Join Kiran Mysore, Chief Data & Analytics Officer at Sutter Health, as he shares insights on scaling AI adoption, building sustainable innovation infrastructure, and transforming healthcare delivery through data-driven approaches. Learn how one of the nation's largest health systems is successfully integrating advanced analytics and AI into clinical practice while maintaining governance and ethical standards.
Kiran Mysore, Chief Data & Analytics Officer at Sutter Health
Shahid Shah, Chairman of the Board, Netspective Foundation
Welcome 0:01 Welcome to Digital Health talks. Each week we meet with healthcare leaders making an immeasurable difference in equity, access and quality. Hear about what tech is worth investing in and what isn't. As we focus on the innovations that deliver. Join Megan Antonelli, Janae sharp and Shahid Shah for a weekly no BS, deep dive on what's really making an impact in healthcare.
Shahid Shah 0:29 Welcome everybody. We're reporting here from vive 2025, in the lovely city of Nashville. It's a lot colder than I wish it was, but otherwise, really, really good, and I'm thrilled to have here. Kieran meisser from Sutter Health. Kieran, tell us a little bit about yourself.
Kiran Mysore 0:47 Shahid, it's great to be here, and I'm really excited about life. I'm the Chief Data Analytics officer at Sutter Health, and so my mission is to really make the business of data and insights as easy as possible for all our stakeholders, that includes our physicians, our hospital staff, our administrators, and, of course, all our patients.
Shahid Shah 1:10 And so what are your biggest challenges in terms of so what as we start to use more and more data in AI, as we start to use more and more data by physicians for research. That self service nature of research data that people probably need is likely one of the biggest challenges. But I'd love to hear from you, what are you seeing as the thing that's causing you the most grief? To get the data out into the hands of people that need it most?
Kiran Mysore 1:37 It's a great question when I started with Sutter. And this is not even unique to Sutter. The classic challenge of every chief data officer, analytics officer is to really look at how to connect the data to make sense for their stakeholders. And what I inherited at the time was a lot of different silos of information. And you know, you have your silos in your electronic health record. You have them in your workday or your systems of record for your employees, your time keeping system and many other proprietary databases. And lot of times the ask is, well, I need an insight about a patient or a physician that connects all of these together. And historical way of doing that would have been to go get extracts from each system and get some sort of a key report or a dashboard and design that. But that's sort of the old school way of doing things these days, in the sense that it takes too long, you don't have the right governance, and it's too cumbersome and too mechanical. And so one of my focus areas has really been to connect the data better. And connecting is not only not just necessarily moving it from point A to point B, but providing an ecosystem where we have access to everything, and we are able to make decisions very quickly through better technology, through some of the advancements in technology that we see, so that the stakeholders can get what they need. And of course, with the advent of AI and generative AI, we are now starting to see tremendous amount of, you know, interest in asking natural language questions. I don't want to wait for a report to be built. I want a simple question and answer so. So what I'm working on, and my team is working on is really first. How do we know where all the data is second? How do we connect it or make available all the different parts of the data? But more importantly, to catalog it. And cataloging is an important part, because we want to know where did it start and where did it end. Where does the data end, and then all the transformations that happened along the way. So we spend a lot of time thinking through all of those things, and then having some level of governance. Governance is a key part of all of this, because you want to make sure that the method, the definitions of all these data elements, are accurate, and there's a business concept or a stakeholder, and then we can measure quality of those data, because we making critical business decisions. So So to summarize, it's all about ultimately driving towards self service, which is a question you asked, but the elements of getting there, you have to find it, integrated catalog and govern it, and then bring it together using modern technology and tools so that it becomes easy to write success. Yeah,
Shahid Shah 4:23 so a lot of our listeners love to hear what tools, and if you want to mention vendors, certainly you can what, where are the tools that are doing really well for you? But then where are the gaps? Like innovators who are listening to this might say, Oh, hey, Karen's got a gap here. I can fill that gap. So tell us what, how, what's working well, what are the tools you're using, and then where are the gaps that you wish people would be filling?
