Blog /

INTERVIEW: Prevent Harm and Reduce Waste with Your Medication Data

Publish Date
Read Time
Ryan Carlson
October 12, 2021

In our video health data series, "Leaders in Leveraging Health Data", we chat with Sam Wilson, CTO and Co-founder of Bainbridge Health.

Prevent harm and preserve resources by monitoring your infusion pump medications.

You can allow your clinicians to operate at their peak by utilizing decision-making companies that specialize in data. Sam Wilson, Co-Founder and CTO of Bainbridge Health, shares how they advise hospitals on the best way to move forward by collecting, aggregating, and analyzing information from infusion pumps.

They have tracked over 50 million administrations, so they know a thing or two about medication usage.


Bainbridge Health with Sam Wilson: Prevent Harm and Reduce Waste with Your Medication Data

Ryan: I'm Ryan Carlson with Healthjump today, we're here with Sam Wilson, the chief technology officer of Bainbridge health. Thank you for being here, Sam. Thanks for having me. So this is the discussion in which we're just focusing on. What would you say you do, uh, at Bainbridge health? Uh, tell us a little bit about that.

Sam: Great. So we work with hospital pharmacies to aggregate and fusion pumps. And look at medication administration from the, uh, you know, the, the sharp end of the syringe, so to speak. Um, so we collect the, we work with the infusion pump vendors to collect data from the infusion pumps, aggregate that, and then go back to the hospital, advise them on how to optimize their decision support, to reduce alert, fatigue, uh, to better comply with joint commission standards.

And then as well too, to look at it from a supply chain perspective, how to reduce costs, how to reduce it. And that can have a big impact on, you know, numerous aspects of, of the, the pharmacy operations, not least of which is addressing diversion, uh, addressing, you know, financial waste, um, and, and, or, or, uh, lack of human resources to efficiently produce meds, source meds, things of that nature.

Ryan: Those infusion pumps, uh, you know, having been from Minnesota where all of the med device manufacturers are located, it's true. You know, I had the opportunity to sit in on a lot of local tech conferences where it was my first exposure years ago to this whole world. I had no idea how much the infusion pumps we're in the stone age and how right.

Sam: And now. Pushing them forward to get it. So you have telemetry that you can pull off, but with a, with a device or, you know, anything along those lines. So, absolutely. W what have you, what have you found, you know, in this world of infusion pumps, has anything gotten any better? A lot has gotten better. So obviously with the, with the advent of the smart infusion pumps, where you're actually tracking the data, you're able to aggregate the data.

To where I think the big focus for most of the vendors, the manufacturers is with the integration to the EHR and really trying to get to that sort of real time integration where you've got either auto proof. So the EHR can send the order information to the pump or the auto documentation so that the pump is sending the documentation back to the EMR in the, in the EHR.

So to keep those records and to facilitate that piece, Um, so now you've got these comprehensive platforms where you know, that you can have a monitor at the nursing station to say, you know, which patients are coming up to the end of an infusion, uh, who needs a bag change to somebody, have a, you know, an, an occlusion.

So maybe a nurse needs to go back and check the lines, um, in, in one central location. Uh, so they've really been focused on that. I think were a big pain point is remains is. So now you've got all this data and what are you going to do? From a governance standpoint, how do you align decision support? Um, how do you really dig into this?

Ryan: In all the various and sundry ways that a hospital pharmacy meal might want to investigate? So is this just patients that, or is it like in-home care as well as in the hospital or what you're dealing with is pharmacies in a hospital, right?

Sam: Correct. Right.

Yeah. So we're focused on the acute care hospital space, as opposed to say a home infusions or infusion therapy centers. We do some work. Some hospitals will have a, like a chemo, uh, Ivy room. And we do some work with them looking at, uh, sort of timing of Medicaid. You know, a nurse might write down in a log book, what time or in an EHR, what time, uh, but with the infusion pump, you know, the exact time that two different meds and with the, uh, chemo drugs, you know, 15 minutes off by one minute can actually have a pretty serious impact on, uh, quality of experience or even outcomes for those patients.

