Jive and Cowpaths

Dr. Victor Lee:
All right. Good morning everyone. Welcome to day two of the Data Quality Theater at Clinical Architecture. It is my pleasure to introduce our first speaker of the day, John Lee. He is a real dynamo. You’re going to like this session. Just a couple words about Dr. John Lee. He is an EPIC whisperer. He’s got a long list of accolades. He’s actually been, he’s a practicing emergency department physician. He has been the Chief Medical Information Officer at two different healthcare organizations and a few years ago at an organization called the Association of Medical Directors of Information Systems. He was voted the physician executive leader of the year. He’s a real renaissance man and I hope you really enjoy his presentation on Jive and Cowpaths. Dr. John Lee.

Dr. John Lee:
Thank you Victor and thank you all for attending. So what I’m going to talk about is Jive and cowpaths and you may wonder what the heck is that? And those are actually the real barriers that I see in implementing healthcare IT in the environment in which we work. Alright, so who am I? As Victor said, I am an Emergency Physician. I’m also an Informaticist. I spend about a third of my time professionally, actually clinically treating patients in the emergency departments. Another third, helping the organization in which I work, do some IT solutions within EPIC and then another third of my time consulting with other organizations. But ultimately at a high level, what I want to do is I want to try to improve the system. Certainly I see patients in the emergency department coming in with their lacerations or appendicitis that needs to go to the OR, MIs that need to go to the cath lab. But more often than not, and increasingly so I see people who slip through the cracks, the patient who can’t see their primary physician for two months. So come into the emergency department and have basically an entire workup done in the emergency department. That is an entirely waste of the system. And as an Informaticist, I see enormous information, gaps in delivery that actually can help solve the systems, the problems that we have.

And I think we all recognize that the system is broken. And I think hopefully most of the people here on the floor understand that the reason why the system is broken is because we don’t have transparency. And the I in HIMSS is Information and we need to have better information delivery. And there’s all sorts of common barriers that people cite EMR, stink people who clinicians have all sorts of pajama time dictating and completing the notes at two o’clock in the morning. You have a payment model that’s fee for service. And value-based care is really difficult to define interoperability where data does not flow. If I see a patient in the emergency department, sometimes the providers who are their primary care providers don’t see that sort of information downstream. You have vocabularies which are highly imprecise and the ones that we use the most like say ICD or CPT, do not really encapsulate the true semantics and the accuracy of what our patients are.

They don’t paint a good picture. And then there’s entirely huge gaps of vocabularies that don’t exist at all. For instance, medical error and harm. There is no vocabulary for that, there’s no standard vocabulary for that. And believe me, I’ve been searching for one for the past half decade or so. But these are all kind of structural issues. What I would contend is that there are actually two more common issues or two more fundamental issues that are rooted in actually human behavior. And that’s what I call Jive and cowpaths. And I think you’ll recognize some of the themes, cowpaths you may intuitively understand. But what the heck do I mean by Jive? And that term was inspired by this scene in the movie Airplane. Some of you, are you guys old enough to remember the movie Airplane? So there is the scene. So just to preface it for you, just to remind you, there is a food poisoning incident in this movie and it’s a spoof, it’s a comedy. And these particular passengers are in pain because of the food poisoning incident and they need help. And so every time I work as an informaticist, I think of this scene and I think you’ll know what I mean, especially if you work as a Clinical Informaticist.

Okay, so just to frame this out for you a little bit better, we have clinical operations who are in pain, they need something done, you have IT coming in, I can help. And they technically supposedly speak the same language, but they don’t understand each other. So you need somebody who speaks Jive to come in and help with the process. And in my case, I’m an informaticist, this is what I do. I speak IT Jive to the clinicians and I speak clinical Jive to the IT folks and hopefully I smooth things out. So what are some examples of this? And this is certainly not exhaustive, but I think specific examples really are helpful to explain this sort of picture. So I remember sitting in a order set meeting, we received an order set request from orthopedics and it didn’t make sense to us because it looked exactly like all the other pre-op orthopedic order sets that they hardly ever used with a handful of exceptions.

