Kelly Kohlleffel (00:01)

Hi folks, welcome to the Fivetran Data Podcast. I'm Kelly Kohlleffel, your host. Every other week, we'll bring you insightful interviews with some of the brightest minds across the data community. We're going to cover hot topics such as AI/ML, GenAI, enterprise data and analytics, and various data workloads and use cases, including things like data culture, and a lot of other areas. Today, I'm pleased to be joined by Erin Rebholz. 

She is president of Merilytic Health,a pharmacy analytics consulting and technology company. Prior to Merilytic, she spent 15 years at McKesson. She was in Corporate Strategy and Analytic Services for McKesson's distribution business, and then Business Development, Innovation and Data Strategy for McKesson's pharmacy technology business. Erin has an MBA from the Wharton School at the University of Pennsylvania and a degree in chemical engineering from Brown.

She's currently pursuing a master's degree in data science through Harvard University's Extension School. Erin, it’s great to have you and welcome to the show.

Erin Rebholz (00:58)

Thanks Kelly, it's fun to be here. I appreciate the invitation.

Kelly Kohlleffel (01:02)

Absolutely. I am really excited to have you on the program and dive into your experience and perspectives on healthcare. Before we get going, why don't you talk a little bit about Merilytic Health and what you're doing right now.

Erin Rebholz (01:17)

You mentioned that I am going back to school for a data science master's degree. Part of that is just trying to imagine the possibilities for how advanced analytics tools can be applied to the pharmacy and healthcare industries. As I think about Merilytic Health, it's my own personal consulting firm as well as technology development. I’m excited about the work that I've been doing from a research standpoint, as well as broader perspectives from the industry as I meet up with folks like you.

Kelly Kohlleffel (01:47)

So there's so much value that your years of healthcare experience can bring to all sizes of firms to help them take advantage of their data and do things around different types of data workloads in a way that is more valuable to the organization. In that vein, what do you feel is essential when you think about a modern approach to data and analytics today?

Erin Rebholz (02:23)

There's no one size fits all. With some of our more data-as-a-service, self-service models and cloud-based offerings, there's more ability to make small investments and really experiment and develop smaller applications in a cost-effective way. Even small and medium-sized companies can now adopt some of the best analytics tools. I think we're in a really great era for analytics tools right now. 

Kelly Kohlleffel (02:50)

I remember so many people that I’ve talked to saying, “Help me get to self-service.” That can be a really tough target to hit. I'd be interested to hear your perspective on that. How close are we with the tools, approaches, technologies and skill sets to really being able to achieve self-service today?

Erin Rebholz (03:11)

I think there are two things. From a data standpoint, being able to access the data with cloud-based data warehouses gives a lot more flexibility to access that data from any analytics tool that you'd like. I think maybe data self-service is different from analytics self-service. On the data self-service side, I think that cloud-based data warehouses definitely enable that, but the other topic that’s probably top of mind is governance. Helping people within your organization understand what that data is, is also important as well. You have to invest in the areas that are going to be most meaningful to your business. 

Kelly Kohlleffel (03:48)

I agree. I'm also interested in the current work that you're doing at Harvard around some of this additional study. How much flexibility is there in the tool sets that you’re looking at and working through as you're studying? Is there one tool that does it all, or is it more that you can bring your own tool to the game to develop these different data science or AI/ML type outcomes?

Erin Rebholz (04:18)

It’s interesting coming at it from a student perspective versus an executive perspective. It’s very different. From a student perspective, we're learning algorithms that can be generated and practiced on data sets that may not be as robust as some of the executive level data sets. From a student-level perspective, it's much more important to understand the algorithm and potential applications of the algorithm, whereas from an executive standpoint, it's thinking about the right architecture to scale applications. 

It’s a little bit of a different mindset. As I'm coming at it from an executive perspective, I think: How do you really scale this more meaningfully?  At least with this program, there are definitely classes being introduced about scaling AI applications because they're realizing that's a skill people need to adopt as well.