Um, there isn't quite as much, uh, availability of. In the, the, the home care setting that's, that's changing with better connectivity, more affordable connectivity. Uh, but that's been an issue. So we haven't really focused on that space quite as much, but we certainly are focused on the acute care. Yeah. I know there's a lot that they're still trying to figure out, you know, at the, at home, you know, as a phone, a gateway, how do we securely manage patient health?

Ryan: You know, so, you know, I, I think that's interesting that you're focusing on where they have that control. Network where they've kind of got things figured out, right. So what is it is the pain that Bainbridge keeps finding that you're addressing over and over again?

Smart or what people say a dumb pump. Yeah. None of them are dying. Right. But like where, where is that? Is that, that, that, um, that issue. Bubbling up that you guys are uniquely addressing.

Sam: Yes. So really what it comes down to is enabling the hospital to have their clinicians operate at the height of their license, as opposed to them being data clerks.

And, and we'd all like to think that, you know, the hospital has an IT department, they've got an informatics team that they can solve these problems. There's 6,000 hospitals in the United States. Why should they all try and solve that problem? 6,000 different times. And you've got an organization like

Deep clinical knowledge of what's going on from a pharmacy standpoint, from a clinical operation standpoint, from a supply chain standards point, we can bring all that together and really enable the hospital to let their clinicians do what they do best. So get them out of Excel, get them out of Tableau.

You know, we can build whether it's a calculator or, or a data discovery tool or our professional services over there. Where we go and do the legwork to, you know, research and, and do the, uh, the documented evidence, um, sort of cross-referencing everything that they need in order to make an informed decision, uh, for their hospital.

And, and so, you know, we want to make a safety change. Well, what's going to be the financial impact of that safety change, or we wanna make a, a financial change. Uh, reduce the, the waste for a particular medication. We're going to do that. We're going to switch concentrations. We're going to switch suppliers.

How do we make sure our decision support is well aligned with that and that we've got an education campaign for the nursing staff so that they can take that transition, um, because none of these decisions happen in isolation, right? Hearing I'm hearing the, the address, the weakest link approach, right?

Ryan: There's also the implementation, the training. I I'd imagine there's some sort of like follow through where you're are you treating accountable?

Sam: Yeah, absolutely. So a big part of what we do, uh, obviously our customers are very interested in making sure that the money they spend on us is, is money well spent.

So we, we review every decision that we make with a customer, uh, over time longitudinally to make sure that the change that we wanted to have happen did happen. So we're, we're tightly integrated with their CQI practice. And, and you know, we're going to make this change. We're expecting this outcome. We come back six months later because we've got our data platform sounds like a scientific method where you make a hypothesis and then you actually put it in and do observation and analysis.

We'll go to a hospital and say, Hey, you should change this concentration to reduce the waste of this very expensive medication. We think it's going to save you $500,000. And we say that to that particular set of clinicians, they go ahead and make the change.

But then if they don't communicate to their buyers, they're still buying that medication and they're just building inventory. All right. And so it's tied into that. I'll use saying now, remember, uh, like, like a good, uh, like someone who does logistics or event management, you know, that, that. Yeah, they're going to go, here's your checklist.

Don't forget this. People often forget a pair of extra shoes and great. We do. We do, we do try to be very comprehensive with our, with our customers. Okay. Try to ask those kinds of questions. We are very interested in expanding into all of these other pools of data that are in the hospital. Um, you know, a lot of the conversation in, I think in healthcare, it is around the EHR is around the idea that there's this central repository.

You know, the, the clinical care information. Yeah. At the end of the day, there's actually vast mountains of data that never make it into the EHR and likely aren't going to make it into an EHR. They're in supply chain systems. Okay. Some very large companies keep inventory in a marble notebooks.

Ryan: Shut the front door.

Sam: I can't make that up. A very large IDM and that's how they manage some of their inventory. So that inventory, it might be an IV management system. In the medication workflow management tools, it could be in the dispensing cabinets.