So we then put it kind of on a back burner, put it in the queue, not prioritized. So they were really upset at this. So then we had to have a meeting. And so we asked the classic informatics question, well what problem are you actually trying to solve? And as it turns out, what was happening was these hip fracture patients, they were being admitted through the ED going up to the floor, getting their or scheduled. And then there was a whole bunch of stuff that did not happen that then delayed the patient from going to the operating room. And there was this mat scramble to get all this sort of stuff done and there were actually delays and canceled surgeries as a result of this. So when we boiled it down, the issue was that the patients were dehydrated because they would go up to the floor and nobody would order IV fluids.

They needed an MRSA test, which is a bacterial test to go to the operating room. And oftentimes what would end up happening is the patient would go to pre-op holding the OR staff would be scrambling around, swabbing the patient’s nose hand, delivering the swab down to the lab and then it would just be a mess. Cardiology patients, patients who had cardiac problems had to have cardiac clearance. And so that had to be determined ahead of time. And I had heard of cases where a patient would go up to pre-op holding and they would call a cardiologist to talk to the patient and then give their blessing or say, no, you got to cancel the surgery, we got to do all these other things to tune the patient up. And then on top of that, these patients were oftentimes in pain and the femoral block that would take them out of pain was often delayed.

So once we heard this, we realized, we explained to them that your problem, your solution is not going to solve the problem. That order set is not actually even going to be entered until the morning of the surgery and we actually have a better solution. What they didn’t know and the gap that existed in the communication is that we actually admitted all our patients through ED admission orders. And then on top of that, what they also didn’t know is that we could add decision support that seamlessly integrated all the requirements into the order set. So what ended up happening was we ended up putting when a patient, this doesn’t happen with all admitted patient, but it just shows up automatically as a result of the patient having a hip fracture. So the MRSA test is automatically ordered. We prompted the end user to hydrate the patient.
And we made this optional because sometimes it’s medically contraindicated to give the hydration. We prompted the end user to order the femoral block and we prompted if the patient had a cardiology problem on their diagnosis list, we prompted them to consult cardiology. Problem solved. All these delays were largely solved. And this was the Jive that occurred here was that there was a gap in understanding of what the system could do and how we operated, we bridged that gap and we were able to explain and solve that problem. So another gap that often occurs is semantics. And I think this is near and dear to Charlie’s heart, so, so for those of you don’t know, in our EMRs we often have problems with these enormous problems, so enormous that they’re completely unintelligible and completely useless. I think the largest one that I ever saw was 80 problems and you can’t make heads or tails in it.

So in EPIC there’s a way to actually organize these and clump them together so you can just at a glance understand the patient has a cardiovascular problem, a neurologic problem, gastrointestinal problem, and a urologic problem. So this requires a grouping, a value set to be implemented so that these diagnoses could be brewed properly. What we didn’t understand initially was that as we implemented this, some of these things, even though technically speaking the definitions are correct semantically, they did not match up with what people thought of them as so clinically. And one of the examples was hemorrhoids. So we started getting complaints, Hey, why are hemorrhoids showing up as a cardiovascular?

But if you think about it, and it was based on the SNOMED ontology, hemorrhoids are a perianal vascular congestion or venous congestion. It’s a venous problem that feeds up into being a vascular problem that feeds up into being a cardiovascular problem. So what did we do? We took that and we actually injected the actual true semantics into this infrastructure. And then we were used some Boolean logic then to then manifest hemorrhoids as the thing that everybody understands it. It’s a gastrointestinal problem. You don’t go to a cardiologist to solve your hemorrhoids, you go to a gastroenterologist. So again, so a gap in knowledge, the technical specifications don’t necessarily match up with the real world understanding what a term actually is.

