Lessons from Aaron Neiderhiser and Coco Zuloaga, CEO and CTO for Tuva Health, on democratizing healthcare analytics
Aaron Neiderhiser and Coco Zuloaga, CEO and CTO for Tuva Health, a modern healthcare data analytics company
Subscribe to our substack for updates and listen on Apple Podcasts or Spotify. Connect with Andrew or Lipsa if you find this post insightful and want to learn more.
Welcome back to the Pear Healthcare Playbook! Every week, we’ll be getting to know trailblazing healthcare leaders and dive into building a digital health business from 0 to 1.
Today, we're excited to get to know Aaron Neiderhiser and Coco Zuloaga. Aaron and Coco are the CEO and CTO of Tuva Health, a healthcare data analytics company simplifying claims and EMR based analytics to supercharge and standardize data-driven innovation.
Founded in 2021, Tuva has raised about $4m from YC, Virtue, BoxGroup, and some angel investors.
Aaron and Coco are experts at using data science and data analytics to help providers understand how to coordinate care and reduce unnecessary healthcare costs. Prior to Tuva, Aaron served as the Senior VP of Technology at Health Catalyst, a top company providing data and analytics technology to healthcare groups, and worked as a statistical analyst at the Colorado Department of Health Care Policy and Financing. During his time there, he met Coco Zuloaga, who would later become his co-founder and CTO. Coco is an experienced data scientist and has held various roles such as senior director of data science at Big Squid Inc and VP of Data Science at Strive Health.
Aaron earned a B.A. in Economics from Coe College and a M.A. in Economics from the University of Colorado Denver. Coco completed a Bachelor's Degree in Engineering and Physics from Technologico de Monterrey, an M.S. in Applied Mathematics from the University of Waterloo, and a Ph.D. from Rice University.
In this episode, we learn how Aaron and Coco are trying to democratize and standardize healthcare analytics, speculate on the costs of the lack of standardization in healthcare, and learn about how open-source projects can fix this.
If you prefer listening, here’s the link to the podcast!
Coco’s transition from quantum physics to healthcare data
Coco (CTO of Tuva) started his academic journey in theoretical physics and math, culminating in a Ph.D. focused on applications of quantum mechanics. He has always been intrigued by the evolving landscape of machine learning and data science.
Coco’s initial foray into healthcare was eye-opening, revealing the uniquely intricate and messy nature of healthcare data. Part of the issue was how much domain knowledge was required to work with messy healthcare data. All the domain knowledge was inaccessible institutional knowledge tied up in the brains of people who have been working with claims data for decades.
But that didn’t dissuade him easily and, despite initially lacking familiarity with healthcare, Coco became a data scientist at Alliance Health.
It was during this tenure that Coco crossed paths with Aaron, a connection that later prompted them to join forces at Health Catalyst. Before starting Tuva with Aaron, Coco held other senior data science positions at Big Squid and Strive Health.
“I came from this mindset of thinking that machine learning is so cool and sexy, and I wanted to do all these cool applications with new technologies in very practical ways that were impactful. So healthcare sounded awesome. But then very quickly, I realized it was very difficult to do any of this because the data was such a mess. And of course, the data is a mess in most industries, but in healthcare it is especially complex and messy.”
Aaron’s healthcare data deep dive
After graduating with a degree in economics, Aaron took a position at Colorado Medicaid on the recommendation of his professor who highlighted the agency’s innovative tendencies.
Aaron had low expectations for the interview, having had no prior experience in healthcare. And yet, like Coco, he found himself immediately captivated when his interviewer began talking about payment reform and value-based care.
The most exciting part about working for Colorado Medicaid was the data. They were sitting on a longitudinal dataset covering over a million people for 20 years. Over three years, Aaron focused on data analysis of claims data, including risk adjustment rate setting, utilization, and outcome measures. He was even on the path to becoming an actuary.
Source: Colorado Medicaid had just expanded in Jan 2014, resulting in a decrease in uncompensated healthcare costs and uninsured rate
Then, he started to explore the idea of working on the underlying software that people would use to super power their analytics. As Aaron explored potential employers in 2013, he saw that Health Catalyst dominated SEO in healthcare analytics and, due to good market timing and their Series B fundraise, he ended up there in 2014 as a data architect.
In 2015, Health Catalyst initiated a business unit focused on consolidating customer data for benchmarking, machine learning, and real-world evidence for Pharma. Aaron played a pivotal role in launching this unit and later recruited Coco to join the venture.
