedure.
00:13:45.840 --> 00:13:48.240
This person needs, I need to give them their CAT scan results.
00:13:48.320 --> 00:13:49.679
I need to tell them that they have cancer.
00:13:49.759 --> 00:13:54.159
I'm not spending a ton of time in the chart trying to like squeeze every dollar out of it.
00:13:54.399 --> 00:13:54.720
That's right.
00:13:54.799 --> 00:13:55.679
And you shouldn't be, right, right.
00:13:55.919 --> 00:13:56.080
Correct.
00:13:56.240 --> 00:13:56.480
Exactly.
00:13:56.879 --> 00:14:07.200
So it is interesting because like when I saw some of the these papers from the insurance company, I was like, hmm, well, maybe we should have been able, maybe we should have been billing that way in the first place.
00:14:07.600 --> 00:14:08.080
I don't know.
00:14:08.240 --> 00:14:08.639
Yeah.
00:14:08.799 --> 00:14:21.200
Um, but of course, from the insurance company standpoint, they're like, well, now, now that this is being billed at a higher complexity and we are having to pay at a higher complexity, we're paying more money.
00:14:21.360 --> 00:14:28.639
And then of course, what that makes me think, geez, okay, well, those costs have to be transferred somewhere, right?
00:14:28.720 --> 00:14:36.639
And so if insurance companies are paying more money, that means that the employers who who buy insurance are paying more money.
00:14:37.039 --> 00:14:45.360
And that means that the patients ultimately are going to be paying more money too, because healthcare costs are rising like crazy in the United States.
00:14:45.600 --> 00:14:54.720
And this if if things are costing more money, that cost is being transferred somewhere, and a lot of times that is to employers and to patients.
00:14:55.200 --> 00:15:03.120
I I think what's also in what's also interesting is that uh the uh costs are going up, but it's complicated, right?
00:15:03.200 --> 00:15:15.519
So people often say, well, let's put AI in place so that the physicians can save time writing notes, which means they can see more patients, which means that they can the hospital can make more money and it's more efficient that way.
00:15:15.919 --> 00:15:33.679
They also say that, you know, by saving time and saving money, you might be it might offset, you know, like the AI, AI's own cost, but we've just talked about the increase in cost because of complexity of cases, uh, might be offset a little bit by increased efficiencies.
00:15:34.559 --> 00:15:34.879
Maybe.
00:15:35.200 --> 00:15:39.840
But we don't, I mean, this is this is where this the whole system comes into play and it gets really complicated, right?
00:15:39.919 --> 00:15:40.879
It gets really complex.
00:15:41.200 --> 00:15:44.879
But we are seeing increased healthcare costs for for everybody.
00:15:45.200 --> 00:15:52.639
And that's AI isn't solving the the point of AI, some of the point of AI is to reduce that so that we can save, you know, you can't do it.
00:15:52.879 --> 00:15:53.759
Well, who says, right?
00:15:53.919 --> 00:15:58.240
Like, this is the problem is like whose goal is it and what are they using AI for?
00:15:58.399 --> 00:16:06.399
Like it might be yours and my goal to try to reduce health care costs overarchingly, but is it the the hospital's goal to do that?
00:16:06.639 --> 00:16:08.799
Like, is it you know what I mean?
00:16:08.960 --> 00:16:14.240
It's like what is the hospital or healthcare system's goal in this standpoint?
00:16:14.559 --> 00:16:20.399
It because they have to make money to exist, like they're gonna try to maximize reimbursement.
00:16:20.559 --> 00:16:20.799
That's right.
00:16:20.960 --> 00:16:27.279
So I kinda like I kind of almost disagree with your point there that everyone wants lower healthcare costs, like you and I do, sure.
00:16:27.600 --> 00:16:34.159
But I think a lot of these healthcare systems, in order to stay afloat, have to do things to try to maximize their reimbursement.
00:16:34.240 --> 00:16:35.679
You can't really blame them for doing that.
00:16:35.919 --> 00:16:43.759
No, I guess I guess my point was that the hospitals themselves, by having pa doctors see more patients uh per hour or whatever, are going to benefit.
00:16:43.840 --> 00:16:44.559
They're gonna make more money.
00:16:44.799 --> 00:16:46.240
Yeah, but where are these patients coming from?