So this is a big thing and definitions. So for those of you who don’t know, there’s this big national push to try to improve sepsis outcomes. And part of that, the metrics associated with that are attached to when did we recognize that the patient was septic? And there are some fairly well-defined definitions of that. And that depends on cs, which is a constellation of vital signs. That’s actually pretty easy because most of that is pretty discrete. But then you have two others, recognition of infection and recognition of end organ damage. Those are really, really fuzzy. So how do you actually define that? And what ends up happening is that there’s a ton of human interpretation, manual abstraction. And so you get a lot of variable results and as a result you get a lot of difficulty in actually implementing these solutions at scale and actually measuring what works and what doesn’t work.

So we’ve been stagnating a lot on the substance and this issue, this issue of definition, something that is conceptually understandable, but translating that into something that’s discreet, definable and scalable is huge. So the issue would be similar to saying, okay, you’re going to run a hundred yard dash stand somewhere over there and then whoever gets to this finish line wins. So you can see where the problems would occur with that. So this particular issue is one of the most vexing problems that I think that we have in healthcare. It is, and it reminds me of the Supreme Court case. GI versus Ohio in the 1960s was a owner of a theater, I think in Cleveland. And he imported gasp this racy French movie back in the 1960s. And someone said that it was obscene and so he was actually prosecuted and he was facing jail time because he showed an obscene movie.

So it went up to the Supreme Supreme Court and the justices watched this movie, and this was a famous quote from Potter Stewart who’s the Supreme Court justice at the time. And this is the famous, I know it when I see it, and it’s not that. So the problem is in healthcare, we do a lot of, I know it when I see it, I know when sepsis starts or I know what a social determinant is, I know what value is, I know what error and harm is. Heck, I know what meaningful use is, right? So meaningful use was like a two sentence or section of the Recovery Act in 2009 and it turned into thousand pages of regulation because we had to define it. And off to the right is a graph from a company of Evidation and they did this study on all the different potential pieces of data that could contribute or be a detriment to a patient’s health.

And what they found was that only about 0.03% of the data that actually contributed to a patient determining a patient’s health was actually truly measured. All the other stuff was just potential stuff that could be measured but is not. It’s ephemeral. It goes poof off into the ether. And in particular I think about social determinants with this. Does the patient get out and walk? Did they see sun? Did they spend 12 hours a day on social media, just clicking stuff? So none of that stuff is actually measured in any sort of meaningful way in our healthcare system and our IT structures. So without that, I would contend that we can’t make any progress and to make any progress with this, we have to start making some headway with definitions.

So the Jive that I’m speaking of then is that there’s this disconnect between what clinical operations wants to do and what we actually can do. There’s often a disconnect between what the technical systems and how they’re structured and how they are implemented, how they actually do they mean anything to the end users. And then we have to get a wrangle. We have to wrangle this issue of system definitions to make any progress. So we need to be able to speak Jive, we need to have informatics or whomever else be able to speak the IT Jive to the clinicians and the clinical Jive to the IT folks. So cowpaths you may already be familiar with, but it’s the phenomenon where you just take the easy path because it’s easy and you’re familiar with it. And that’s human nature. And I don’t think anybody really knows where it came from.

But the story that I like best is that it came from Boston. So Boston, 16 hundreds agrarian colony of the British Empire, lots of farms, lots of cows. So they start running pathways because that’s where they trample. Boston becomes a city, they need streets and roads. Well are you going to cut through the trees and the grass and the brush or are you going to put roads where things are already trampled down? And the reason why Boston resonates with me is I’ve tried to drive in Boston and anybody who’s actually tried to drive in Boston knows what I mean, not really well thought out, but now it’s too late. It’s what has been built.

There’s two books that I think illustrate this sort of phenomenon really well. The first is Clayton Christensen’s Innovator’s Dilemma. And in it he talks about sustaining versus disruptive technologies. And I would contend that if you look around on this floor, the vast majority of the stuff that we see here are sustaining technologies. One of my good friends, Dale Sanders asked once, well, are we paving paths and roads to places where we want to go to? Or should we actually make the effort into building new roads? And it’s harder, but that’s where we need to go.