Kiran Mysore 4:47 It's a great question. And our ecosystem is also evolving, so we're making a big transition from some of our on prem systems, which we used to use historic. Need to go to the cloud and and that's an important step for us, because some of the advanced capabilities we want to enable are actually in the cloud. And so we've predominantly built our ecosystem on the Microsoft stack of things. They're very strong partner of ours. We've spent a lot of time with them. And so in terms of data storage and moving data to the cloud, I mean, we use a lot of Microsoft tools and ecosystem. More recently, we made a choice to evolve to Databricks as a cloud data lake for us. And the reason the data lake concept is important is because Databricks again, one of the many leaders in the industry that's certainly one that we saw a lot of value from, is really the pioneer in getting different types of data and building the data lake house. And the lake house is important for us because we want to make sure we have version of the truth of the source system, a version of the truth that is a normalized view and a version of truth that ultimately presents to the end user. And these three are connected and they're harmonious, but we think of it as a medallion architecture, so we're building that capability in terms of visualization. You know, we know different from many other health systems. We have many tools around visualization of data. But more recently, with the advent of AI, we've actually spent a lot of time going deep and understanding the LLM architecture, the large language model architecture, and we've built some of our own capabilities to integrate different types of algorithm based models to provide a Q and A in terms of gaps, and that's a good one. Frankly, I think we've got the range of tools we need. A lot of times it's about integration of the tools. Sometimes there are two options for the same business problem, right? So think about catalog, for example, or data quality, or when you think about loading data, there's so many different choices we have, even in the ecosystem that we've purchased. And so we also think about an architecture framework of, you know, what is best in class, Western breed for what we are trying to solve, and make sure that we we have a scalable open architecture so we don't sort of close down to, you know, just one. So as we evolve, I think I'm sure we'll find more gaps in the ecosystem. A lot of it is, how do you get more scale and speed? How do you sort of have low touch in terms of data visibility and accessibility, and then just overall monitoring and observability of information and driving quality? These are the key areas we focus
Shahid Shah 7:38 on. Yeah, no, I love the idea that you guys are doubling down on Microsoft. Obviously, things that Microsoft is currently missing usually appear a month later or a year later, but they will end up Databricks is a fantastic vendor as well for Lake houses. One of the challenges is that scheme, both the Databricks and Microsoft often offer things like schema less systems, but in the end, there are no schema less systems that are useful in the utilization of structured reporting, et cetera. Where do you guys fit in? How you do schema management, and what do you require? Early bound in terms of schema, what do you like to be laid down in terms of schemas?
Kiran Mysore 8:22 Again, a great question, and something we think about all the time. The way we think about schema is more from a standpoint of what is the overall data model that we should create so that the key answers that we drive toward can be easily answered. Now, there's two parts to this. You know. One part is, who are your stakeholders? You know, are they your classic you know, executives, leaders who don't care as much as the schema as much as you know, can I get my information right? But we also have others that are more deep analysts or data scientists who actually want to understand, you know, how everything is connected behind the scenes. And they want to really traverse from a summarization, for example, to the source, source of the data. And for them, schema is really important. Other data models are really important. The other way we slice and dice. This is executive reporting versus operational reporting, right? Executive reporting or integrated reporting as well, you have to connect a lot of different things and normalize your information. So in those cases, we think of a specific schema for those right so, for example, if I want to connect, you know how many surgeries happen in a hospital, but then I want to connect that to, well, how did my labor actually performed? Hospital staff performed, well, they're in two different systems, and so if an executive report is, what is my labor productivity? That's my question. I have to connect a lot of things together behind the scenes. And so that's for executive reporting, but for operational reporting. Sometimes it may be how many misses happened yesterday that would. A very simple question we may have that in our electronic health record. So all we need is just to go into the system, find where the tables are and get the answer. So the way we think about it is persona first, you know, who as who's asking the question, what will they do with the data? The second thing is, what is the actual use case? And you know, is it executive? Is it operational? And there are many others, even data science, for example, who might want, like, a whole list of data elements, right so that they can build their AI models. So we think about all of these things, and then we tailor the answer. So it's not one size fits all, but it's certainly one where we we have competency in each area, and then we surround the solution with the right combination of competencies,
Shahid Shah 10:43 yeah, that makes a ton of sense. So if you, if you think about the raw data, there's going to be some schema somewhere, right? So with a natural way that Databricks is working, it's going to be columnar, and then you're kind of like layering on top. So what, where do you fit things like the fire model, or the CCDA or the omop. Where does all of this fit in?
