High Dimensional Health Data

David Sontag, AI researcher and CEO of Layer Health, joins hosts Phil Ballentine and Aaron Neiderhiser to discuss how AI can meaningfully transform healthcare. From his formative years at Google and MIT to founding Layer Health, David shares what it takes to build AI that truly understands clinical data and improves care. We cover the evolution of machine learning, reasoning in large language models, why healthcare has lagged in AI adoption, and where the biggest opportunities lie today.TIMESTAMPS:(00:00) Introduction and guest background(01:12) Early interest in AI and machine learning(01:53) Academic journey and influences(03:27) Internship at Google and key experiences(06:41) Milestones in AI development(11:32) PhD research and medical applications(14:30) Transition to healthcare AI(27:49) Founding of Layer Health(36:01) Current focus and future prospects(46:23) Conclusion and contact information🔔 SubscribeFor deep dives and guest interviews about healthcare data, AI, and the future of health tech.🎙️ About High Dimensional Health Data (HD²) HD² is the podcast for people building healthcare's data and AI future. Hosted by Aaron and Phil, HD² explores the intersection of healthcare, data, and AI, and focuses on conversations with builders and practitioners.Hosts Phil Ballentine – Sr Director, Data Engineering at Atropos Health 🔗 LinkedInAaron Neiderhiser – Co-Founder and CEO at Tuva Health 🔗 LinkedInGuest David Sontag – CEO at Layer Health 🔗 LinkedIn📩 Interested in working together or coming on the podcast as a guest? Email us at hello@highdimensional.health

In this episode, we chat with Alyssa Antonopoulos, a bioinformatician at Natera who spends her days wrangling some of the messiest data in healthcare. We get into the chaos of clinical abstraction, why large language models are exciting (but not magic), and how a single phrase like “last day of therapy” can derail an analysis. Alyssa walks us through Natera’s journey from prenatal testing to cancer monitoring and kidney disease, and we talk about what real-world evidence could look like when the data finally starts playing nice.TIMESTAMPS(0:00) Intro(1:59) Understanding Natera’s Flagship Test(3:34) Natera’s Expansion into Therapeutic Areas(5:09) Real-World Data and Monitoring Frequency(6:40) Data Integration and Abstraction Challenges(9:27) Leveraging Large Language Models for Data Extraction(11:30) Evaluating and Improving Data Accuracy(21:43) Flexible Data Structures for Analysis(23:07) Data Modeling and Oncology Challenges(23:40) Importance of Semantic Definitions(24:27) Data Dictionaries and Abstraction Logic(26:24) Oncology and Chronic Kidney Disease Analyses(27:35) Genomic Landscapes and ctDNA Dynamics(28:25) Kaplan–Meier Analyses in CKD(34:56) Future of Real-World Data and Evidence(38:55) Leading a Data Team: Advice and Insights(42:38) Closing Remarks and Future Plans🔗Links & Resources:Natera Inc: https://www.natera.com/🔔 Subscribe For deep dives and guest interviews about healthcare data, AI, and the future of health tech.🎙️ About High Dimensional Health Data (HD²) HD² is the podcast for people building healthcare's data and AI future. Hosted by Aaron and Phil, HD² explores the intersection of healthcare, data, and AI, and focuses on conversations with builders and practitioners.HostsPhil Ballentine – Sr Director, Data Engineering at Atropos Health🔗 https://www.linkedin.com/in/phil-ballentine/Aaron Neiderhiser – Co-Founder and CEO at Tuva Health🔗 https://www.linkedin.com/in/aaronneiderhiser/GuestsAlyssa Antonopoulo – Senior Director, Data and Informatics (RWD) at Natera 🔗https://www.linkedin.com/in/alyssa-antonopoulos/ Interested in working together or coming on the podcast as a guest? Email us at hello@highdimensional.health

EP 03: Mara Alexeev on Messy Clinical Data, the Limit of EHRs, and the Future of Clinical InformaticsIn this episode, we chat with Mara Alexeev, a pediatrician-turned-clinical informaticist with a knack for turning medical chaos into clean, usable data. We talk EHR chaos, learning R on maternity leave, and why smelling liver failure is a real diagnostic skill.TIMESTAMPS(00:00) Meet Mara(04:00) Becoming chief after attending(06:00) The broken ritual of rounding(09:00) Fixing cancer workflows with templates(12:00) Building rogue order sets at Kaiser(16:00) Learning R during maternity leave(20:00) "I am the data process"(24:00) When a black box warning hits(30:00) Why EHR ≠ the patient(33:00) Diagnosing disease by smell(40:00) Over-measuring what doesn’t matter(46:00) The future: GLP-1s & off-the-shelf care(52:00) The dirty secrets of health data(56:00) Mara’s dream: Data Talk for real-world messes🔗Links & Resources:🔔 Subscribe For deep dives and guest interviews about healthcare data, AI, and the future of health tech.🎙️ About High Dimensional Health Data (HD²) HD² is the podcast for people building healthcare's data and AI future. Hosted by Aaron and Phil, HD² explores the intersection of healthcare, data, and AI, and focuses on conversations with builders and practitioners.HostsPhil Ballentine - Sr Director, Data Engineering at Atropos Health🔗⁠ https://www.linkedin.com/in/phil-ballentine/Aaron Neiderhiser - Co-Founder and CEO at Tuva Health🔗⁠ https://www.linkedin.com/in/aaronneiderhiser/⁠GuestsMara Alexeev - ​​Clinical Informatician and Pediatrician🔗https://www.linkedin.com/in/maraalexeev/Interested in working together or coming on the podcast as a guest? Email us at ⁠hello@highdimensional.health

