Transcript:
Jordan Cooper: Hello. We are here today with Sameer Sethi, the senior vice president and chief data and analytics officer at Hackensack Meridian Health. Samir, thank you for joining us today.
Sameer Sethi: Jordan, thanks for having me. I’m really excited for this conversation.
Jordan Cooper: So for those who don’t know, Hackensack Meridian is a health system headquartered in Edison, New Jersey, with 4444 inpatient beds at 18 hospitals with 11,000 primary and secondary care providers. Today we’re going to be speaking about different kinds of data analytics process automation. To start, Samir, would you walk us through the four types of areas of focus that you have? Hackenzac Meridian as a chief data and analytics officer, and then we’ll just run from there into different process automation projects that you’ve been working on.
Sameer Sethi: Yeah, absolutely, Jordan. Thank you. So, yes. So there are four areas that me and my team help deliver from a functionality perspective. One is, and I tend to divide this into buckets. The first bucket is descriptive analytics. This is where a team of folks within my team help ingest the right kind of data, do quality checks on data, make the data available, and then start to show the business what has happened and at times why that’s happened. The second portion of the team is our predictive health team. These are the folks. Then they take that large amounts of data, starts to look at trend, and start to detect early, detect or predict things. This actually involves artificial intelligence as well. The third is robotics process automation. This is a group of folks in my team that write code that emulates human function versus emulating human thought, which is AI. So that’s the robotics process automation folks. Last but not least, we have a group of folks in my team that build or write software, and this is where we start to write software. So these capabilities, whether it be robotics, process automation, or AI, or data and analytics, start to become available in a way by which people can consume it. The right kinds of messages are being sent to folks at the right time, so it’s most meaningful. So that’s the software development piece or the glue that brings it all together.
Jordan Cooper: So is the software development team kind of integrating different vendors solutions for the descriptive analytics and predictive and robotic, or are they building native solutions to hack and sac meridian?
Sameer Sethi: It’s a combination of both. Right? At times, we get software that doesn’t integrate well with our other software. So we become, this software development team becomes the. At times, we actually do what’s called full stack development, which is everything from soup to nuts. Another variation of this is our predictive health team would build an AI use case, which is giving a predictive signal, for example. And we already have a software, for example, our EMR, we use Epic at hack and sack marine health, and our doctors use that. So they want that signal to be. To be provided to the doctor within their clinical workflow, which exists in epic, another software. So this software development team will build mechanisms to deliver that signal at the right time. So that’s another variation of software development.
Jordan Cooper: So you’re referring to your team triggering custom best practice advisories, bpas.
Sameer Sethi: That’s right, yes. So that’s what we do. So, for example, our team built a capability by which we were, you know, I guess, in essence, predicting mortality in patients. Right. And the reason we were doing that was because we wanted to trigger the right use of end of life care, palliative care, or hospice. And the way we did that is through those, what you just mentioned, which is a BPA. So we would take the signal, which was built in our AI capabilities on our Google platform, which is in vertex, and then we were taking that signal as a BPA into epic.
Jordan Cooper: Got it. So what would be the data source for that signal and how would you integrate it so that it appears natively in epic?
Sameer Sethi: Yeah, so a couple of things. One is the data source is data that we have collected about hundreds and thousands of patients that helps us build a model together which starts to predict mortality. In this case, we then run the model within our Google platform. We call that ecamm. That’s the name of our platform that we have internally developed. And then we send those signals as a BPA, as you mentioned, into epic. So the workflow that we solve for is a physician, and I’m using an example here, is prescribing pain medicine for a patient as a part of that workflow. As they prescribe the medicine and put that order in, this BPA pops up right within epic, and it tells the patient, it tells the provider that there’s an x percentage of chance of this person passing away. So you should consider palliative care, consultation by a palliative care or hospice care team. And the way that’s done is that’s also in order. If the physician or the clinician accepts that, as soon as they press, yes, another order is created. So now what’s happened is two orders are created. One is for the pain medicine and others is for a palliative care consultant.
Jordan Cooper: Got it. Because this is healthy data podcast, a lot of our listeners are very interested in actual data that drives all of these different processes. So I’m really interested in delving deeper into how you actually build the model. Are you populating that model with CRM data? Is it marketing data? Is it claims data? Is it data from referring providers? How are you building that model?
Sameer Sethi: So it’s mostly, I would say, 90% to 95% of the data comes from our EMR, right? So, Epic, that’s where we know the condition of the patient, the demographics of the patient. We also have trend data, just not of the patient, but almost. So, for example, the model that was, that we are using here uses almost hundred attributes, right? That come to conclude on a signal. So 95% of is coming from the EMR. There is some data about the patient that we host in other parts of, you know, in other systems as well, and those are utilized, but mostly 95% is all at the EMR.
Jordan Cooper: Got it. And the descriptive analytics team, they deal with data ingestion and they’re doing quality checks on the data. What sort of data sources are they ingesting and how are you able to ensure that it is clean data?