It could be in their carousel systems. A lot of times those systems are in some ways tied back to EHR, but similar to the infusion. A lot of the source data never actually makes it in.

Ryan: How are you pulling these disparate pieces together? So what makes Bainbridge uniquely suited to solve this problem? Is there something that you do that differentiates your ability to address this versus

Sam: I started joking, like to say it's that we have a high threshold for pain, but basically we integrate with all of those systems.

So we pull the data from those differences. And bring it all together in our own clinical data warehouse, that's very similar to data warehouse at the hospital might have yep. Very focused on medication administration and the medication supply chain side, as opposed to, you know, uh, tracking patients, managing billing, uh, you know, managing, uh, you know, uh, patient transfers and things of that nature.

The EHR is really good at that piece of the pie. These other operational systems, you could ask the question, why aren't you querying EHR for the data that you're finding in Workday? It's a very similar problem. There are these other operational areas within the hospital that aren't fully integrated.

Some hospitals have tremendous data warehouse platforms where they've brought all this data together. But even in that case, that data is really managed and owned by a group of people that are experts in data. They're not experts in the specific clinical unit. And so we really free that data up and bring it right to the specific clinical uses where I hear a lot of data information assists, where I'm like, all right, COVID hit.

What changed? What growing pains just got amplified. Right. And they just said, we just have more work. There's no downtime. Right. And so they've always got a backlog of just some of it's just like shift reports and all of the stuff. Right. And, and, you know, Uh, I think this is why I hear some people call data lakes, data swamps.

Is it just things get stagnant on used, you know, like, no, one's got the time, uh, when you're in charge of everything, right. You're really in charge of specializing in nothing. Right. Right. And so you just have all of this information. So what I'm hearing is you'll leverage these existing data. Go out and you know, where, where to shine the light and, and bring that information.

There's always a new buzzword, right. And this isn't exactly a new buzzword. But if you're just getting into data lakes, well go one step further into it. I've heard the data lake houses.

I recently moved into my data lake house. But no, the data mesh, or the idea of a data mesh is really starting to gain some traction. And part of the idea is, you're going to have these different systems that are going to produce this data. And then you're going to have these various, domain specific applications that need to consume that data, rather than trying to dump everything into one massive cesspool of unmanaged data.

Right. Or centrally managed data with a group of people and resources that are very finite and can't expand, take a services approach, and really try to build a platform where you can begin to compose these different aspects of the system together. And so this is really a part of that data mesh. Okay. So, so the mesh though, Not bringing all the data, but just the reference point.

So indexing to the different sources of how to access it or more of a mindset for how you're going to manage data within an enterprise. So we're not a data mesh providers. This is a philosophy, was it was more of a management philosophy of data. Right. I think you're going to find there are products out there that, that, that really try to productize the concept of a data mesh.

Sure. But I think if you use it more as a strategic sort of mental model for how you might think about data within your area, Whether it's a hospital, large manufacturing or retail organization, uh, you know, or university, whatever it is, you're going to have lots of different data sources. You want to bring all of that together.

The idea with the data lake is to centralize it and have one central tool or group of folks. In this case, it's really have good governance have good standards and, and then really try to make that data accessible to them. Domain experts that can then put it together. And those domain experts are now as in house.

So if you've got this data mesh sort of strategy in place or the desire to create one or, or the desire to create one. Or if you want to sort of understand where an organization like ours kind of fits in.

Yeah. It's really, from there, we have some of our customers that take our data and they bring it in to their data, their data lake. Yeah. We transform sort of that raw data into something that's more clinically relevant for the institution. Now we were experts in interpreting that we're experts in interpreting that and so we can provide the professional services over the top, whether it's clinical or analytical to really improve that. But we can also take sort of our sort of secondary data and bring that in to enable whatever. Uh, analysis they may, may want to perform.

Ryan: So as the secondary data you're bringing in, is it enriching existing data sets or is it just providing a tool for prioritizing or, or, or the cases?