Kiran Mysore 11:05 Well, they all fit in in different levels again. So when you think about fire, for example, we exchange information a lot with our third parties, and you know, other groups and partners that we work with all the time. And when you exchange information, you want to have it in a very standard, secure format, and fire, obviously, is one of the examples of how we do that. So we take advantage of the platform capabilities to drive the right level of integration or exchange information exchange. So that is, that is one, one aspect of how we think about it, when, when you sort of step back and look at the ecosystem that you need to support different use cases, one of the things we have really thought about is you said, Omar, right? So we've actually thought about what is the right way to connect all these different parts of our ecosystem. And by the way, a lot of this is not unique to Sutter. It's actually a common healthcare model, right? So you want to take advantage of what the industry has already done, and to the extent possible, connect things together. And this is where we lean on our partners a lot. So we take advantage of what has been built, or we look at it very carefully, but at the same time. We also customize it to start because we do certain things in a certain way. We are integrated health delivery network, which means we have a combination of acute and ambulatory facilities. And lot of times, they operate in slightly different ways. And so when you think about referrals, for example, from you know one part of the ecosystem to another, you want to make sure that all of the information is very consistent and modeled in a common way. So those are some of the business challenges that we actually think about when we start to look at modeling in terms of the home of itself. We are starting to gravitate towards that. We've had versions of that in the past on the on prem side of things. As we move through data fix the cloud. We're actively looking and saying, as different data sets come in, whether it's our labor time keeping system, our employee system, or EHR, or any other, like I said, proprietary system, we're actively looking to see how to connect this once so that we don't have silos of information. And omop is a is one example by which we can actually model that really well.
Shahid Shah 13:24 Yeah, that's a great idea. And in fact, just as a side note, for listeners, what they should do in their data models and their data modeling tasks, data science tasks, is you want to try to look as much as possible as other agents, as other institutions, so that you can get people from other institutions on board quickly and begin to use it. There are a lot of people who have been used. They have very old schemas, and they're great because they work. But then the people are not very portable. When you bring somebody new, you have, you know, six months of training and other things. So do you feel like as you move towards, you know, a deeper Microsoft stack, a deeper data brick stack, that your ability to hire retain, continuously trained workforce becomes easier. Or what could you do to make that easier? So,
Kiran Mysore 14:16 great, great question. And the whole topic of talent and hiring and retention, right? Is actually so much on top of mind these days, like across the board, you know, for us, you know, I'm actually grateful for for a fabulous team. When I joined Sutter, you know, one of the things I spoke about, and I continue to speak about, is the deep Sutter knowledge that is resident within the team. There are many folks that have been around for a long time, and over time, they've actually developed the institutional knowledge of how things work, and that's super valuable. And especially as you introduce new technologies and you go to the cloud, obviously we have done a ton of training for them. We continue to train and upskill and bring in. Other partners that know how to do this, experts that know how to do this, so they can do two in a box and learn. So we're doing all of those things right as we go. Attention for one thing is all about giving them opportunities to practice what they have learned, and that is one where we continue to work on which is, let's say I go to a great data fix class, and I learned about how to manage Databricks environments and load the medallion architecture, but I want to actually do it in real life, right? So I want to be able to have a sandbox where I can play with things, and I can extract and pull information and play with it, and maybe, in the case of AI, build some AI models as well. So the key thing here is training has to be coupled with real world experience coupled with the right infrastructure setup and the right time for experimentation and play so that they can learn. And so those are all things we're thinking about as we go through our next evolution of the cloud journey to make sure that we have the highest qualified folks who are building great capabilities for
Shahid Shah 15:57 us. So I'm an advisor to a number of different universities and other programs where we help them with their data science curricula. So if you were able to talk to those people that helped design these data science programs based on what you're seeing of personnel that come into your shop, what do you wish people knew already when they came in. Is it something as broad as you know, healthcare knowledge in general, or is it very specific to particular specialties? What do you wish your staff knew when they came in? So? So I think about it in three layers, right in the way I think about hiring talent. Obviously, the first one is domain knowledge. So some healthcare knowledge is super important, critical.