Sameer Sethi: Yep. So the discrete Linux team and a portion of that is what I call my data engineering or data ops team. This is the folks that are involving, are responsible for ingesting the data, normalizing data into a data model. Our larger source, as you can imagine, is again EMR. So epic is our largest data. We have data from Peoplesoft. We have a lot of claims data that we use, especially in use cases where the data doesn’t have to be as real time. But as you may know, and as your audience may know, claims data is extremely rich, especially when we start to look at data that is within our value based care contracts, because we see the lineage. So we use a lot of data after we de identify to build these models.
Jordan Cooper: That’s excellent. I’m glad that you brought up the value based care contracts. It’s something that’s facing many health systems across the United States. And as health systems are looking to manage these populations, they’re looking to do exactly what your team does. They’re looking to build out a team that can identify trends and identify the high risk populations and manage the population health and take a share of that shared savings. So how is your team? Can you walk us through a concrete use case of how your team has definitively added value to Hackensack Meridian for its value based care contracts?
Sameer Sethi: Yeah, absolutely. So I think a lot has to do with perhaps if you go to the basics, which is preventive care. Right. This is where we start to go after the value based care dollar. So depending on the condition, whether it’s, you know, that somebody having, you know, some, somebody that’s a smoker and making sure that we get their. We have lung scan in place or, or a female that, that has to have a mammal done. So on the preventive side of the house, you know, we have to build various models to make sure that not only is there preventive, you know, all this preventive care actually happens, but also there’s proper reimbursements for it, right? So a big portion of what happens with value, value based care, there’s an intent incentive attached there. And if you don’t, of these incentives, you know, we lose our money or leave money on the table. So this is a double edged sword, is what I call it right here, if it’s not done correctly. So we need to make sure that we address both sides of this. So what we do is that we ingest data, again, from claims, you know, because at times, those procedures happen outside of our health system. Right. But we want to take advantage of the signals that are and the data that’s available within that claims data. We fuse it with the data that we have for the patient within our EMR, and we try to create the next best action. In this situation, the next action would potentially be calling out a patient and saying, hey, you need to come back here and need to do a chest scan, a mammogram, or a lung scan.
Jordan Cooper: So I appreciate that concrete result. And I want to bring us to a related topic that I’ve heard you mention on other previous podcasts, which is a topic of reporting. You do these analytics, you say, all right, these people need these preventive measures, maybe will trigger an action, but maybe we need to create a report for a specific audience. Now, I think I heard you said that there are about 17,000 assets in Hackensack, Meridian Health’s library, and some of them are not actually used, and there’s different maturity levels to the data. I’m wondering, as you’re trying to produce, for example, this, the integration of claims data with Peoplesoft data, with EHR data, who are you developing these reports for and how to ensure that they’re actually used by the intended audience to actually drive value, since many reports, as you might mentioned in previous conversations, are not actually. They’re created, but then they aren’t. They just sit there in the library.
Sameer Sethi: Yeah, that’s a great question. So the 17,000 number is something that I have used in previous conversations, and unfortunately, that is what I inherited as a part of taking this position as a data analytics officer, and this happens, by the way, I don’t think anything bad or wrong was done with creating these 17,000 assets. But Hackensack Median Health has been doing data and analytics for quite some time. And while hindsight is 2020, perhaps things could have been better and we could have limited that to less than 17,000. Our goal at hack and SAC now, as a result of us creating this data platform called eCAMM that I mentioned, is to start to harmonize what the needs of the organization are. So what we are doing is by virtue of us understanding the business needs one, and also by virtue of us watching which of those 17,000 reports are being used, we’re coming to a common ground of what truly the business needs. Our intent is to make sure that we don’t have to maintain those 17,000 reports. Our intent make sure that those are somewhere from four to 800 assets that we have to maintain. And the process of that is us sitting with the business, getting a sense of what their actual needs are, understanding why they need it. Look at the logic behind how those things are calculated and build the right asset for the right person. Right. And honestly, if this becomes 17,000 assets, it is what it is. I don’t think that’s the case. That will never happen. I hope that’s not the case. This has to be a lot less. It has to be most meaningful. It has to be customized to the needs of the business today versus the needs of the business before.
Jordan Cooper: So for the risk bearing relationships, the shared savings programs that Hackensack participates in, whether it be an ACO or whatever the model is, what have you found? Your business stakeholders have been clamoring for the most in terms of reports, some listeners, or at other health systems, they’re thinking, well, we’re in this value based risk sharing contract. What are the assets that are most valuable to your organization? May be helpful to a listener.