Kelly Kohlleffel (05:14)

If I'm a small or medium-sized organization, I may not have the specialized skill sets available or all the tools in the bag, but can I bring existing skill sets to the table that will allow me to get something done relatively quickly without a massive investment and grow from there?  

With the cloud, are you seeing more approachability and accessibility today than five years ago, regardless of the organization size?

Erin Rebholz (05:48)

Yeah, and if you look at the core tools that are offered through cloud providers, they're much more robust. They aren't as robust as some enterprise scale tools, but if you just want to do ETL and grab files and load them into your database within the cloud, you could do that. If you need to do that on an ongoing basis or you're managing multiple jobs, you still need to invest in certain tools, but those tools are available within your cloud environment. You can easily run them up. It's a cost-based model, not a license where you need to have a server. I think that process is much more robust now than it was five years ago. There are definitely a lot of strides being made across the cloud providers.

Kelly Kohlleffel (06:31)

I agree. What are the top tech trends that you're seeing right now? Maybe there are things that you're talking about with colleagues or other folks at Harvard or at Merilytic right now.

Erin Rebholz (06:47)

We’ve spoken about AI/ML and workloads, and adopting the right tools for your business and figuring out how to manage the adoption of some of these advanced analytic tools and protect the confidentiality of your data. I think that those are definitely top of mind in both places. 

The other thing that excites me is data interoperability within healthcare right now. I've been doing a lot of deeper dives there, especially within pharmacy and within the provider community. There's a lot more broad access to data. There's a common framework for getting access to the data across covered entities, so how do we connect across our healthcare delivery networks in a more meaningful way? That’s definitely very exciting.

Kelly Kohlleffel (07:42)

It is. Access and interoperability are two huge issues in healthcare. How far have we come? 

Erin Rebholz (07:50)

I do think we've come a long way and I think that we have a long way to go.  Our national government is establishing more of a framework for data exchange through TEFCA — a common framework for accessing data and regulation around that. 

I was speaking to someone familiar with the European market and they were somewhat taken aback by all this talk about interconnectivity, because for them that just happens through the government and you don't need to have various stakeholders to organize and define those frameworks. 

Academic institutions may have adopted more than some of the regional areas — pharmacy in particular. I was just at a conference held by the NCPDP, and they're just this year being asked to opine on how to bring pharmacy data into that framework. I definitely think that there are a lot of opportunities in terms of better connecting our data. That helps the clinical outcomes for patients as well, so pharmacists don't have to call the provider to get access to the data that they need. 

I was recently at an exchange of information around the Boston area with various medical centers exchanging charts and imaging data. How do you do that in order to really improve your patient care? That's definitely top of mind within the industry.

Kelly Kohlleffel (09:17)

Like you said, it should be about better clinical outcomes and better patient outcomes. Are there certain groups, organizations or areas within healthcare that you feel are doing a better job of pushing this interoperability? 

Erin Rebholz (09:32)

Sure. Recent federal legislation established QHINs, which are ways that regional areas or groups within the federal government (they're actually commercial entities) can exchange information between healthcare entities for clinical purposes. In the future, they’ll be able to exchange for patient purposes, and longer term, maybe for other things like treatment or operations or payment. They focus primarily on treatment and clinical applications first. That piggybacks on some of the EHR-based efforts and industry efforts, but then the federal government joined in and built upon that to create better accountability across the organization.

I think the reason why people are doing it, though, is centered on patient care and better access to information so that a provider who may not be part of your EHR system, or even a health system with multiple EHR systems, can combine information across EHR systems. It enables better access to information across the ecosystem. I think the next phase will be patient access to the information. Some of that is already in place, just as we think about HIEs, or health information exchanges, that have formed commercially or in local regions.

Kelly Kohlleffel (10:55)

Interesting. Are there areas that, as an industry, whether on the tech side or the health care side, we should place more value on going forward? 

Erin Rebholz (11:08)

I think the two things you need to be focused on are, one: data governance and better cataloging, curating and tagging your data, enabling its use within your organization. The second thing that I think is undervalued is cost management and monitoring the use of your teams. 