Um, in a lot of cases, there's some enrichment. We don't generally bring our enriched data back in for various reasons. I can think of about a hundred, but one way to think about the raw data that we consume is it sort of like web click data, right?

So, or clickstream data and, and you can take that clickstream data and you could load it directly into your data lake. And then over the top of that, you can build various, analysis on top of that to try to understand, uh, extract individuals, whether it's a clinician or a stand in for a patient.

Most of the data we have is not identified. And we don't reidentify the data. What we really try to look for is anonymous sort of like series of activities that might represent a clinician's interaction with a patient, um, to then derive some further statistics on that to say, well, what is your override rate?

What's your alert, salience rate, you know, some of these higher level statistics, uh, what's the result in action. So, uh, a clinician gets a decision support alert. What did they do? Right. So, so I'm told, Hey, uh, that's too high. Did I override that or better yet? Did I cancel train reprogram the same thing?

Get the same error. Did I do that four times in a row and then ultimately override? Or did I just switch to a different drug that wasn't going to alert and now try to do that really messing up the data stream. So I'd imagine so he can troubleshooting right. And that's every time you've tried to program something, you realize that you'd actually saved the second.

Right? If you, if you've, if you've ever been in a hospital, sat next to a patient bed, while two nurses are struggling to get an infusion pump to, to do what they want it to do to get really creative at certain point, because they've got a patient there that, that needs the medicine. Yeah. And they're not going to let the infusion pump get in the way of that.

So, so we can start to look at the data and try and identify those situations. And, and sometimes the solution. Hey, look, let's make that decision support better. Let's make the training around that decision support better, um, or, you know, what's maybe addressing an individual practice problem that that needs to be addressed.

Ryan: Sure. Are you ever identifying issues that are hardware related that is valuable for the manufacturers? I know that might be a loaded question.

Sam: Uh it's it's not, it's not all a loaded question. I just have to think about how I can answer that question. Um, so the device manufacturer. Have to go through FDA approval and sometimes they need information, uh, from outside their own sort of, um, sort of device use where they may need to understand, uh, how infusions are used generally.

Yep. Um, and so we can provide them with assistance there. Uh, we are providing some, uh, device manufacturers with, uh, what I call sort of our platform services, where, you know, Hey, we're experts in this. We understand this data and we can manage it maybe better than that. Um, or, or maybe we can manage it, uh, maybe not better than they can, but maybe, uh, in a way that's maybe a little more flexible from like a product resource perspective.

Do they want to spend their time, you know, with their product people on, on sort of this aspect of it? Or do they want to spend their time focusing on, you know, really making sure that the connection with EHR is exactly what it needs to be. Yeah. I, I see a lot of parallels between like industry 4.0 and the smartphone.

Where an initially a lot of the sensing and tracking at first was, oh, so we can now hold our vendors to, like, we can prove that this anomaly is occurring, right. An unintended, not consequence, but benefit was what they first thought was going to be. Now we can prove it and they're gonna, you know, like, right.

Prove the problem. The manufacturers, even one step further removed would say, could we. Could we work with your factory and yeah. And actually get that data because we should make an actual improvement yes. On the next rev. Absolutely. And so it was really cool to see that one level of abstraction where the manufacturers themselves were not part of that initial conversation about accountability.

Exactly. When they could then start seeing. When you, when you think about a manufacturer who typically sells through distribution is they're the having a customer contact. So manufacturers oftentimes are just, I, I hope everything's going right, right, right. But they don't, they lose some of the contexts.

So, uh, correct. I don't know. I'm, I'm seeing health. Really following a lot of these trends are like maybe five, six years trailing behind other industries that have kind of proven out the model. Yeah. And, and built enough tolerance for a risk around security, wireless, you know, encryption. And, you know, I mean, in some ways, you know, it might be a question around, uh, sort of the, the hospitals.

Uh, having a better appetite for risk, but it might also just be that those technologies have matured to the point where now they can meet the risk profile that the hospital has. And that's certainly, I think in the context of COVID, uh, been been something that a lot of the conversation is really focused around sort of health, the obvious pain points of the pandemic, right?