Sameer Sethi: Yeah. So I think it really comes down to making sure the measures are defined and obviously turned into code the right way. That’s what I see as a missing piece. Right. What the audience is looking for is when they say, and I’ll just use an example of not something that is ideally related to value based care, but let’s talk about length of stay, which the whole of healthcare talks about. Right. I think using that example, what the business is asking for is a definition of the length of stay to say, when you say length of stay, what does that truly mean? When does it start? When does it end? What does it involve. Right. What we have done at Hackensack is we have heard, we have heard of all the different definitions of length of stay, and it’s okay to have a few. I understand how it’s to come to one definition, but we have done is we have taken all these measures and agreed on it and defined them, and that’s how we put those reports out there. And that’s when these reports and these assets start to become most meaningful. So whether it be for value based care or any other group that’s out in the team that is doing some great work, what we have made sure is that not only do we provide the data, but the measures that calculate and use this data are appropriately defined and are very clear cut and available for people to use. That starts to create a lot of, you know, organization around. When we say that this has happened, that’s what it means.
Jordan Cooper: Interesting. So it sounds like governance processes are being used to create consensus, which is what drives the value of the reports that your teams create.
Sameer Sethi: Absolutely. So we. I, as somebody that’s leading a division, this is one of my KPI’s and one of those KPI’s, the way I measured at the end of the year is to say how many measures I have in all the reports and assets that exist out there and which of them have been agreed and defined on. And we keep, and keep your watch on that and in full transparency. Some, you know, some months, those are really good, and some, and as a result of what the business needs, at times they lower. But that’s how I am measured. That’s how I’m held accountable by saying, not only am I putting the right assets out there, but these assets have proper definitions. These measures have proper definitions that have been agreed on with the business.
Jordan Cooper: Sometimes there are measures that are very clearly defined, for example, by National Quality forum, by NCQA. There are Hedis measures or measures that are clearly defined by CMS, and some are process measures and some are outcomes measures. How do you take those different measures and I guess tie them, I guess pull them into a library? Let me put it another way. Some of these measures require your team to pull information from ehrs in different ways. Have you ever come across a measure that has consensus, but you’ve had difficulty getting the actual numerators and denominators for that? How have you grappled with collecting the right kinds of data to satisfy the requirements of particular measures?
Sameer Sethi: Yeah, so I. So there are sources, as you know, of the definitions of measures, the struggle is coming to a consensus that’s the hard part. Right? So all these organizations come up with what the measure is, why it’s there, what the numerator, to your point, is, what the denominator. It is what we have done at Hackensack, and this was the first thing that we did, is we set up a measures catalog, and we have connectivity between the report and the measures catalog. So at any point, anybody can, when they go and they’re looking at a number or something that’s calculated, they can click on it on top, either by virtue of a hoverboard, which shows or at least a brief description of it, or they can click in it further, and it shows the metadata about the measures, which says what the measure is, what the numerator and denominator is, who is involved. When was it approved? When the last time it was renewed? These are the steps that we are taking. It is, by the way, and I think to your point, point, it isn’t our measures all the time. We are using measures because we have to from external entities out there that develop these things, and we have to follow those things. So this is a common. Our measures catalog is a combination of the measures that we have put together and agreed on as an organization or measures that we have had to inherit, you know, by virtue of just being a part of the healthcare continuum.
Jordan Cooper: So, Samir, we are approaching the end of this podcast episode, and with that, I’d like to make. Pose a final question to you. If you could think of one request that you could pose to Hackensac Meridian Health chief Executive officer, chief information officer, the board of directors for Hackensack Meridian, someone in leadership, and have this request granted so as to benefit patient safety, so as to improve quality, measure outcomes, improve the profitability of the enterprise, you know, what would be that desired goal and what would be the request?
Sameer Sethi: Yeah. So I think it’s a very deep question, and thank you for asking that. I think about this quite a bit. And, you know, for me, what’s missing at times is the push to change a process. So, as I mentioned through all my buckets, all of these are more technology enablement. Right. And those are very important. My ask would be for us to, when we take technology and technology enablement to patients, to organize that, we are empowered to work on changing the process. So whether it be the patient or the provider or the nurse or the back office staff, that they inherit this technology in a way that is most meaningful to them. So what, at times, what happens is we put great technologies out there, but we don’t focus on changing processes or making the technology a part of the process. Technology is a part of the solution. Data and analytics, AI, robotics, process automation, all this is a part of the solution, but it is not the full solution. What has to come with it is the people process. So as a, you know, going back to what we all have learned, people, process and technology. So my ask of the organization and I think all organizations out there would have be help us change not just technology, help us, but let’s bring the process into this so we can deliver technology in the right way. Otherwise we’ll build 18,000 assets that will not be used. Right. It’s going to happen.
Jordan Cooper: So for our listeners, this has been Sameer Sethi, the SVP and chief data and analytics officer at Hackensack Meridian Health. And we covered a lot of ground today, Samir, and ran through data analytics and ran through governance and measures. I think one interesting takeaway that you just mentioned is not just attack, but also humans and workflows and processes that you can have the best solutions, but if nobody uses them, what value is it? So I think we spoke a little bit about governance and ensuring that not only that you have the right measures and you have the data sources to drive those right measures, but you also have consensus that these are the right measures and that consensus is driven to meet and serve the business needs of the organization. So I’d like to thank you very much for joining us here today.
Sameer Sethi: Yeah, thank you so much, Jordan, for having me. This has been great.