As you adopt cloud technologies at scale, you have the ability to keep it running by accident, or scale up a server in a way that you didn't know was happening. I think you need to keep on top of things very actively. A number of the cloud-based data warehousing tools are developing more monitoring tools, but I think that that's something that’s important to make sure you're developing as an executive leader.

Kelly Kohlleffel (11:57)

For data governance and cost management optimization, do you feel like it's more effective when done centrally, or should some of that be pushed down to a business unit?

Erin Rebholz (12:14)

I think the data governance might have a different answer than the cost side. On the data governance side, I do think central investment is best. What happens in large organizations is you develop these data silos. How do you create something across your organizations so that people realize what you have so that you don't recreate it across your organization? You also have to incent people to use it, and make sure that it's kept up for those high value data sets. 

On the cost side, I think you have to do it at the business level and the cost center level, because that's really who is paying the bills. I think that there's the incentive to do something about the information. If you're running up your own bill, there's an incentive to dedicate some time to making sure you're not going over your costs and budget.

Kelly Kohlleffel (13:06)

You mentioned data silos in healthcare and that we are seeing improvements in that area. Any comments on that? How prolific is that challenge? 

Erin Rebholz (13:16)

There are a couple of things that are interesting within healthcare. One, just for everyone's information, you have a contractual obligation to your patient to keep their information private. People within healthcare take that very seriously. That said, there are ways within HIPAA (which governs disclosures of personal health information) of de-identifying that data. There are ways of making sure that people have the right use of that data for treatment, payment or operations.

Now there are regulatory frameworks to examine there. There's also interesting information where people are taking models and pushing some of the model load down to the specific user provider and just returning an answer. Federated analytics is something that's being examined in a number of different technology providers within the healthcare space so that the covered entity for the patient information doesn't always have to disclose the PHI, they can actually disclose the answer. 

Kelly Kohlleffel (14:24)

You mentioned PHI. There’s a whole other level of security and governance when you're talking about personal information in the context of healthcare that we've got to be mindful of as we're dealing with all these data sets.

Erin Rebholz (14:37)

Yeah. I think people within the healthcare industry take that seriously.  Some of the movement to the cloud recently has been because there have been privacy incidents within health systems or healthcare providers that were managing their own data centers and maybe not keeping abreast. I think people are realizing some of the larger-scale cloud providers have made great inroads in terms of their own protections.

Obviously, you have to elect into those frameworks as you adopt the cloud, but I think that there is a lot of movement towards the cloud operators at this point because of that.

Kelly Kohlleffel (15:12)

I'd be interested in your perspective on the balance between risk and innovation. As you've been leading these data teams, there are probably way more requests than you can fulfill.

How do you balance the risk innovation/profile of each individual request?

Erin Rebholz (15:31)

I do think part of it is making sure that the request is valuable and realizing that you will need to go through a review. It’s important to make sure there's going to be value from the risk. You can do a lot to validate a business opportunity before you actually see data. There’s actually a lot of  synthetic data being adopted for some of those purposes because it's easier to do a light POC with some synthetic data versus accessing PHI.

It’s something the healthcare industry has taken very seriously in terms of going through a review process for any data use and reviewing your customer agreements to make sure how you plan to use the data aligns with those customer permissions. That all said, I do think there are a lot of opportunities around innovation within the space, but you need to put some guardrails around it because of the risks.

Kelly Kohlleffel (16:31)

Is synthetic data something that is used quite a bit in healthcare today?

Erin Rebholz (16:35)

As you do your setup for any analytics, oftentimes you have production or dev and testing. You often don't have real data until you get into some of the testing so that you're mitigating any risk associated with live PHI. 

Part of the way it's used right now is to create better synthetic data in that development phase. Especially around analytics, you need to have a data set that is actually meaningful. I do think that data is getting better and better, and I do think it's a good first pass for some analytics to figure out if an algorithm works. You’ll have to go back and validate, obviously.