A co-founder meet-cute
Aaron and Coco initially met while playing squash in downtown Salt Lake City, around 2015, while Aaron was at Health Catalyst and Coco was still at Alliance Health. They played squash many times before discovering they both worked in data science with healthcare data.
After being recruited to Health Catalyst and working with Aaron for a year, Coco explored various opportunities but remained connected with Aaron.
Even after Coco moved on to Strive, their squash game conversations increasingly centered on common issues in healthcare data management. Over months of discussions, Aaron and Coco recognized the persistent nature of these challenges, sparking the idea for Tuva.
Born out of shared experiences, Tuva aimed to address common problems in healthcare data management, reflecting the duo's desire to make a meaningful impact and revolutionize how healthcare analytics is done today.
Aaron and Coco’s strong partnership is based on truth and rooted in friendship
Aaron and Coco's collaboration was rooted in a strong friendship formed outside of work which allowed them to be more direct in their communication and made working together enjoyable and productive.
“What I [Aaron] treasure most about our working relationship is our intense discussions and debates. Usually, I’m the one with strong opinions, feeling a lot of urgency about something. But, for us, it always comes back to seeking the right answer from a total truth-seeking standpoint. That foundation is the bedrock of everything we do.”
Their working dynamic thrives on a shared commitment to truth-seeking and intellectually honest problem-solving, fostering compatibility in their work philosophy. They often engage in intense discussions and debates which are rooted in a dedication to truth and avoidance of self-deception.
They also had meaningful shared experiences that helped them both see clearly the intricate challenges of working with complex data issues.
Aaron fondly recalls beating Coco in ping pong, a rare victory against someone who excels in every racquet sport, showcasing the lighter side of their professional relationship.
Tuva Health’s Mission
A recent report found that only “20% of healthcare organizations fully trust their data.” Having messy data is a big issue for any healthcare company that relies on data to show they're cutting costs and making things better. But before they can analyze the data, they need to make sure that their data has a solid foundation. As Coco mentioned, the biggest bottleneck is the quality of the underlying data and the transformation required to make it usable.
That’s where Tuva comes in. Tuva Health cleans up raw healthcare data and turns it into actionable, analytics-ready data. For Aaron, Tuva aims to accelerate the creation of high-quality analytics, evidence in AI, leveraging real-world healthcare data from diverse sources such as claims, medical records, and other clinical data.
One of the bottlenecks in healthcare analytics is the challenge posed by data transformation, particularly when dealing with raw claims data that lacks specific identifiers for healthcare events.
Imagine you're running a digital health company and you’re trying to show how your product results in reduced hospital readmissions. The administrative claims data you get from your payer partner on your population will not contain indicators for readmissions. So how would you tell which inpatient admission was a readmission? How do you tell the difference between someone going in for a planned surgery and someone ending up there unexpectedly? How do you link the data collected by your engineering team with the data you get from claims?
Source: Tuva’s website explains the underlying method to calculate readmissions in simple to understand terms.
Tuva undertakes all necessary data transformation processes to generate analytics-ready data, beginning with the establishment of a standard data model for healthcare data.
Tuva also comes with a comprehensive set of free tools, including measures, groupers, geocoding, social determinants, and terminology sets, which were developed to facilitate this transformation.
Source: Tuva’s model allows you to transform all the different types of data sources on the left into the common healthcare concepts on the right by enriching it along the way with terminology and reference sets.
While anyone is free to use the online tools, Tuva also has a paid service called Data Factory, where they will transform an organization’s raw data into the Tuva Data Model for a fee.
By utilizing the Data Factory, organizations can access analytics-ready data in ~2 weeks, and redirect their focus to analyzing the data, minimizing the need for extensive in-house engineering efforts in managing complex data transformation pipelines.
A democratization of healthcare analytics
“We didn't want to be just yet another healthcare analytics company. We had to do something different. That’s why we got really excited about open-source.”
Embracing an open-source model, Tuva makes all code and knowledge related to data transformation available online for free, advocating for transparency and consensus in the industry. Open-source projects allow anyone to become a healthcare data analyst and data engineer.
Source: Tuva’s Knowledge base breaks down concepts and provides valuable reference data marts that can enrich the value of your existing data.
