Epic unveils Comet, an AI platform for predicting patient health journeys

Every day, doctors make judgment calls without a crystal ball. A patient walks in clutching their stomach — do they need immediate admission, more tests, or just rest at home? Another is recovering in the hospital — send them home now, or risk an early readmission? These are high-stakes decisions, often made with incomplete information. Epic, the electronic health records giant, believes it can help. The company has just introduced Comet, an AI-powered platform that aims to forecast where a patient's health journey is heading next. Built on data from more than 100 billion medical events, Comet promises to shine a light into the fog of clinical uncertainty. As Epic puts it, the platform is designed to "help care teams anticipate and act."

How does it work?

At its core, Comet is a simulation engine. It's trained on a massive subset of Epic's Cosmos database, which aggregates millions of de-identified patient records from health systems across the country. Much like today's large language models learn from billions of words, Comet learns from time-stamped medical events:

• Diagnoses - everything from a first asthma attack to a late-stage cancer detection
• Lab results - blood tests, imaging findings, and biomarkers that shape care decisions
• Medications - prescriptions filled and discontinued
• Encounters - hospital admissions, discharges, ER visits, follow-ups

Feed it a patient's current state, and Comet generates multiple "what-if" scenarios. One path may suggest a quick recovery and discharge; another hints at complications or a potential readmission. Clinicians don't get a single prediction carved in stone — they get a spectrum of likely futures, distilled into clear insights like: Will this patient stay longer than expected? How high is their 30-day readmission risk? Could an underlying condition, like pancreatic cancer, emerge?

Why does it matter?

Healthcare is often a game of educated guesses. Clinicians rely on training, intuition, and fragmented data. Comet wants to tip the odds in their favor by grounding those decisions in data-driven foresight. In a study of 78 use cases, Comet outperformed individual, condition-specific models most of the time. That matters because even small improvements in anticipating risk can mean fewer unnecessary admissions, better resource allocation, and — most importantly — lives saved.

As Epic notes, the shift is about moving from "reactive to anticipatory" care. Instead of waiting for a crisis, providers could prepare for the most likely scenarios, tailoring care plans to each patient's unique trajectory. It's not about replacing clinical judgment; it's about giving that judgment a sharper edge.

The context

Comet is housed entirely within Epic Cosmos, a secure, cloud-based platform built for collaborative health research. Cosmos updates continuously as care is delivered, meaning Comet's predictions evolve alongside medical practice. Privacy is a centerpiece: the system only uses de-identified data and is evaluated carefully before being integrated into any real-world workflow.

The rollout starts in February 2026, when researchers from health systems participating in Cosmos will get access to a virtual lab. There, they'll stress-test Comet against real cases, probing for strengths, blind spots, and new applications.

It's a cautious but significant step in how AI could be woven into healthcare. For decades, hospitals have leaned on retrospective studies, guidelines, and gut feel. Now, a tool trained on an ocean of real patient journeys is offering to project what tomorrow might bring. If it delivers, Comet could mark a turning point: care that doesn't just respond to illness but stays one step ahead of it.