Epic's EHR Vision for Space Medicine and Genomic

My entire health history has lived in a digital record. I have never known anything different. I've logged into portals on my phone to read my own lab results before a doctor called me. I've watched pharmacogenomic flags pop up next to my prescriptions—warnings that my specific variants might make a standard dose hit wrong. For me, the patient record isn't abstract infrastructure. It's a thing I interact with, negotiate with, occasionally argue with.

So when Peter DeVault, a senior leader at Epic, stood up at a UCTV-recorded event earlier this year to argue that the same EHR stack powering American hospitals could underpin healthcare in space, I was not automatically skeptical. I was actually curious in a very specific way: does he mean this infrastructure is ready, or does he mean it's directionally right?

Having watched the talk, I think the honest answer is: both, depending on which layer you're looking at.

The interoperability case is the strongest one

DeVault opens with scale, and it lands. Epic's Care Everywhere platform exchanges roughly 20 million patient records per day—a figure DeVault says he double-checked right before the talk. More than half of those exchanges are Epic to non-Epic systems, including the Department of Defense and VA. The Cosmos platform (yes, named with this talk's theme in mind) holds what DeVault describes as approximately 300 million longitudinal records from health systems across the U.S., Lebanon, Saudi Arabia, Switzerland, and Canada. Epic self-reports this figure, and it hasn't been independently audited, but the order of magnitude is consistent with Epic's known market footprint.

The interoperability framing is genuinely smart for the space medicine context. An astronaut crew on a long-duration mission will have come from different countries, trained under different health systems, and been tested at different labs. The nightmare scenario isn't a missing file—it's a medication prescribed in orbit that interacts badly with a genetic variant that was tested three years ago at a clinic that uses a different system. DeVault's point is that Epic already solves the version of this problem that exists on Earth:

"Astronauts who might receive care and indeed come from many countries around the world can have their records assembled in wherever they go. And indeed here or up in space, their records can follow them. And when we say records again, it's not just the paper—it's structured data."

Structured data. That's the thing. A scanned PDF of a genomic report is almost useless at the point of care. A structured pharmacogenomic flag that fires when someone tries to prescribe a CYP2D6-metabolized drug to a poor metabolizer? That saves a life. Epic claims—through DeVault, who is their spokesperson, so read accordingly—to have been the first EHR to build a native, structured home for genomic and genetic data inside the patient record. That claim hasn't been independently verified, but the capability itself is real and the clinical value is not in dispute.

The AI piece is where I want to slow down

This is where I stopped nodding and started scribbling.

DeVault describes a large medical model built on Cosmos data that doesn't just generate text—it generates futures. Tokenize the events in a patient's health timeline, train a model on 300 million records, and instead of completing a sentence, you complete a health trajectory. Hundreds or thousands of parallel scenarios, narrowed toward specific risk profiles. Apply that to telemetry from a spacecraft, genomic and proteomic and transcriptomic data collected before, during, and after a mission, and you have something that could genuinely change what "monitoring astronaut health" means.

"We can generate potential futures for a person based on what has happened before... We can expand this to new types of data, whether it's telemetry data being collected in space, genomic and proteomic and transcriptomic data collected before, during, and after a mission."

Here's what I keep coming back to, though: a model trained on Earth-based health data, drawn primarily from U.S. and Western health systems, is going to have a complicated relationship with the bodies of astronauts undergoing microgravity-induced physiological changes that have no real analog in the training set. The model will have seen millions of cardiovascular events. It has seen very few cases of fluid shifting cephalad at 17,500 miles per hour for six months straight. I'm not saying the model is useless—I'm saying the extrapolation zone is huge, and DeVault doesn't spend much time there.

The Cosmos governance question deserves a real answer

Three hundred million longitudinal records in one platform, owned by a private, employee-owned company based in Verona, Wisconsin, with offices in 16 countries. DeVault frames this as a feature—single code base, unified governance, learnings from a Finland deployment flow back to the U.S. and everywhere else. That is genuinely a good thing for software consistency.

But I need someone to explain to me, in specific terms, what the governance structure looks like for patients whose data flows into Cosmos. Who decides what research gets run on it? What happens when a health system in Saudi Arabia and a health system in Switzerland have different national rules about what can be inferred from a patient's genome? What does meaningful consent look like at that scale? DeVault gestures at a "shared governance structure" but the talk doesn't go deep, and I find myself genuinely wanting the long answer.

This isn't a gotcha. Epic has worked on patient privacy longer than most players in this space, and DeVault's framing throughout is patient-centered in a way that doesn't feel like marketing gloss. But "patient-centered" and "patient-governed" are different things, and at 300 million records that distinction starts to matter enormously. 🧬

The training data problem isn't a footnote

I want to be direct about this because it keeps getting treated as a caveat when it's actually the central challenge.

If Epic's AI model learns what "healthy" looks like from data that skews toward insured, English-speaking, U.S.-based patients, then the model's predictions are calibrated for that population. Apply it to a Nigerian astronaut, a Japanese astronaut, or—less dramatically—an uninsured American who only shows up in the record when something goes wrong, and you have a system that is confidently wrong in ways that are very hard to detect. The pharmacogenomics version of this is well-documented: early PGx databases were staggeringly non-representative, and the clinical guidance built on them carried real risk for patients from underrepresented ancestries.

Cosmos has international nodes now—Lebanon, Saudi Arabia, Switzerland, Canada. That's progress. But "international" doesn't automatically mean "representative," and aggregating more data from more wealthy, connected health systems doesn't solve the underlying bias problem. It just makes the blind spots harder to locate.

DeVault doesn't address this in the talk. That's not an accusation—it was an 11-minute overview, not a technical seminar. But it's the question I'd walk up to the microphone and ask.

On the Sagan ending

DeVault closes with Carl Sagan—specifically the Cosmos series line that we are "star stuff," and his own riff on it: that it's not enough for the universe to know itself through us; we should be the way the universe heals itself.

I'll be honest: I found it genuinely moving. Not despite the fact that it was a little on-the-nose, but because it revealed something real about why people spend careers building healthcare infrastructure. That impulse—to make the data follow the person, to make the medicine as portable as the human—is not cynical. It predates the commercial opportunity by decades.

The Sagan quote as DeVault rendered it during the talk reads as slightly paraphrased rather than verbatim; the UCTV recording is the authoritative source for his exact words. But the spirit of his adaptation is clear: the work of space medicine isn't just technical ambition. It's a bet that connected health data, at scale, with the genomics baked in from the start, could protect humans in genuinely unprecedented environments.

I believe he believes that. I also think the version of Epic's platform that can do everything he described—reliably, equitably, in the microgravity edge cases—is still being built.

Which is fine. Important things take time to build correctly. The question I'm sitting with isn't whether Epic belongs in the space medicine conversation. It does. The question is whether the urgency of the space medicine timeline will push the deployment faster than the hard representativeness and governance problems get solved—and whether the people making those deployment decisions will even know what they're skipping.

Mei Zhang is Buzzrag's biotech and genetics reporter. She covers the genomics revolution with enthusiasm and a healthy suspicion of anyone who says the hard parts are already solved.