Kelly Kohlleffel (17:20)

You talked about some of the work you're doing right now, maybe a smaller focus data set. I feel like there's still value that can be gained from a focused contextual data set, especially with some of the new ways that we're approaching AI today with generative AI.

What are you finding as you're running various tests and experiments and looking at different ways that AI can be applied in healthcare? Do you feel like it always has to be a large data set or does a focused contextual data set provide some value as well?

Erin Rebholz (17:53)

First of all, generative AI and large language models in general really do benefit from a large context of information. You're taking advantage of speech. You're taking advantage of photos and image data and sound. You can use that in order to create and build on top of it.

You can actually leverage the data that the big giants are amassing. You build upon that in creating your own models with transfer learning and adopting some of the large language models. I think that that's maybe a separate conversation. I do think that you can optimize your operating efficiency and think through smart ways to use some of those tools to improve your industry. 

I think we'll see a lot of that in the next year or two. Part of that is looking at operational efficiency and figuring out how to use those models for your own segment moving forward. That said, there's also this reflex in the analytics community that simpler models are much more explainable. 

Especially within healthcare, if you're making a diagnosis, you want to take advantage of those larger data contexts, but also make your analytics explainable so that the practitioner can actually interpret them and make a decision associated with the patient. I do think some of the smaller models may be more explainable, so there is a focus there, but then some of the neural networks help reduce the operating tasks within the industry.

You could isolate where a tumor may be on an image but then rely on the radiologist to actually make that interpretation. You could use image data to look at size and radiomics associated with the tool, but then create a smaller model that can be more interpretable that actually makes the diagnosis. 

Kelly Kohlleffel (20:03)

Are there any other areas where you’re seeing LLMs and GenAI impacting healthcare? I know it’s early days.

Erin Rebholz (20:11)

I actually have been. I've been to a number of conferences recently, and one of the things that strikes me is that a lot of technology companies are adopting LLMs and generative AI, and they're putting that in their technology offering. As an organization, you're going to purchase technology that's built on these models. I think each business needs to think about the core capabilities they need to develop for their operations to really focus the adoption of those technologies.

Kelly Kohlleffel (20:45)

This is a side question. People, process or technology: which is the toughest one to get right?

Erin Rebholz (20:52)

Process. I think that that's the toughest. People is a hiring thing. Technology evolves and you need to be flexible, but process is often undervalued.

Kelly Kohlleffel (21:08)

In healthcare, you've got some really long-lived, very stable companies that have been around a long time and have process and approach firmly embedded. Having to shift that within a large organization can be a huge challenge.

Erin Rebholz (21:27)

Yeah. Within smaller companies, you can work more nimbly, but oftentimes, the problem is that you need to develop those processes to mature. 

Kelly Kohlleffel (21:37)

This has been outstanding. 

I'd love to get some personal perspectives, like your best practices and your learnings. A lot of folks in the data community would be interested to know your advice for individuals looking to modernize their organization’s approach to data.

Erin Rebholz (22:03)

I definitely would advise them to look for areas of experimentation. I think what's been helpful in my career is taking opportunities, even if it's a small investment, to broaden your team's perspective in terms of what's out there. 

Oftentimes people get used to what they're doing and they're not really looking for the next thing. So it's helpful to use smaller investments in order to build capabilities within your team so that people don’t get stuck in a specific area. 

I've also really valued education for both my team and myself because I think it's a lot easier to train a person to do something new than it is to find a person that does exactly what you want to do, especially within technology. That’s definitely something that I think is important from a leadership perspective as well.

Kelly Kohlleffel (23:01)

I love both of those. I look forward to keeping up with everything that you're doing with Merilytic Health. I'm sure we'll have more conversations in the future. 

Thank you to everybody who listened in today. We appreciate each one of you and highly encourage you to subscribe to the podcast on any of the major platforms, Spotify, Apple, Google, et cetera. You can find us also on YouTube. 

We'll see you soon. Thanks a lot.