The second book is Eric Olsen’s Second Machine Age, and he illustrates how when they made the transition from steam to electric power in the late 18 hundreds and the early 19 hundreds, factories because of the physics of steam power had to have these enormous power plants and they distributed power out through belts and cables to all the places where they were doing all the work. And so when electrical power came on board, they did it in the familiar way that they were familiar with. They created an enormous electrical engine belts and pulleys out to the periphery. But then somebody after a while started realizing, hey, we don’t have to create electrical motors and motor plants, power plants this way and a textile plant. We don’t have to have this enormous power plant and distribute energy up. We can have small sewing machines. But that transition took decades and he illustrates a few other examples.

When we had computers in the 1980s, it wasn’t until the 1990s that we started doing productivity gains, apple and iPhones and keyboard less phones. These sorts of things take a while to get ingested. And that’s what I think we’re experiencing here in healthcare. What are some examples of stuff that I’ve personally experienced where this sort of thing has occurred? I’m involved in a project right now where we’re doing some research recruitment and I was asked to help because they were having all sorts of problems finding these patients in EPIC, in the epic databases. So I helped them with the SQL queries and it was better than what they had before, which was kind of almost random. But then I asked them, Hey, have you tried using Epic Slicer er? And I got two answers, what’s Slicer dicer? Even though it had been there for many years or yeah, we’ve heard of Slice of dicer but no one knows how to use it.

So I sat down and typically in about half an hour to an hour, I was able to actually solve their problems and the research coordinators were able to do this query on their own, find the patients and do something that had previously taken weeks or months to do. It’s something that completely changed how they approached, how they do the research recruitment. And then on top of that, I asked them what they did with these lists. They export them into an Excel spreadsheet and send them off to a mail merge vendor and send out recruitment letters to the patient’s physical mailboxes. And so my next stage of helping them is, you know, can do this all electronically, you can create something where you press a button and it sends it to their MIT trade account. So the second example is sepsis. So I talked about the tagging of when does sepsis start?

And so this data Abstracter asked me, Hey, one of the things that we want to try to tag is when did the doctor enter the sepsis diagnosis into the record? Because they were trying to use the patient’s note. And what they didn’t realize at the time was that initially was that the note is often signed hours actually after the documentation occurred. So they wanted some sort of labeling and tagging within the note itself. So I was able to figure out a way, it’s discrete data, I pulled it into the note and this automatically shows up on every sepsis patient in the system. But then I asked them, what do you do with this? And they said, oh, we take that time and put it into a spreadsheet. And I said, the reason why I was able to do this is discreet data, I can just pull it into that spreadsheet for you. And she said, no, I’m just used to doing it this way. And I would actually, one of the things that I actually discovered, and this is not the only abstracter that I’ve had this conversation with, she explicitly told me that they don’t want to make their workflows more efficient because they’re afraid that they’re going to need less fewer of them. And I’ve had this conversation at multiple organizations where I’ve had this in conversation. So it’s actually 21st century example of a metaphor of the Luddites in England.

Alright, so this is for Sarah. There’s all sorts of things that we have with operational improvement that I think that are potential and are emblematic of a lot of the things that happen that are being displayed here on the floor. One of the most prominent is ambient documentation. So if you think about what’s happening here, you take an analog conversation, a dialogue turns into a bunch of discreet tables, do lots of complex linear algebra on them and then spit out a note. Great technology transforming clinical care. But what ends up happening, what people may not necessarily realize is the same thing that I just talked about with the sepsis. That there’s abstracters who come in after the fact and then they look at the note, they pull out data and they use it for secondary uses. So the opportunity here is what if we could take the data out, put it into some sort of other data structure, and then use that data structure for those secondary uses prior auth.