There's sick people, you know, different, different, uh, uh, you know, uh, acuity for the patients that are coming in, it's impacting our supply chain, all of those kinds of things. Uh, going on at the same time. Uh, there's, there's a, there's a cyber war going on out there that we don't talk about a lot. Um, but especially around vaccine development and collecting some of this patient information, a number of our hospital partners, we're certainly a part of, uh, I think, um, you know, that effort and themselves became, you know, much greater targets than they ever were in the past.

I mean, they're already big targets, but this certainly raised their risk profile. Yeah. Um, I think, you know, I think every hospital CIO, CTO or security risk manager has the same nightmare of, of, of getting a call in the morning while they're enjoying their coffee. Uh, I'm gonna, you know, I'm going to do patient in this room because I can control the infusion pump remotely.

Everybody's terrified about that. Yeah. So security is absolutely at the forefront of everybody's mind. Yep. And, and securing this data is absolutely at the forefront of everybody. The data and using the data, uh, to, to understand the risks, whether it's, you know, device failures, whether it's, you know, better managing the medication, being able to use that data to, to drive those decisions is fundamentally important.

So you can't throw the baby out with the bath water, so to speak. You've got to have the data in order to drive the decision-making and, and, you know, I think one of the nice things about how we've really tried to build and model our platform. Yeah. To really acknowledge what we do and what we don't need.

You know, we don't need all necessarily, uh, you know, patient information. We don't necessarily need some of these other things. We can really try to manage that risk and, and really work with the hospital to how do we implement this in such a way that we're respecting and managing risk for you? Yeah. But still empowering you to make decisions with that data.

Ryan: And that's where I think there's an interesting conversation. That we've been having at Healthjump is the idea of health metadata. There's a complete patient record. That's identifiable and then it's de-identified. Right. But what if the data that you're pulling never knew who the person was?

It just happens to be a series of factors that could be attributed to a random number, but it's never associated with the person, but it is a condition. Like history of like this med device with the, the, you know, these, uh, health readings. Yeah. And we also know these medications, right? Like, yeah. And then, and maybe even new, like the, uh, the outcome of the patient, but we didn't ever knew it was never associated.

So these are like little breadcrumbs, right. Never could be re identified. It's not the redacted document. That's got the Sharpie, that's mark things out and you kind of hold it up yeah. To the light, you know, I think it's, it's, it's a conversation. I'm interested in, in having in healthcare is, is what, what can we do with these valuable, right.

We don't need all 10,000 record or fields. Yeah. We don't even need all 300, unless we're doing something fairly comprehensive. Right. Sometimes it's just three. Right, right. Or whatever it is that, that, uh, the, the hypothesis, uh, would need to support the actual conclusions. So I'm fascinated. And how we can see that risk profile, whether that's going to also be adapting over time, because data is crucial in making decisions.

And everyone's so worried about the liability that they're worrying about getting that phone call about, oh my gosh, this. So I think some of the, some of the risk management does come down to making recommendations of, well, how about we change how we're getting the data? I think, I think there's. You know, if you think about sort of the, the inter-operability standards we have today, right?

They're ancient name, practice then, right. All you're really trying to address, I think sort of the, the very rough, uh, direct care for an individual patient where you actually need all of them, is that information because it's, it's the, the data supports the care for you or me, or a loved one or my patient.

Right. You know, and I think, I think there's, I think. There needs to be maybe a new generation of, of inter-operability standards that really try to look at and address some of these considerations. Yeah. And when you think about sort of dealing with, with aggregate data, the kind of data that we deal with, I think there's a whole fascinating area.

If you look at the way, for example, apple deals with credit cards, you know, apply that same thing maybe to prescriptions where you can have intermediaries that never know. So sure. On the transactional side of things as a whole yeah. Technology, uh, technological improvements that could be made there, but then when you come back and you think about, okay, now I've got this massive amount of data and I want to do, uh, population studies, or I want to do a sort of machine learning, or I want to do a larger sort of governance types of analysis.