00:16:46.320 --> 00:16:47.440
Are they magically appearing?
00:16:47.600 --> 00:16:52.000
Or are you just like understaffing and assuming that they you know what I mean?
00:16:52.240 --> 00:17:00.000
No, I I I guess my point is again, the the point of having an AI scribe is that the doctor saves time by doing, doesn't have to write the notes and write do all of that.
00:17:00.240 --> 00:17:04.079
And so by having an AI scribe, you now save the doctor some time.
00:17:04.240 --> 00:17:13.279
And it's not that Or maybe you just have like a less Harried doctor or provider who can like spend more of their time with the patient.
00:17:13.440 --> 00:17:17.119
I don't know that the goal has to be that they see more patients.
00:17:17.279 --> 00:17:30.319
Maybe the goal could be that they actually get to spend 15 minutes with the patient and one minute with the chart instead of five minutes with the patient or and the rest of that time with the chart.
00:17:30.640 --> 00:17:30.880
Okay.
00:17:30.960 --> 00:17:32.240
Yeah, yeah, that's fair.
00:17:32.559 --> 00:17:44.720
Um yeah, and I I I was just pointing out the fact that there are the the system is complex enough that a saving here might result in an increased cost elsewhere, and an increase and a you know, saving there might reduce, you know, do something else here.
00:17:44.799 --> 00:17:47.200
And so I think that's kind of what my point was about this.
00:17:47.599 --> 00:17:49.440
Um so it gets pretty complicated.
00:17:49.519 --> 00:17:54.960
So it's not as simple as AI throw AI in to make some cost savings and increase profit.
00:17:55.119 --> 00:18:05.200
It's more like throw AI in, uh save some money here, but oh, here's some um, you know, uh unintentional uh consequences uh of that.
00:18:05.359 --> 00:18:06.319
And for sure.
00:18:06.480 --> 00:18:09.200
And that's what this these the these papers are kind of identifying.
00:18:09.440 --> 00:18:14.400
And I here the end result is an overall increased cost in healthcare, which is not great.
00:18:14.720 --> 00:18:15.200
Exactly.
00:18:15.359 --> 00:18:20.480
And I think it's worth going into at least one of the examples that Blue Cross Blue Shield put out.
00:18:20.640 --> 00:18:28.480
Um, this one is from last month where they were looking at major bowel procedures during uh inpatient hospital stays.
00:18:28.880 --> 00:18:32.720
And basically what they found was like the costs were going up.
00:18:33.039 --> 00:18:41.359
Um, and it wasn't because more surgeries or more procedures were happening, it's because of the way that things were being billed and coded.
00:18:41.440 --> 00:18:41.599
Yeah.
00:18:41.839 --> 00:18:50.160
And that it was being billed or coded at a higher complexity because extra sort of secondary diagnoses were being added on.
00:18:50.720 --> 00:18:56.559
And it was specifically in places that have these AI coding and billing software.
00:18:56.720 --> 00:19:03.759
And a lot of times it was because of a single lap result that they were able to say, hey, this is more complex.
00:19:03.920 --> 00:19:04.160
Yeah.
00:19:04.400 --> 00:19:10.319
So oftentimes it was like, oh, their blood count was a little bit low or their sodium level was a little bit low.
00:19:10.640 --> 00:19:16.720
And the AI coding systems recognizing that could say, hey, that's that's a more complex care.
00:19:16.880 --> 00:19:26.559
And they were able to sort of bump it up in in level and bump it up in billing level, get reimbursed more for it based on this like single lab value.
00:19:26.799 --> 00:19:36.160
This is something that uh you can speak to probably, but like, is that an easy task for AI to do to like go in and look at a lab and be like, oh, that that that lab makes it more complex?
00:19:36.559 --> 00:19:37.039
Yeah.
00:19:38.000 --> 00:19:39.839
It's just like a rules-based thing, right?
00:19:40.000 --> 00:19:40.400
Yep.
00:19:40.720 --> 00:19:43.680
And so it's it's lab data mining, yep.
00:19:44.000 --> 00:19:52.640
That you know, as a coder, you're kind of pouring through all of that and having to remember all of those rules.
00:19:53.200 --> 00:19:56.960
If you program an AI bot to do it, it's like easy.
00:19:57.200 --> 00:19:58.160
Yeah, yeah.