Kiran Mysore 16:41 I think that healthcare knowledge gives you a little bit of an acceleration, and potentially a way ahead, a step ahead. So that's one thing. The second thing is the technical skills, right? So, and these are skills that you hone over a period of time, but part of the technical skills that are also about, not necessarily one technology or the other, but just a learning mindset, right? Which is the ability to sort of absorb new things quickly and experiment and play with and, you know, be intellectually curious and try things out. The third one is, to me, some of the softer skills, which I think, frankly, are the most important pieces of, you know, skills that you need, right? So things about things like teamwork, collaboration, things like communication, the ability to take risks, but at the same time, balance risk with the right level of realism, if you will. So these are all skills that you need. So I think when I look at, you know, candidates coming from universities, sometimes they are very deep in the first two they're very good. You know, from a healthcare standpoint, they know the tech really well. They've done projects. I also look to see, are they leaders? Can they drive and lead? Can they communicate? Well? Can they work in a team? Are they intellectually curious? Are they do they have a learning mindset? I think these are all very important aspects of skills that I think we're looking for. The last thing I would say is, and this doesn't necessarily apply to my team, but across the board is, you know, we're in the business of helping patients get better, and so do they have the right empathy and compassion, right that goes with it, that do they exhibit that in every interaction, I think those are important elements.
Shahid Shah 18:22 Yeah, what's fascinating is that we're just obviously emerging out of the COVID environment where social skills were difficult to develop. And so if you had kids go to universities during that very, very important time, how do you suss out to do your interview process? What are the kind of things you asked for that will tell you, well, they need more training to be more socially adept, etc, or you just want it, but you're willing to train for it as well when you bring it online. You know. Andrea,
Kiran Mysore 18:52 right. We've gone through a tough period after COVID, where a lot of our work habits changed. People are, you know, a lot of folks love to be remote now and work from home and so on. And we have a very robust hybrid policy. You know, we let our employees work from home three days a week, and we have them come in two days a week. But over the last few years, this transition has happened where, you know, people have sort of become more individualistic, you know, tell me what to do. I'll go do it. Or, you know, I know, from a team's perspective, you know, electronically, you're collaborated and so on. But there's nothing to speak for some of the soft skills that come when you actually interact physically, you know, in real life. So part of the change I think we have to make is, as we start to hire again in the new age, is to provide them ample opportunities during the interview process to interact in person as well, whether it's case studies, whether it's panel interviews, whether it's actually taking a problem in it and sort of whiteboarding something out. I think there are many different techniques and. I know companies are trying many different things today. What I'd love to see is just sort of a way to simulate a real, you know, works situation, and see how candidates do in that in that scenario. I'll give you one quick example. Recently, we did a team workshop with my leaders on my team, and, you know, we went and broke up into smaller teams, and we gave tasks to each team to go and build an object using Legos. Okay, so some very simple tasks, you have the instructions, but along the way, what we did is we gave the instructions only to one person, and we separated that person from the teams, and the only way they could communicate is through a cell phone. And so the key was, then they had to really narrate that, narrate the instructions to build it. But as they were building it, you know, we took a couple of key people away from the team to see what the rest of the team would do, right? And and the teams that one were very clear about what the objectives were. They had clear roles and responsibilities. They were able to be resilient in the face of change and build something innovative, and some, in some cases, they're not afraid to break the rules to get the thing done, but in a safe, controlled way. So again, this is an experiment, of course, but more to say that we need to simulate different scenarios in the real life as we bring people