Sam: Yes. I think we really need to start to ask the question questions around Federation and Federation of data so that the data can remain where it needs to remain. Um, but maybe the analysis. And their aggregates can come out and, you know, it's certainly very complicated. And I think the, you know, sort of the, maybe the naive or the trivial approach is to, is to just try to, uh, push up some of those, um, those aggregates and assume that because I've got an aggregate, it's not identifiable.

Ryan: Um, did you know that Cesar Milan, the dog whisper already solved this one? Oh no, no. So data don't tell me I'm barking up the wrong. Oh, oh, wow. Okay. First ad job. Alright. That's great. Uh, so, so, so follow me here, right? You've got your, your, your troubled dog that needs to learn some new behaviors, right? It's a machine learning model that needs to be exposed to a bunch of real live data.

Right. That dog gets walked to the sidewalk. Caesar comes out and takes it and leaves the owner on the sidewalk and takes it into it. Sensed compounds like 80 other trained dogs. They've all been rehabilitated, but they're a healthy pack that has all of the skills, all of the things they need. Six weeks later, they do the reveal.

Cesar walks that dog back out Avar after having been trained by not just him, but the path. Right. And goes here is your dog that is now a new and improved dog that had the benefit of all of these other dogs. But you're not bringing these home right. You, you have no idea who these were, but you have all of the outcomes that you were choosing to have, which out without ever, you know, letting them out of that dogs at rest.

Although that never happens. My dogs are either they're at rest and in motion. So just like data. So, um, yeah, I, I never thought about that. Uh, some data names for some dogs. So how that same idea though, if you're, if you've got models that need to be brought into data. To be informed and learned there. I really believe there are safe, completely secure ways.

Sam: And it's like, I don't think that's the challenge is how do we have the conversation with people who don't understand data in motion, data at rest, uh, how machine learning models are informed, you know, how, you know, what moves, what, what actually, he comes out like great, all these things. So I think that it's, that.

How many years needs to go by how many use cases do we need to see that for, for that comfort level? Yeah. I think, you know, a lot of the technology still has to be validated. I think, you know, you talk about, um, you know, federated learning and things like that. I think, I think there are still some questions from an adversarial stance.

Sam: Um, you know, Hey, new concept. Okay. What are all the ways that might be exploited? Um, so I think that, I think some of that still needs to kind of go through the proverbial ringer, uh, for, for folks to feel comfortable with it. Yeah. Um, so similar to some of these other technologies where, you know, it may have started out in manufacturing and, you know, it's maybe not a question of, uh, hospitals developing more of an appetite for risk, but maybe those technologies sort of getting through the DRAM.

Um, to, to get them to a state where they make sense in the hospital environment. But, but I think there are also sort of industry activities, right? You think about something like the fire standard, right. Very much about point to point. Now, you know, there, there are efforts in there to, to, to try and get aggregate data through fire.

Uh, and, and some of those efforts, what we really need is we need strong incentives, uh, or compulsion for, for vendors to participate in those standards in a holistic manner. Yeah. It's one thing to say, well, there's an HL seven standard for this. It's another thing to say. Okay. Uh, do you do anything other than ADT records?

Yeah. Right. Yep. Do, do you actually, do you actually support anything other than ADT records? You can have conversations with folks all day long about, you know, when patients show up and when they leave the hospital. But if you want to know whether or not the drug they were given was approved by a pharma.

Ryan: You know. Yeah. You know, how well supported is that? And how standardized is that across all of the well deployed platforms that are out there? Well, I think that's actually another conversation that I'd love to have at some point in the future. So, uh, Sam, thank you so much for being here, talking about Bainbridge house.

Sam: Thank you guys so much. This is great. So thank you.

Health Data Extraction

Collect and standardize health data within minutes, regardless of which EHR systems you use

Contact us

Achieve your boldest ambitions

Explore how Datavant can be your health data logistics partner.

Contact us