Tuva is trying to streamline the common industry practice of building concepts on top of raw data. When definitions for common concepts like readmissions vary or are otherwise opaque, it becomes ripe for errors and discrepancies.
Tuva advocates for transparency. By showing how data transformation is executed they can facilitate consensus and establish a clear source of truth in the industry.
On why open-source projects in healthcare are rare
Healthcare is highly regulated and fragmented. Especially for analytics vendors, you are often building for one user and selling to another, all while juggling the regulatory and cost complexity.
Open-source appears to further complicate business models, especially in an industry with an already intricate structure. So most businesses may be hesitant to add an extra layer of complexity to their existing challenges.
“Open-source really is a give-to-get business model. From a business standpoint, we're giving away something, what is the business getting in return for it?”
Most healthcare data companies, whose business revolves around specializing in data transformations, choose closed solutions. This means you pay them to analyze your data, but you don’t always know how they do it or have the means to reproduce it on new data.
For these companies, it’s harder to justify open-source. If the core offering is open-sourcing data transformations, what is the business selling?
The decision to keep products closed is seen as an easier path, given the intricacies involved in maintaining a balance between openness and business viability.
However, this results in the underlying opacity in healthcare analytics that causes variation in what should be simple metric definitions like readmissions.
On variation in healthcare analytics
Standard definitions exist for certain healthcare measures, such as CMS readmission measures, with annually published logic. Despite standardized definitions, variations can arise in code implementations, leading to different results.
On the other hand, many healthcare measures lack standardization, relying on heuristics and rules of thumb for calculation. Concepts like member months may be defined differently, resulting in variations of up to 30-40% in calculated metrics. Small variations in defining concepts can significantly impact analytics results, with the propagation of concepts amplifying differences.
“The way people define concepts like member months or PMPM can vary slightly. We’ve done some exercises where we calculate things like member months slightly differently and we see variations in the corresponding PMPM that are as high as 30-40%. That’s crazy - for some of these simple calculations you can get such a different answer if you define a concept slightly differently.”
Source: Tuva is trying to educate users on how simple calculation discrepancies in PMPM could inflate the metric.
By sharing code and documentation openly, Tuva is inviting industry-wide collaboration and agreement on these non-standardized metric definitions. Contributions through pull requests facilitate consensus on these complicated topics.
Another benefit of this approach is that members of the industry can engage in conversations about the best ways to define concepts and best practices. Healthcare analysts can learn from other experts and unlock the institutional knowledge that exists in their brains.
Building a knowledge-sharing slack community
Tuva’s open-source project is paired with a *free-to-join* public slack community. The goal is to break down silos and unlock the wealth of knowledge existing in the minds of healthcare data experts. This addresses a knowledge problem prevalent in healthcare data by tapping into the expertise of professionals.
Despite being in the early stages, community members have already provided valuable insights and expertise, including sharing unique approaches and extensive experiences on specific topics. Aaron and Coco are humble leaders – happily leveraging this expertise to make Tuva better.
“Often community members will tell us, ‘Hey, I've actually done this [metric definition] a lot. And this is how I do it.’ And we've been like, ‘Wow, we thought we knew how to do that. But you've spent like, way more time than us thinking through this. And so now this is how we're gonna do it.’”
Contributions may take the form of pull requests, Slack discussions, or direct messages, enhancing the collective knowledge base. The Slack community serves as a platform for in-depth discussions and knowledge exchange among healthcare data professionals.
Source: Check out how Tuva’s slack community of 900 people discusses issues and discrepancies. Collaboration is in this community’s DNA.
The billion dollar problem of not standardizing healthcare definitions
The healthcare industry has been rebuilding similar solutions for more than three decades, leading to redundancy. The economic cost and time wasted by every healthcare organization and data team repeatedly doing the same tasks is incalculable.
Coco comments on the differences between the worlds of theoretical physics and healthcare data. In physics, all knowledge is acquired through deductive reasoning – assuming that certain basic facts are true. In healthcare, the very foundation is different from company-to-company.
A severe reproducibility crisis in scientific research highlighted the need for reliability of data-informed decisions. Without standardization, we are risking arriving at incorrect conclusions and producing unreliable evidence.
“I wasn't even that familiar with this severe reproducibility crisis in science, especially in medical science, now that I'm acutely aware of that, I'm fascinated by it. Even in rigorous peer reviewed scientific research, we can't reproduce any studies. Now imagine trying to reproduce results across companies, where they are just calculating on their own data with data engineers who are not trained in rigorous study design.”