In this episode, we chat with Vera M., a cancer researcher-turned-chief scientific officer-turned-biotech founder. We dive into what makes healthcare data “good,” the challenges of connecting siloed clinical trial data to the real world, and the evolving infrastructure powering data-driven medicine. Vera shares her journey from the lab to consulting to startups, and we explore the current and future state of real-world data, patient matching, and tokenization.TIMESTAMPS:(00:00) Meet our first guest: Vera Mucaj (01:00) From cancer research to Datavent  (03:00) Why consulting became the next step  (05:00) Startup life(06:00) Linking trials to real-world data  (09:00) The problem with siloed trial data  (11:00) What happens after a trial ends?  (15:00) Are we underusing trial data?  (17:00) The evolution of informed consent  (20:00) Ethics vs access  (26:00) Real-world data is still early  (29:00) Observational limits & FDA guidance  (32:00) Can RWD replicate trial results?  (34:00) Data context is everything  (36:00) Could consumer data help healthcare?  (39:00) Why data decisions take so long  (44:00) Product tip: No one wants your dashboard  (47:00) Vera’s Nature paper on patient matching  (51:00) The penny-drop moment for linkage  (52:00) What Vera’s building next  (55:00) Advice for pivoting PhDs  (57:00) Wrap-up and reflections🔗Links & Resources:Vera’s Nature article on patient matchingFlex CapitalDataventFDA’s guidance on real-world evidence🔔 Subscribe For deep dives and guest interviews about healthcare data, AI, and the future of health tech.🎙️ About High Dimensional Health Data (HD²) HD² is the podcast for people building healthcare's data and AI future. Hosted by Aaron and Phil, HD² explores the intersection of healthcare, data, and AI, and focuses on conversations with builders and practitioners.HostsPhil Ballentine – Sr Director, Data Engineering at Atropos Health🔗 https://www.linkedin.com/in/phil-ballentine/Aaron Neiderhiser – Co-Founder and CEO at Tuva Health🔗 https://www.linkedin.com/in/aaronneiderhiser/GuestsVera Mucaj – Venture Partner at Flex Capital & Former Chief Scientific Officer at Datavent 🔗https://www.linkedin.com/in/veramucaj/ Interested in working together or coming on the podcast as a guest? Email us at hello@highdimensional.health

In our pilot episode, we dive into how large language models are being used to clean up messy healthcare data—plus what that means for the future of clinical workflows. Phil shares details on a tool he built and evaluated at Atropos Health to map medication descriptions to RxNorm, and we unpack where AI is actually working in healthcare data (and where it's not working quite yet).(00:00) Enter: The Phantom Menace(00:28) How AI is reshaping health data(01:04) The messiness of strings in clinical data(03:11) The chaos of terminology in healthcare(04:13) Using UMLS to find structure(05:53) AI + RAG reaching 99.7% accuracy in structuring messy text?(09:19) Reimagining how data gets created(18:38) Will LLMs kill traditional clinical NLP?(24:57) SQL: annoyingly unbeatable (for now)(27:49) Where the health tech money’s going(29:12) RAG explained… kind of(39:21) What happens next? Let’s speculate🔔 Subscribe for deep dives and guest interviews about healthcare data, AI, and the future of health tech. 🎙️ About High Dimensional Health Data (HD²):HD² is the podcast for people building healthcare's data and AI future. Hosted by Aaron and Phil, HD² explores the intersection of healthcare, data, and AI, and focuses on conversations with builders and practitioners.YouTube: https://www.youtube.com/@highdimensionalhealthdataSocials:Phil Ballentine, Sr Director, Data Engineering at Atropos Health : https://www.linkedin.com/in/phil-ballentine/Aaron Neiderhiser, Co-Founder and CEO at Tuva Health: https://www.linkedin.com/in/aaronneiderhiser/Interested in working together or coming on the podcast as a guest? Email us at hello@highdimensional.health.