So if you’re not familiar, it’s a big pain point in care nowadays, what typically happens is that you have to first figure out, Hey, I’m ordering something expensive, does it qualify for prior auth? And if it does, then I need to collect all the data, send it off to the insurance company, they process it, and in some time they come back with an answer. And if they decline it, then I have to collect more data and then it goes on and on and on. So that sort of repetitive process, you would think that’s great for a bot, we can apply technology to that. But the problem is that the insurance companies also have their bots, so they are now able to more effectively nitpick your submission. And on top of that, they are now using the data to come up with even more requirements so that then what you end up is a bot versus bot war.

And we really haven’t gotten much progress with that. But if you think about these sorts of data points that we’re collecting, they are no different than the data points we collect in population health. So we have situations where this is an old slide deck, I think, anyways, so what you can do is you can take the potential here is you can take the data points, pull them into tables, and instead of intervening at each ordering encounter and requiring prior auth, you can aggregate that and then grade somebody say, Hey, you did, you ordered MRIs at a rate that was properly with all the different requirements and turned that into essentially a population health metric rather than interrupting the flow of patient care and turning that again into population health tool. And then population health right now, right now, if you ask providers, it is basically care gap. And what we’re not doing enough is actually doing that same sort of aggregation so that we can help identify patients who are at risk for something using data, using the analytics and the tools that are presented to us. And I think we defer, and the cow path that we stay on typically is we want to do it at the point of care. And so we default to that. So we end up doing that a lot, but we don’t end up using some of the more analytic tools to actually deliver care more explicitly.

So the other thing that happens that I’ve observed is that cowpaths and this miscommunication gap that I reference as Jive are tightly interwoven. A lot of times the reason why we’re on the comp is that people don’t know that there’s an alternative or that they don’t know that they have to do things like data definitions to actually implement and execute on some of the strategies that they want to actually accomplish. And simultaneously they don’t ask for that sort of interpretation and try to bridge that gap because they don’t know that that exists. So it becomes this vicious circle. So what we can do now though is work amongst ourselves and work within our own organizations to start chipping away at these issues internally, little by little and highlighting the places where we may need to bridge the gap, where we may need to do more definition work or educate the clinical operations, what it is that the systems can do. And then on top then afterward we can then guide ourselves away from these cow paths that we’ve ingrained in ourselves. So if you want to connect with me, that’s my LinkedIn QR code and clinical architecture wanted you to give a review on this talk. Thank you.

Dr. Victor Lee:
Thank you so much, Dr. Lee. I have one quick question, then I’ll open it up as well to the audience for additional questions. The question is related to some of the opportunities for improving operational efficiency. I’m wondering if you could share your thoughts on some of the biggest data quality problems that you’ve witnessed, whether it’s as a practicing emergency doc or as A-C-M-I-O related to the problems you’ve articulated with Jive and Calas. And I say this because I know that you’ve also been participating in the Picky Alliance. So from a data quality perspective, what are some of the data quality issues that have hampered your efforts the most?

Dr. John Lee:
I don’t know necessarily explicitly hampered, but I think I would, I would characterize it as an untapped opportunity. There are all sorts of, and I think the biggest one is probably social determinants.
There’s stuff that’s out there that we really should be documenting and collecting as discreet data and doing it in a very normalized fashion that we just aren’t, because there’s no placeholder for them, there’s no data definition for them. No, even if we had a data definition for them, we don’t know where to send that and surface that up. So it ends up being kind of a vicious circle because we don’t know where to surface that up and deliver on that information. So we don’t ask for it. And so we just kind of keep going along on the cow path.

Dr. Victor Lee:
And sometimes it’s also unclear whose responsibility it is to document that data.

Dr. John Lee:
And I think that’s where a lot of technology can help in. So I’ve had some conversations with ambient vendors and a lot of that stuff can be collected at the time of care so that as you’re talking to somebody, if you say, Hey, are you taking your medications? No, that sort of documentation may make it to the note sometimes. A lot of times it’s just kind of a conversation that happens between a patient and a physician. But even if they document it, it’s typically documented only as free text so that it’s only really useful to the person who views that text. Now what if you could have that captured as discrete data? Then you could actually have a dashboard of patients who are noncompliant with medication or are not taking their medication properly, and then you can exert some other resources to help them take their medication or maybe change their medication because they’re not taking their medication because of side effects, for instance. And right now it’s the whack-a-mole phenomenon where you have to do it because you can only enact change on a one-to-one encounter, whereas we need to flip the script and do it at a population level.