When companies define the same concept differently, trust in data-informed decisions becomes challenging. The analogy of everyone "pretending to build products" resonates, emphasizing the need for real-world evidence based on consensus and standardized definitions.
We saw with the example of PMPM how easy it is to manipulate metrics, such as reducing spending, by tweaking calculation methods. We could be misrepresenting outcomes as well as costs based on adjusted calculations.
“So all these companies are claiming that they are making data informed decisions and trying to show that their care model works by rigorously analyzing data and things like that. And we can't even trust any of that.”
Tuva’s future and embrace of AI
Various tools and transformation processes are currently in place, including financial, quality, and utilization metrics, as well as groupers for conditions and procedures. Recent additions include geocoding and social determinants, enhancing capabilities for geospatial analytics.
Plans for 2024 involve implementing the latest versions of CMS HCCs and HHS HCCs, further expanding and refining the project. Ongoing work includes building applications and data products on top of the established data model.
The project's website serves as a comprehensive resource, offering documentation and a knowledge base on healthcare data and analytics. Future enhancements will introduce a searchable interface, potentially leveraging chat-based GPT models for improved user interaction.
They are exploring the integration of Large Language Models (LLM) to facilitate chat-to-SQL interactions on the data model. This feature, part of the Data Factory offering for paying customers, aims to simplify data exploration and querying.
“Chat-to-SQL performs best on data models that are close to being analytics ready, meaning like, the data tables are very wide, and they have all the columns that you need for answering your questions — exactly the types of data that Tuva’s data mart produces.”
They also anticipate the launch of a journal club podcast in 2024, providing insights into the latest developments and industry discussions. Recent partnership announcements, including one with Metriport (universal API for healthcare data), indicate the company's active engagement and collaboration within the healthcare data landscape.
Advice for technical founders
Aaron and Coco both emphasize the value of gaining deep knowledge in a specific niche domain. They suggest observing areas where the existing solutions are inadequate, which can help you recognize gaps and envision better solutions.
They both encourage individuals, particularly younger ones, to find a job in a domain they're passionate about to learn extensively. They also advocate getting lost in a problem one cares deeply about as a foundation for starting a company.
Deciding on an initial business model
Aaron and Coco would not recommend blindly adopting open-source models without a clear understanding of the underlying reasons. Their initial motivation for embracing open-source was driven by altruistic ideals, aiming for something cool. But they quickly recognized the need to build a sustainable business model alongside the open-source approach.
Tuva’s paid offering is called Tuva Data Factory, which is a fully-managed version of the open source. With Data Factory, customers can access their data in the Tuva Data Model in ~2 weeks, rather than spending months setting up the open source and preprocessing their raw healthcare data sources. The cost of Data Factory starts at $10k per month, which is significantly less than the cost of hiring a data engineering team to manage the open source.
It is challenging to build open-source in healthcare compared to classic models like MongoDB or Elasticsearch. They hope to build a sustainable financial glide path through their data factory and add value through the community.
Advice for staying abreast of the industry
Since their initial goal for the project came from a deep understanding of the pain points in previous healthcare jobs, they constantly looked for effective tools and technologies to address these challenges.
Being close to the problem as technical founders and experiencing its pain intensified the search for optimal solutions.
“Everybody likes to consume information in different ways. Some people read newsletters, some people listen to a lot of podcasts, like it's all different. Whatever your modality is, try and figure out what the good newsletters to follow, what the good podcasts follow are etc. And follow those.”
Recognizing that individuals prefer diverse modalities for consuming information, Aaron suggests finding the right resources such as newsletters or podcasts that align with personal preferences.
Aaron shares his preference for podcasts and highlights the value of platforms like the Healthcare Playbook and Health Tech Nerds for staying informed and connected in the healthcare space.
Thanks for reading Pear Healthcare Playbook! Subscribe for free to receive new posts and support our work. If you found this post insightful or want to learn more, please feel free to connect with Andrew or Lipsa.
Interested in Tuva Health? Learn more on their website and LinkedIn or join their Slack.
A note from our sponsor: PacWest
Looking for guidance, connections, resources, opportunity? Pacific Western Bank’s banking products and services are built to support your evolving needs as you navigate the challenges of growing a successful business. As you continue to scale, our team will be with you every step of the way. Ready to take your business to the next level? Learn more: pacwest.com