00:19:58.480 --> 00:19:58.799
Yeah.
00:19:59.119 --> 00:20:11.599
And so that's what they found was that these sort of lab derived values, they were able to increase complexity of the billing and therefore bump up how much they would get reimbursed for it.
00:20:11.839 --> 00:20:13.279
Yeah, yeah, no, that's absolutely right.
00:20:13.440 --> 00:20:13.680
Yeah.
00:20:13.759 --> 00:20:18.000
So this is uh I think this is an I think this is an ongoing topic.
00:20:18.079 --> 00:20:21.119
I think we have to we have to return to it because it's very interesting.
00:20:21.359 --> 00:20:30.400
And as more and more AI tools enter the system uh in different places, not just these two, uh what does that do to overall healthcare costs?
00:20:31.359 --> 00:20:31.920
And why?
00:20:32.079 --> 00:20:34.799
I mean, what motivations have we set up for people, right?
00:20:34.960 --> 00:20:35.200
Yeah.
00:20:35.440 --> 00:20:42.799
It's like the motivation of uh a healthcare facility is gonna be to optimize uh coding.
00:20:43.200 --> 00:20:52.640
I also think that here is an example of multiple AI tools that individually might be helpful, but when paired together might actually make things worse.
00:20:53.039 --> 00:21:05.519
Uh and so here we have a case where uh the coders were identifying things you know almost too much because the scribes were writing in maybe almost too much.
00:21:06.000 --> 00:21:11.920
And it's that it's that interaction between those two that really is making things much worse.
00:21:12.000 --> 00:21:21.200
And so I as you get more and more AI tools in the mix, you're gonna have these kinds of higher order interactions between them that we have to be able to model and think about and see what happens.
00:21:21.519 --> 00:21:28.319
Yeah, I think that's a really interesting point that it's the interaction between these different agents that's like creating some of this.
00:21:28.559 --> 00:21:40.559
I also think it's interesting because if you have a coder, like a human coder going into a chart and and going through things, they may not upcode for certain things.
00:21:40.720 --> 00:21:52.400
Like one of the examples given in some of these articles where was where they had um a procedure done or gave a deli had a delivery uh on the maternity ward and had an anemia afterwards, like their blood counts went down.
00:21:52.559 --> 00:21:58.720
Not an uncommon thing to happen afterwards from bleeding for or whatever.
00:21:59.039 --> 00:22:09.119
A lot of times these things don't need to be treated because you know they didn't go down below a certain threshold and the patient's doing fine.
00:22:09.359 --> 00:22:12.640
We don't need to transfuse them, we know it will get better over time.
00:22:12.880 --> 00:22:18.799
Like I personally might not say, okay, well, that's that's an acute anemia from hemorrhage from this.
00:22:18.880 --> 00:22:20.880
I I might not put that into the chart.
00:22:21.039 --> 00:22:24.400
Maybe even a human coder wouldn't, unless they got transfused.
00:22:24.559 --> 00:22:24.720
Yeah.
00:22:24.960 --> 00:22:32.720
Whereas an AI coder might be say, might, might actually say, Oh, well, I could bill for it, so I might as well bill for it because we could get reimbursed for it.
00:22:32.880 --> 00:22:34.559
And it's true based on the lab values.
00:22:34.720 --> 00:22:34.960
Yes.
00:22:35.200 --> 00:22:41.279
And so this is another example of where, hey, the the patients are not getting more complex.
00:22:41.440 --> 00:22:48.240
The treatment's not getting more complex, but we're trying to squeeze every last ounce of of what we can bill out of this system.
00:22:48.400 --> 00:22:50.319
And that's the way that it's set up.
00:22:50.640 --> 00:22:51.039
All right.
00:22:51.200 --> 00:23:00.000
So I think this interaction between one AI scribes and two AI coders is something that can definitely be driving up the cost of care.
00:23:00.240 --> 00:23:03.200
Um, and it's something that we'll definitely have to keep an eye on.
00:23:03.440 --> 00:23:03.839
Yep.
00:23:04.319 --> 00:23:05.359
I think we'll call it there.
00:23:05.440 --> 00:23:06.079
Thank you for joining us.
00:23:06.319 --> 00:23:08.000
We'll see you next time on Code and Cure.
00:23:08.240 --> 00:23:08.960
Bye-bye.