Dr. Victor Lee:
Great. Thank you so much for an excellent presentation. Any questions from the audience?

Charlie Harp:
Dr. Lee, thank you so much for the great presentation. One of the questions I had is when you’re thinking about the Jive broker in your scenario, what do you do in your organization to affect that? Because what I find is the people that are working with a system, this is based upon working with folks over the years is there’s kind of a workaround culture. I can’t immediately see how to do it, so I’m just going to put together some Rube Goldberg workaround. Are you guys doing anything systematically to try to raise awareness and enable that?

Dr. John Lee:
So I’m biased, and I mentioned it on the slides, but I think the key missing piece in a lot of organizations is good clinical informatics teams because those people can understand and actually anticipate that there’s a better way to do it without even people asking for it much in the same way as that order said they were asking for a different thing, but that then prompted them us to explore and say, Hey, there’s a better way to actually address the problem. And so I would emphasize doubling down on good clinical informatics.

Charlie Harp:
I think that’s a great advice. And I think we’re in a time where between AI and everything else, it’s kind of shifting the other direction. It’s a little bit scary, but

Dr. John Lee:
Shifting the other direction. Which way? What do you mean?

Charlie Harp:
Well, I know a lot of organizations that as they buy COT solutions for their systems, they kind of feel like, well, maybe we don’t need informatics people. And I think the opposite is true.

Dr. John Lee:
Well, yeah, and actually the data quality issue is even more important with ai and I’ve encountered more than one organization where they’re going full in on the AI hype, but they don’t even use simple dashboards and analytics all that well. And they don’t understand that they have to have a basic level of data literacy almost as a prerequisite to using ai.

Charlie Harp:
Absolutely.

Dr. John Lee:
Because otherwise it ends up being kind of an unguided missile, a really powerful unguided missile that can do all sorts of collateral damage.

Charlie Harp:
I’m going to steal that metaphor.

Dr. John Lee:
Yeah, I use it a lot.

Charlie Harp:
Thank you. You’re welcome.

Audience Member:
Another question. Thank you for your presentation, Dr. Lee. So I work with a lot of independent smaller practices. So I wanted to ask if you could build a data strategy for maybe a solo doc or a two doc with a couple of mid-levels where informaticists is an MA that’s been working there for 20 years doing billing and documentation, so they don’t have a lot. So I’m curious your thoughts about a broad data strategy you would recommend for that type of physician.

Dr. John Lee:
Well, that probably is a multi lesson course, but what I would say is, and one of the fundamental underlying principles that I would like people to get from this is that the reason why these things exist, these friction points exist is because we don’t pause, take a breath and go back to first principles and try to figure out what exactly are we trying to accomplish with this? Are we actually trying to accomplish a project or are we actually trying to accomplish delivering healthcare? So for the instance, the data strategy for the small practices, what are they trying to accomplish? Are they trying to accomplish care gap, whack-a-mole, or are they trying to deliver the best care for their patient at a population and at a level of value? And then that spawns the question, well what do we mean by value? And then so, and there may be some conflict with payment model and all that kind of stuff, but I think going back to first principles and trying to ask, what is it that you’re trying to accomplish? Do kind of lean methodology of five why’s. Why are you, well, we want to put in an order set. Well, why do you want to put in the order set well or it’s getting delayed. Well why is your or getting delayed? Well these three things are missing. Well, okay, now you should have said that in the first place. We could have solved that two months ago.

Dr. Victor Lee:
I sense that there might be an entire presentation just on that topic. But I think that’s all the time we have for today. Dr. Lee, thank you for sharing your words of wisdom.

Dr. John Lee:
Alright, thank you. Thank you for having me.

Dr. Victor Lee:
I got to take off. Thank you again.