Shahid Shah 21:27 up. No, no, I love those. That was a perfect example of how there's no way you can find out how people remotely work unless you test them also remotely working. So that's great. One other thing I recommend, whether you're young or old transitioning, or you've been doing this for a while, is spend a lot of time in open source. There are lots of open source data sets. Look you can load it in your own Microsoft or data bricks environments just on your personal account. Load them up. You can play with things. Lots of places where schemas are already built out, you know, whether it's an omop or fire Microsoft, for example, has fire services and servers running. So what do you like to see in terms of open source or, you know, like these days, if somebody is coming to me as an engineer of any sort, and I ask them for their GitHub ID, and then I just go view their GitHub, what are you doing? What are you starring? What are you looking at? And normally, if I, if I have a choice, if like, if I have 10 people that are candidates, I will actually increase the priority to hire somebody that has a robust GitHub environment where they're doing work, writing their own code, publishing it to other people, liking projects, etc. Is there something like that that gives you a little bit of insight we
Kiran Mysore 22:43 take? I mean, certainly GitHub is one, one great source of information about a candidate's work or their past life. With regard to your question about open source, something we think about all the time too. One of the key things for us, right, is obviously open source gives us tremendous opportunities to drive great value and experiment with different ways of working. We want to, we want to definitely explore it. I think we haven't quite done enough. Frankly, I think Databricks and you know, the Microsoft platform certainly gives us some springboard and experimenting with some I think in this case of AI, for example, there's a lot of open source models out there now that, as you know, we can start to leverage and use and we are playing with those all the time. We build some tools and some AI products with open source as well today, and we continue to explore that even more. I think the key for us is we need to still continue to amp up our engineering expertise and prowess. Historically, we've used external parties to help us, but really we want to invest in our own and drive that higher. And so I think this year and beyond, right is really the opportunity for us to really double down, go deep, get the right expertise and or incubate the expertise entirely
Shahid Shah 24:00 no that's great. And the last couple of minutes that we have talk to us a little bit about what do you need to do to your data sets and your data sourcing for the upcoming agentic AI world. So for example, whether the agents are supervised or unsupervised, whether they're interactive or non interactive, could dramatically change what kind of data you need, because today you could grab data, potentially batch it, but in an interactive agent driven world, could batching work, or do you have to go more to real time? What are you thinking about in terms of agentic AI?
Kiran Mysore 24:34 So we're thinking both, actually. So to give you some examples in cases where the agentic AI needs to provide you with some real time actions, yes, we need to go to some of the real time data. In other cases, we're seeing agentic AI to help us with automating some tasks that require us to not only query but write back. Call Center is a great example. In some. Cases where we may need to assimilate a lot of information from many different sources to maybe check for insurance eligibility, for example, which our agents do all the time. In those cases, you have to grab a lot of information, and once you grabbed it and you have it, you can sort of process it and query it. So we're seeing both. And I think the future in terms of where we go, right, I think there is, there is no doubt that good quality data and access to it is the key to successful agenting architectures, the successful AI models in the future as well. And for us to build that ecosystem, that layer, we're actually thinking about how we can make it easy for even the agents themselves, yeah, to not only recognize a data set to parse a catalog, but then access it and actually run it so and maybe even write back, yeah. So, so the layer that actually abstracts it is a really important layer for us, and it's work that we need to do more. And I think that's the journey we're
Shahid Shah 26:03 on that's fantastic. So really appreciate your time here. Get on and Sutter is obviously a great place to work. So if anybody's interested, definitely call. Get on up and
Kiran Mysore 26:14 review there. We welcome great minds, and we're always looking for great minds to join us. Awesome. Thank you so much. Thank you.
Thank you 26:24 Thank you for joining us on digital health talks, where we explore the intersection of healthcare and technology with leaders who are transforming patient care. This episode was brought to you by our valued program partners, automation anywhere, revolutionizing healthcare workflows through Intelligent Automation, netera, advancing contactless vital signs, monitoring elite groups delivering strategic healthcare. IT solutions, sell point, securing healthcare, identity management and access governance. Your engagement helps drive the future of healthcare innovation. Subscribe to digital health talks on your preferred podcast platform. Share these insights with your network and follow us on LinkedIn for exclusive content and updates. Ready to connect with healthcare technology leaders in person. Join us at the next health impact event, visit health impact forum.com, for dates and registration until next time, this is digital health talks, where changemakers come together to fix healthcare.